In a nutshell:

You can start building software and tech skills by trying out learning to code, and then doing some programming projects before applying for jobs. You can apply (as well as continue to develop) your software and tech skills by specialising in a related area, such as technical AI safety research, software engineering, or information security. You can also earn to give, and this in-demand skill set has great backup options.

Why are software and tech skills valuable?

By “software and tech” skills we basically mean what your grandma would call “being good at computers.”

When investigating the world’s most pressing problems, we’ve found that in many cases there are software-related bottlenecks.

For example, machine learning (ML) engineering is a core skill needed to contribute to AI safety technical research. Experts in information security are crucial to reducing the risks of engineered pandemics, as well as other risks. And software engineers are often needed by nonprofits, whether they’re working on reducing poverty or mitigating the risks of climate change.

Also, having skills in this area means you’ll likely be highly paid, offering excellent options to earn to give.

Moreover, basic programming skills can be extremely useful whatever you end up doing. You’ll find ways to automate tasks or analyse data throughout your career.

What does a career using software and tech skills involve?

A career using these skills typically involves three steps:

1. Learn to code with a university course or self-study and then find positions where you can get great mentorship. (Read more about how to get started.)
2. Optionally, specialise in a particular area, for example, by building skills in machine learning or information security.
3. Apply your skills to helping solve a pressing global problem. (Read more about how to have an impact with software and tech.)

There’s no general answer about when to switch from a focus on learning to a focus on impact. Once you have some basic programming skills, you should look for positions that both further improve your skills and have an impact, and then decide based on which specific opportunities seem best at the time.

Software and tech skills can also be helpful in other, less directly-related career paths, like being an expert in AI hardware (for which you’ll also need a specialist knowledge skill set) or founding a tech startup (for which you’ll also need an organisation-building skill set). Being good with computers is also often part of the skills required for quantitative trading.

Programming also tends to come in handy in a wide variety of situations and jobs; there will be other great career paths that will use these skills that we haven’t written about.

How to evaluate your fit

How to predict your fit in advance

Some indications you’ll be a great fit include:

• The ability to break down problems into logical parts and generate and test hypotheses
• Willingness to try out many different solutions
• High attention to detail
• Broadly good quantitative skills

The best way to gauge your fit is just to try out programming.

It seems likely that some software engineers are significantly better than average — and we’d guess this is also true for other technical roles using software. In particular, these very best software engineers are often people who spend huge amounts of time practicing. This means that if you enjoy coding enough to want to do it both as a job and in your spare time, you are likely to be a good fit.

How to tell if you’re on track

If you’re at university or in a bootcamp, it’s especially easy to tell if you’re on track. Good signs are that you’re succeeding at your assigned projects or getting good marks. An especially good sign is that you’re progressing faster than many of your peers.

In general, a great indicator of your success is that the people you work with most closely are enthusiastic about you and your work, especially if those people are themselves impressive!

If you’re building these skills at an organisation, signs you’re on track might include:

• You get job offers at organisations you’d like to work for.
• You’re promoted within your first two years.
• You receive excellent performance reviews.
• You’re asked to take on progressively more responsibility over time.
• After some time, you’re becoming someone in your team who people look to solve their problems, and people want you to teach them how to do things.
• You’re building things that others are able to use successfully without your input.
• Your manager / colleagues suggest you might take on more senior roles in the future.
• You ask your superiors for their honest assessment of your fit and they are positive (e.g. they tell you you’re in the top 10% of people they can imagine doing your role).

How to get started building software and tech skills

Independently learning to code

As a complete beginner, you can write a Python program in less than 20 minutes that reminds you to take a break every two hours.

A great way to learn the very basics is by working through a free beginner course like Automate the Boring Stuff with Python by Al Seigart.

Once you know the fundamentals, you could try taking an intro to computer science or intro to programming course. If you’re not at university, there are plenty of courses online, such as:

• Udacity’s Intro to Theoretical Computer Science
• MIT’s Introduction to Computer Science and Programming
• Stanford’s Programming Methodology

Don’t be discouraged if your code doesn’t work the first time — that’s what normally happens when people code!

A great next step is to try out doing a project with other people. This lets you test out writing programs in a team and working with larger codebases. It’s easy to come up with programming projects to do with friends — you can see some examples here.

Once you have some more experience, contributing to open-source projects in particular lets you work with very large existing codebases.

Attending a coding bootcamp

We’ve advised many people who managed to get junior software engineer jobs in less than a year by going to a bootcamp.

Coding bootcamps are focused on taking people with little knowledge of programming to as highly paid a job as possible within a couple of months. This is a great entry route if you don’t already have much background, though some claim the long-term prospects are not as good as if you studied at university or in a particularly thorough way independently because you lack a deep understanding of computer science. Course Report is a great guide to choosing a bootcamp. Be careful to avoid low-quality bootcamps. To find out more, read our interview with an App Academy instructor.

Studying at university

Studying computer science at university (or another subject involving lots of programming) is a great option because it allows you to learn to code in an especially structured way and while the opportunity cost of your time is lower.

It will also give you a better theoretical understanding of computing than a bootcamp (which can be useful for getting the most highly-paid and intellectually interesting jobs), a good network, some prestige, and a better understanding of lower-level languages like C. Having a computer science degree also makes it easier to get a US work visa if you’re not from the US.

Doing internships

If you can find internships, ideally at the sorts of organisations you might want to work for to build your skills (like big tech companies or startups), you’ll gain practical experience and the key skills you wouldn’t otherwise pick up from academic degrees (e.g. using version control systems and powerful text editors). Take a look at our our list of companies with software and machine learning internships.

AI-assisted coding

As you’re getting started, it’s probably worth thinking about how developments in AI are going to affect programming in the future — and getting used to AI-assisted coding.

We’d recommend trying out using GitHub CoPilot, which writes code for you based on your comments. Cursor is a popular AI-assisted code editor based on VSCode.

You can also just ask AI chat assistants for help. ChatGPT is particularly helpful (although only if you use the paid version).

We think it’s reasonably likely that many software and tech jobs in the future will be heavily based on using tools like these.

Building a specialty

Depending on how you’re going to use software and tech skills, it may be useful to build up your skills in a particular area. Here’s how to get started in a few relevant areas:

Depending on how you’re aiming to have an impact with these skills (see the next section), you may also need to develop other skills. We’ve written about some other relevant skill sets:

• How to get started building research skills
• How to get started building specialist knowledge relevant to a pressing problem
• How to get started building skills for communicating ideas

For more, see our full list of impactful skills.

Once you have these skills, how can you best apply them to have an impact?

The problem you work on is probably the biggest driver of your impact. The first step is to make an initial assessment of which problems you think are most pressing (even if you change your mind over time, you’ll need to decide where to start working).

Once you’ve done that, the next step is to identify the highest-potential ways to use software and tech skills to help solve your top problems.

There are five broad categories here:

• Use software and tech skills in research. Lots of technical research relevant to the world’s most pressing problems makes heavy use of software and tech skills — most notably, AI safety technical research. To be successful, you might also need a research skill set, which we’ve written about separately. For some paths, you’ll also need specialist knowledge in an area related to a pressing problem — e.g. hardware for becoming an expert in AI hardware.
• ML engineering for AI safety research. Most AI safety researchers work closely with engineers (and in many organisations, no clear distinction is made). This is a particularly high-impact way of using software and tech skills because we think risks from AI is one of the world’s most pressing problems.
• Build software for organisations working on pressing problems. Most organisations working on everything from global health to reducing the risk of nuclear war need software engineers to manage computer systems, apps, and websites. The key feature that draws this work together is that you’ll be building a product for others to use. Read more about software engineering careers and organisation-building skills.
• Protect sensitive information. Some organisations need help protecting information that could be hugely dangerous if it was known more widely, such as harmful genetic sequences or powerful AI technology. Breaches in areas like these could have disastrous consequences — which makes information security a great option for people who want to have a high-impact career. Read more about information security.
• Earn to give. Most jobs that use software and tech skills, whether software engineering, information security, data science, or something else entirely, command high salaries (particularly in the US) — and so they offer a great option for earning to give. Skilled software engineers can earn $300,000 a year or more at big tech companies. Probably the highest-paying routes are trading in quantitative hedge funds or founding a tech startup.

While some of these options (like protecting dangerous information) will require building up some more specialised skills, being a great programmer will let you move around most of these categories relatively easily, and the earning to give options means you’ll always have a pretty good backup plan.

Find jobs that use software and tech skills

See our curated list of job opportunities for this path.

View all opportunities

Career paths we’ve reviewed that use these skills

• AI safety technical research
• Information security in high-impact areas
• Expert in AI hardware
• Forecasting and related research and implementation
• Data collection for AI alignment
• Software engineering
• Found a tech startup
• Earning to give in quantitative trading
• Machine learning PhDs
• Early-stage startup employee
• Working at a leading AI lab
• Computer science PhD
• Data science
• Product manager in tech
• Web designer

Read next:  Explore other useful skills

Want to learn more about the most useful skills for solving global problems, according to our research? See our list.

Continue →

The post Software and tech skills appeared first on 80,000 Hours.

There’s still a lot of uncertainty about which strategies for AI governance and coordination would be best, though parts of the community of people working on this subject may be coalescing around some ideas. See, for example, a list of potential policy ideas from Luke Muehlhauser of Open Philanthropy¹ and a survey of expert opinion on best practices in AI safety and governance.

But there’s no roadmap here. There’s plenty of room for debate about which policies and proposals are needed.

We may not have found the best ideas yet in this space, and many of the existing policy ideas haven’t yet been developed into concrete, public proposals that could actually be implemented. We hope to see more people enter this field to develop expertise and skills that will contribute to risk-reducing AI governance and coordination.

“What you’re doing has enormous potential and enormous danger.” — US President Joe Biden, to the leaders of the top AI labs

Why this could be a high-impact career path

Artificial intelligence has advanced rapidly. In 2022 and 2023, new language and image generation models gained widespread attention for their abilities, blowing past previous benchmarks the technology had met.

And the applications of these models are still new; with more tweaking and integration into society, the existing AI systems may become easier to use and more ubiquitous in our lives.

We don’t know where all these developments will lead us. There’s reason to be optimistic that AI will eventually help us solve many of the world’s problems, raising living standards and helping us build a more flourishing society.

But there are also substantial risks. AI can be used for both good and ill. And we have concerns that the technology could, without the proper controls, accidentally lead to a major catastrophe — and perhaps even cause human extinction. We discuss the arguments that these risks exist in our in-depth problem profile.

Because of these risks, we encourage people to work on finding ways to reduce these risks through technical research and engineering.

But a range of strategies for risk reduction will likely be needed. Government policy and corporate governance interventions in particular may be necessary to ensure that AI is developed to be as broadly beneficial as possible and without unacceptable risk.

Governance generally refers to the processes, structures, and systems that carry out decision making for organisations and societies at a high level. In the case of AI, we expect the governance structures that matter most to be national governments and organisations developing AI — as well as some international organisations and perhaps subnational governments.

Some aims of AI governance work could include:

• Preventing the deployment of any AI systems that pose a significant and direct threat of catastrophe
• Mitigating the negative impact of AI technology on other catastrophic risks, such as nuclear weapons and biotechnology
• Guiding the integration of AI technology into our society and economy with limited harms and to the advantage of all
• Reducing the risk of an “AI arms race,” in which competition leads to technological advancement without the necessary safeguards and caution — between nations and between companies
• Ensuring that those creating the most advanced AI models are incentivised to be cooperative and concerned about safety
• Slowing down the development and deployment of new systems if the advancements are likely to outpace our ability to keep them safe and under control

We need a community of experts who understand the intersection of modern AI systems and policy, as well as the severe threats and potential solutions. This field is still young, and many of the paths within it aren’t clear and are not sure to pan out. But there are relevant professional paths that will provide you valuable career capital for a variety of positions and types of roles.

The rest of this article explains what work in this area might involve, how you can develop career capital and test your fit, and where some promising places to work might be.

What kinds of work might contribute to AI governance?

What should governance-related work on AI actually involve? There are a variety of ways to pursue AI governance strategies, and as the field becomes more mature, the paths are likely to become clearer and more established.

We generally don’t think people early in their careers should be aiming for a specific job that they think would be high-impact. They should instead aim to develop skills, experience, knowledge, judgement, networks, and credentials — what we call career capital — that they can later use when an opportunity to have a positive impact is ripe.

This may involve following a pretty standard career trajectory, or it may involve bouncing around in different kinds of roles. Sometimes, you just have to apply to a bunch of different roles and test your fit for various types of work before you know what you’ll be good at. The main thing to keep in mind is that you should try to get excellent at something for which you have strong personal fit and that will let you contribute to solving pressing problems.

In the AI governance and coordination space, we see at least six large categories of work that we expect to be important:

• Government work
• Research on AI policy and strategy
• Industry work
• Advocacy and lobbying
• Third-party auditing and evaluation
• International work and coordination

There aren’t necessarily openings in all these categories at the moment for careers in AI governance, but they represent a range of sectors in which impactful work may potentially be done in the coming years and decades. Thinking about the different skills and forms of career capital that will be useful for the categories of work you could see yourself doing in the future can help you figure out what your immediate next steps should be. (We discuss how to assess your fit and enter this field below.)

You may want to — and indeed it may be advantageous to — move between these different categories of work at different points in your career. You can also test out your fit for various roles by taking internships, fellowships, entry-level jobs, temporary placements, or even doing independent research, all of which can serve as career capital for a range of paths.

We have also reviewed career paths in AI technical safety research and engineering and information security, which may be crucial to reducing risks from AI, and which may play a significant role in an effective governance agenda. People serious about pursuing a career in AI governance should familiarise themselves with these fields as well.

Government work

Taking a role within government could lead to playing an important role in the development, enactment, and enforcement of AI policy.

Note that we generally expect that the US federal government will be the most significant player in AI governance for the foreseeable future. This is because of its global influence and its jurisdiction over much of the AI industry, including the top three AI labs training state-of-the-art, general-purpose models (Anthropic, OpenAI, and Google DeepMind) and key parts of the chip supply chain. Much of this article focuses on US policy and government.²

But other governments and international institutions may also end up having important roles to play in certain scenarios. For example, the UK government, the European Union, China, and potentially others, may all present opportunities for impactful AI governance work. Some US state-level governments, such as California, may also offer opportunities for impact and gaining career capital.

What would this work involve? Sections below discuss how to enter US policy work and which areas of the government that you might aim for.

But at the broadest level, people interested in positively shaping AI policy should aim to gain the skills and experience to work in areas of government with some connection to AI or emerging technology policy.

This can include roles in: legislative branches, domestic regulation, national security, diplomacy, appropriations and budgeting, and other policy areas.

If you can get a role out of the gate that is already working directly on this issue, such as a staff position with a lawmaker who is focused on AI, that could be a great opportunity.

Otherwise, you should seek to learn as much as you can about how policy works and which government roles might allow you to have the most impact, while establishing yourself as someone who’s knowledgeable about the AI policy landscape. Having almost any significant government role that touches on some aspect of AI, or having some impressive AI-related credential, may be enough to get you quite far.

One way to advance your career in government on a specific topic is what some call “getting visibility” — that is, using your position to learn about the landscape and connect with the actors and institutions that affect the policy area you care about. You’ll want to be invited to meetings with other officials and agencies, be asked for input on decisions, and engage socially with others who work in the policy area. If you can establish yourself as a well-regarded expert on an important but neglected aspect of the issue, you’ll have a better shot at being included in key discussions and events.

Career trajectories within government can be broken down roughly as follows:

• Standard government track: This involves entering government at a relatively low level and building up your career capital on the inside by climbing the seniority ladder. For the highest impact, you’d ideally end up reaching senior levels by sticking around, gaining skills and experience, and getting promoted. You may move between agencies, departments, or branches.
• Specialisation career capital: You can also move in and out of government throughout your career. People on this trajectory will also work at nonprofits, think tanks, industry labs, political parties, academia, and other organisations. But they will primarily focus on becoming an expert in a topic — such as AI. It can be harder to get seniority this way, but the value of expertise and experience can sometimes outweigh seniority.
• Direct-impact work: Some people move into government jobs without a longer plan to build career capital because they see an opportunity for direct, immediate impact. This might look like getting tapped to lead an important commission or providing valuable input on an urgent project. We don’t generally recommend planning on this kind of strategy for your career, but it’s good to be aware of it as an opportunity that might be worth taking at some point.

Research on AI policy and strategy

There’s still a lot of research to be done on the most important avenues for AI governance approaches. While there are some promising proposals for a system of regulatory and strategic steps that can help reduce the risk of an AI catastrophe, there aren’t many concrete and publicly available policy proposals ready for adoption.

The world needs more concrete proposals for AI policies that would really start to tackle the biggest threats; developing such policies, and deepening our understanding of the strategic needs of the AI governance space, should be high priorities.

Other relevant research could involve surveys of public opinion that could inform communication strategies, legal research about the feasibility of proposed policies, technical research on issues like compute governance, and even higher-level theoretical research into questions about the societal implications of advanced AI. Some research, such as that done by Epoch AI, focuses on forecasting the future course of AI developments, which can influence AI governance decisions.

However, several experts we’ve talked to warn that a lot of research on AI governance may prove to be useless, so it’s important to be reflective and seek input from others in the field — both from experienced policy practitioners and technical experts — about what kind of contribution you can make. We list several research organisations below that we think would be good to work at in order to pursue promising research on this topic.

One potentially useful approach for testing your fit for this work — especially when starting out in this research — is to write up analyses and responses to existing work on AI policy or investigate some questions in this area that haven’t been the subject of much attention. You can then share your work widely, send it out for feedback from people in the field, and evaluate how much you enjoy the work and whether you might productively contribute to this research longer term.

But it’s possible to spend too long testing your fit without making much progress, and some people find that they’re best able to contribute when they’re working on a team. So don’t overweight or over-invest in independent work, especially if there are few signs it’s working out especially well for you. This kind of project can make sense for maybe a month or a bit longer — but it’s unlikely to be a good idea to spend much more than that without meaningful funding or some really encouraging feedback from people working in the field.

If you have the experience to be hired as a researcher, work on AI governance can be done in academia, nonprofit organisations, and think tanks. Some government agencies and committees, too, perform valuable research.

Note that universities and academia have their own priorities and incentives that often aren’t aligned with producing the most impactful work. If you’re already an established researcher with tenure, it may be highly valuable to pivot into work on AI governance — this position may even give you a credible platform from which to advocate for important ideas.

But if you’re just starting out a research career and want to focus on this issue, you should carefully consider whether your work will be best supported inside or outside of academia. For example, if you know of a specific programme with particular mentors who will help you pursue answers to critical questions in this field, it might be worth doing. We’re less inclined to encourage people to pursue generic academic-track roles with the vague hope that one day they can do important research on this topic.

Advanced degrees in policy or relevant technical fields may well be valuable, though — see more discussion of this in the section on how to assess your fit and get started.

Industry work

While government policy is likely to play a key role in coordinating various actors interested in reducing the risks from advanced AI, internal policy and corporate governance at the largest AI labs themselves is also a powerful tool. We think people who care about reducing risk can potentially do valuable work internally at industry labs. (Read our career review of non-technical roles at AI labs.)

At the highest level, deciding who sits on corporate boards, what kind of influence those boards have, and to what extent the organisation is structured to seek profit and shareholder value as opposed to other aims, can end up having a major impact on the direction a company takes. If you might be able to get a leadership role at a company developing frontier AI models, such as a management position or a seat on the board, it could potentially be a very impactful position.

If you’re able to join a policy team at a major lab, you can model threats and help develop, implement, and evaluate promising proposals internally to reduce risks. And you can build consensus around best practices, such as strong information security policies, using outside evaluators to find vulnerabilities and dangerous behaviours in AI systems (red teaming), and testing out the latest techniques from the field of AI safety.

And if, as we expect, AI labs face increasing government oversight, industry governance and policy work can ensure compliance with any relevant laws and regulations that get put in place. Interfacing with government actors and facilitating coordination over risk reduction approaches could be impactful work.

In general, the more cooperative AI labs are with each other³ and outside groups seeking to minimise catastrophic risks from AI, the better. And this doesn’t seem to be an outlandish hope — many industry leaders have expressed concern about extinction risks and have even called for regulation of the frontier technology they’re creating.

That said, we can expect this cooperation to take substantial work — it would be surprising if the best policies for reducing risks were totally uncontroversial in industry, since labs also face huge commercial incentives to build more powerful systems, which can carry more risk. The more everyone’s able to communicate and align their incentives, the better things seem likely to go.

Advocacy and lobbying

People outside of government or AI labs can influence the shape of public policy and corporate governance via advocacy and lobbying.

As of this writing, there has not yet been a large public movement in favour of regulating or otherwise trying to reduce risks from AI, so there aren’t many openings that we know about in this category. But we expect growing interest in this area to open up new opportunities to press for political action and policy changes at AI labs, and it could make sense to start building career capital and testing your fit now for different kinds of roles that would fall into this category down the line.

If you believe AI labs may be disposed to advocate for generally beneficial regulation, you might want to try to work for them, or become a lobbyist for the industry as a whole, to push the government to adopt specific policies. It’s plausible that AI labs will have by far the best understanding of the underlying technology, as well as the risks, failure modes, and safest paths forward.

On the other hand, it could be the case that AI labs have too much of a vested interest in the shape of regulations to reliably advocate for broadly beneficial policies. If that’s right, it may be better to join or create advocacy organisations unconnected from the industry — supported by donations or philanthropic foundations — that can take stances that are opposed to the labs’ commercial interests.

For example, it could be the case that the best approach from a totally impartial perspective would be at some point to deliberately slow down or halt the development of increasingly powerful AI models. Advocates could make this demand of the labs themselves or of the government to slow down AI progress. It may be difficult to come to this conclusion or advocate for it if you have strong connections to the companies creating these systems.

It’s also possible that the best outcomes will be achieved with a balance of industry lobbyists and outside lobbyists and advocates making the case for their preferred policies — as both bring important perspectives.

We expect there will be increasing public interest in AI policy as the technological advancements have ripple effects in the economy and wider society. And if there’s increasing awareness of the impact of AI on people’s lives, the risks the technology poses may become more salient to the public, which will give policymakers strong incentives to take the problem seriously. It may also bring new allies into the cause of ensuring that the development of advanced AI goes well.

Advocacy can also:

• Highlight neglected but promising approaches to governance that have been uncovered in research
• Facilitate the work of policymakers by showcasing the public’s support for governance measures
• Build bridges between researchers, policymakers, the media, and the public by communicating complicated ideas in an accessible way to many audiences
• Pressure corporations themselves to proceed more cautiously
• Change public sentiment around AI and discourage irresponsible behaviour by individual actors, such as the spreading of powerful open-source models

However, note that advocacy can sometimes backfire. Predicting how information will be received is far from straightforward. Drawing attention to a cause area can sometimes trigger a backlash; presenting problems with certain styles of rhetoric can alienate people or polarise public opinion; spreading misleading or mistaken messages can discredit yourself and fellow advocates. It’s important that you are aware of the risks, consult with others (particularly those who you respect but might disagree with tactically), and commit to educating yourself deeply about the topic before expounding on it in public.

You can read more in the section about doing harm below. We also recommend reading our article on ways people trying to do good accidentally make things worse and how to avoid them.

Third-party auditing and evaluation

If regulatory measures are put in place to reduce the risks of advanced AI, some agencies and organisations — within government or outside — will need to audit companies and systems to make sure that regulations are being followed.

One nonprofit, the Alignment Research Center, has been at the forefront of this kind of work.⁵ In addition to its research work, it has launched a program to evaluate the capabilities of advanced AI models. In early 2023, the organisation partnered with two leading AI labs, OpenAI and Anthropic, to evaluate the capabilities of the latest versions of their chatbot models prior to their release. They sought to determine in a controlled environment if the models had any potentially dangerous capabilities.

The labs voluntarily cooperated with ARC for this project, but at some point in the future, these evaluations may be legally required.

Governments often rely on third-party auditors as crucial players in regulation, because the government may lack the expertise (or the capacity to pay for the expertise) that the private sector has. There aren’t many such opportunities available in this type of role that we know of as of this writing, but they may end up playing a critical part of an effective AI governance framework.

Other types of auditing and evaluation may be required as well. ARC has said it intends to develop methods to determine which models are appropriately aligned — that is, that they will behave as their users intend them to behave — prior to release.

Governments may also want to employ auditors to evaluate the amount of compute that AI developers have access to, their information security practices, the uses of models, the data used to train models, and more.

Acquiring the technical skills and knowledge to perform these types of evaluations, and joining organisations that will be tasked to perform them, could be the foundation of a highly impactful career. This kind of work will also likely have to be facilitated by people who can manage complex relationships across industry and government. Someone with experience in both sectors could have a lot to contribute.

Some of these types of roles may have some overlap with work in AI technical safety research.

One potential advantage of working in the private sector for AI governance work is you may be significantly better paid than you would be in government.

International work and coordination

US-China

For someone with the right fit, cooperation and coordination with China on the safe development of AI could be a particularly impactful approach within the broad AI governance career path.

The Chinese government has been a major funder in the field of AI, and the country has giant tech companies that could potentially drive forward advances.

Given tensions between the US and China, and the risks posed by advanced AI, there’s a lot to be gained from increasing trust, understanding, and coordination between the two countries. The world will likely be much better off if we can avoid a major conflict between great powers and if the most significant players in emerging technology can avoid exacerbating any global risks.

We have a separate career review that goes into more depth on China-related AI safety and governance paths.

Other governments and international organisations

As we’ve said, we focus most on US policy and government roles. This is largely because we anticipate that the US is now and will likely continue to be the most pivotal actor when it comes to regulating AI, with a major caveat being China, as discussed in the previous section.

But many people interested in working on this issue can’t or don’t want to work in US policy — perhaps because they live in another country and don’t intend on moving.

Much of the advice above still applies to these people, because roles in AI governance research and advocacy can be done outside of the United States.⁶ And while we don’t think it’s generally as impactful in expectation as US government work, opportunities in other governments and international organisations can be complementary to the work to be done in the US.

The United Kingdom, for instance, may present another strong opportunity for AI policy work that would complement US work. Top UK officials have expressed interest in developing policy around AI, perhaps even a new international agency, and reducing extreme risks. And the UK government announced in 2023 the creation of a new AI Foundation Model Taskforce, with the expressed intention to drive forward safety research.

It’s possible that by taking significant steps to understand and regulate AI, the UK will encourage or inspire US officials to take similar steps by showing how it can work.

And any relatively wealthy country could use portions of its budget to fund AI safety research. While a lot of the most important work likely needs to be done in the US, along with leading researchers and at labs with access to large amounts of compute, some lines of research may be productive even without these resources. Any significant advances in AI safety research, if communicated properly, could be used by researchers working on the most powerful models.

Other countries might also develop liability standards for the creators of AI systems that could incentivise corporations to proceed more cautiously and judiciously before releasing models.

The European Union has shown that its data protection standards — the General Data Protection Regulation (GDPR) — affect corporate behaviour well beyond its geographical boundaries. EU officials have also pushed forward on regulating AI, and some research has explored the hypothesis that the impact of the union’s AI regulations will extend far beyond the continent — the so-called “Brussels effect.”

And at some point, we do expect there will be AI treaties and international regulations, just as the international community has created the International Atomic Energy Agency, the Biological Weapons Convention, and Intergovernmental Panel on Climate Change to coordinate around and mitigate other global catastrophic threats.

Efforts to coordinate governments around the world to understand and share information about threats posed by AI may end up being extremely important in some future scenarios.

The Organisation for Economic Cooperation and Development is one place where such work might occur. So far, it has been the most prominent international actor working on AI policy and has created the AI Policy Observatory.

Third-party countries may also be able to facilitate cooperation and reduce tensions betweens the United States and China, whether around AI or other potential flashpoints, should such an intervention become necessary.

How policy gets made

What does it actually take to make policy?

In this section, we’ll discuss three phases of policy making: agenda setting, policy creation and development, and implementation. We’ll generally discuss these as aspects of making government policy, but they could also be applied to organisational policy. The following section will discuss the types of work that you could do to positively contribute to the broad field of AI governance.

Agenda setting

To enact and implement a programme of government policies that have a positive impact, you have to first ensure that the subject of potential legislation and regulation is on the agenda for policymakers.

Agenda setting for policy involves identifying and defining problems, drawing attention to the problems and raising their salience (at least to the relevant people), and promoting potential approaches to solving them.

For example, when politicians take office, they often enter on a platform of promises made to their constituents and their supporters about which policy agendas they want to pursue. Those agendas are formed through public discussion, media narratives, internal party politics, deliberative debate, interest group advocacy, and other forms of input. The agenda can be, to varying degrees, problem-specific — having a broad remit of “improving health care.” Or it could be more solution-specific — aiming to create, for example, a single-payer health system.

Issues don’t necessarily have to be unusually salient to get on the agenda. Policymakers or officials at various levels of government can prioritise solving certain problems or enacting specific proposals that aren’t the subject of national debate. In fact, sometimes making issues too salient, framing them in divisive ways, or allowing partisanship and political polarisation to shape the discussion, can make it harder to successfully put solutions on the agenda.

What’s key for agenda setting as an approach to AI governance is that people with the authority have to buy into the idea of prioritising the issue, if they’re going to use their resources and political capital to focus on it.

Policy creation and development

While there does appear to be growing enthusiasm for a set or sets of policy proposals that could start to reduce the risk of an AI-related catastrophe, there’s still a lack of concrete policies that are ready to get off the ground.

This is what the policy creation and development process is for. Researchers, advocates, civil servants, lawmakers and their staff, and others all can play a role in shaping the actual legislation and regulation that the government eventually enforces. In the corporate context, internal policy creation can serve similar functions, though it may be less enforceable unless backed up with contracts.

Policy creation involves crafting solutions for the problem at hand with the policy tools available, usually requiring input from technical experts, legal experts, stakeholders, and the public. In countries with strong judicial review like the United States, special attention often has to be paid to make sure laws and regulations will hold up under the scrutiny of judges.

Once concrete policy options are on the table, they must be put through the relevant decision-making process and negotiations. If the policy in question is a law that’s going to be passed, rather than a regulation, it needs to be crafted so that it will have enough support from lawmakers and other key decision makers to be enacted. This can happen in a variety of ways; it might be rolled into a larger piece of legislation that has wide support, or it may be rallied around and brought forward as its own package to be voted on individually.

Policy creation can also be an iterative process, as policies are enacted, implemented, monitored, evaluated, and revised.

For more details on the complex work of policy creation, we recommend Thomas Kalil’s article “Policy Entrepreneurship in the White House: Getting Things Done in Large Organisations.”

Implementation

Fundamentally, a policy is only an idea. For an idea to have an impact, someone actually has to carry it out. Any of the proposals for AI-related government policy — including standards and evaluations, licensing, and compute governance — will demand complex management and implementation.

Policy implementation on this scale requires extensive planning, coordination in and out of government, communication, resource allocation, training and more — and every step in this process can be fraught with challenges. To rise to the occasion, any government implementing an AI policy regime will need talented individuals working at a high standard.

The policy creation phase is critical and is probably the highest-priority work. But good ideas can be carried out badly, which is why policy implementation is also a key part of the AI governance agenda.

Examples of people pursuing this path

How to assess your fit and get started

If you’re early on in your career, you should focus first on getting skills and other career capital to successfully contribute to the beneficial governance and regulation of AI.

You can gain career capital for roles in many ways, and the best options will vary based on your route to impact. But broadly speaking, working in or studying fields such as politics, law, international relations, communications, and economics can all be beneficial for going into policy work.

And expertise in AI itself, gained by studying and working in machine learning and technical AI safety, or potentially related fields such as computer hardware or information security, should also give you a big advantage.

Testing your fit

One general piece of career advice we give is to find relatively “cheap” tests to assess your fit for different paths. This could mean, for example, taking a policy internship, applying for a fellowship, doing a short bout of independent research as discussed above, or taking classes or courses on technical machine learning or computer engineering.

It can also just involve talking to people currently doing a job you might consider having and finding out what the day-to-day experience of the work is like and what skills are needed.

All of these factors can be difficult to predict in advance. While we grouped “government work” into a single category above, that label covers a wide range of positions and types of occupations in many different departments and agencies. Finding the right fit within a broad category like “government work” can take a while, and it can depend on a lot of factors out of your control, such as the colleagues you happen to work closely with. That’s one reason it can be useful to build broadly valuable career capital, so you have the option to move around to find the right role for you.

And don’t underestimate the value at some point of just applying to many relevant openings in the field and sector you’re aiming for and seeing what happens. You’ll likely face a lot of rejection with this strategy, but you’ll be able to better assess your qualifications for different kinds of roles after you see how far you get in the process, if you take enough chances. This can give you a lot more information than just guessing about whether you have the right experience.

It can be useful to rule out certain types of work if you gather evidence that you’re not a strong fit for the role. For example, if you invest a lot of time and effort trying to get into reputable universities or nonprofit institutions to do AI governance research, but you get no promising offers and receive little encouragement even after applying widely, this might be a significant signal that you’re unlikely to thrive in that particular path.

That wouldn’t mean you have nothing to contribute, but your comparative advantage may lie elsewhere.

Read the section of our career guide on finding a job that fits you.

Types of career capital

For a field like AI governance, a mix of people with technical and policy expertise — and some people with both — is needed.

While anyone involved in this field should work to maintain an evolving understanding of both the technical and policy details, you’ll probably start out focusing on either policy or technical skills to gain career capital.

This section covers:

• Generally useful career capital
• Policy-related career capital
• Technical career capital
• Other specific forms of career capital

Much of this advice is geared toward roles in the US, though it may be relevant in other contexts.

Generally useful career capital

The chapter of the 80,000 Hours career guide on career capital lists five key components that will be useful in any path: skills and knowledge, connections, credentials, character, and runway.

For most jobs touching on policy, social skills, networking, and — for lack of a better word — political skill will be a huge asset. This can probably be learned to some extent, but some people may find they don’t have these kinds of skills and can’t or don’t want to acquire them. That’s OK — there are many other routes to having a fulfilling and impactful career, and there may be some roles within this path that demand these skills to a much lesser extent. That’s why testing your fit is important.

Read the full section of the career guide on career capital.

Policy-related career capital

To gain skills in policy, you can pursue education in many relevant fields, such as political science, economics, and law.

Many master’s programmes offer specific coursework on public policy, science and society, security studies, international relations, and other topics; having a graduate degree or law degree will give you a leg up for many positions.

In the US, a master’s, a law degree, or a PhD is particularly useful if you want to climb the federal bureaucracy. Our article on US policy master’s degrees provides detailed information about how to assess the many options.

Internships in DC are a promising route to evaluate your aptitude for policy work and to establish early career capital. Many academic institutions now offer a strategic “Semester in DC” programme, which can let you explore placements of choice in Congress, federal agencies, or think tanks. The Virtual Student Federal Service (VSFS) also offers part-time, remote government internships. Balancing their academic commitments, students can access these opportunities during the academic year, further solidifying their grasp on the intricacies of policy work. This technological advance could be the stepping stone many aspiring policy professionals need to ascend in their future careers.

Once you have a suitable background, you can take entry-level positions within parts of the government where you can build a professional network and develop your skills. In the US, you can become a congressional staffer, or take a position at a relevant federal department, such as the Department of Commerce, Department of Energy, or the Department of State. Alternatively, you can gain experience in think tanks — a particularly promising option if you have a strong aptitude for research — and government contractors, private sector companies providing services to the government.

In Washington, DC, the culture is fairly unique. There’s a big focus on networking and internal bureaucratic politics to navigate. We’ve also been told that while merit matters to a degree in US government work, it is not the primary determinant of who is most successful. People who think they wouldn’t feel able or comfortable to be in this kind of environment for the long term should consider whether other paths would be best.

If you find you can enjoy government and political work, impress your colleagues, and advance in your career, though, that’s a strong signal that you have the potential to make a real impact. Just being able to thrive in government work can be an extremely valuable comparative advantage.

Technical career capital

Technical experience in machine learning, AI hardware, and related fields can be a valuable asset for an AI governance career. So it will be very helpful if you’ve studied a relevant subject area for an undergraduate or graduate degree, or a particularly productive course of independent study.

We have a guide to technical AI safety careers, which explains how to learn the basics of machine learning.

The following resources may be particularly useful for familiarising yourself with the field of AI safety:

• Redwood Research’s MLAB bootcamp curriculum
• BlueDot Impact’s AI Safety Fundamentals
• Center for AI Safety’s Introduction to ML Safety course

Working at an AI lab in technical roles, or other companies that use advanced AI systems and hardware, may also provide significant career capital in AI policy paths. (Read our career review discussing the pros and cons of working at a top AI lab.)

We also have a separate career review on how becoming an expert in AI hardware could be very valuable in governance work.

Many politicians and policymakers are generalists, as their roles require them to work in many different subject areas and on different types of problems. This means they’ll need to rely on expert knowledge when crafting and implementing policy on AI technology that they don’t fully understand. So if you can provide them this information, especially if you’re skilled at communicating it clearly, you can potentially fill influential roles.

Some people who may have initially been interested in pursuing a technical AI safety career, but who have found that they either are no longer interested in that path or find more promising policy opportunities, might also decide that they can effectively pivot into a policy-oriented career.

It is common for people with STEM backgrounds to enter and succeed in US policy careers. People with technical credentials that they may regard as fairly modest — such as computer science bachelor’s degrees or a master’s in machine learning — often find their knowledge is highly valued in Washington, DC.

Most DC jobs don’t have specific degree requirements, so you don’t need to have a policy degree to work in DC. Roles specifically addressing science and technology policy are particularly well-suited for people with technical backgrounds, and people hiring for these roles will value higher credentials like a master’s or, better even, a terminal degree like a PhD or MD.

There are many fellowship programmes specifically aiming to support people with STEM backgrounds to enter policy careers; some are listed below.

This won’t be right for everybody — many people with technical skills may not have the disposition or skills necessary for engaging in policy. People in policy-related paths often benefit from strong writing and social skills as well as a comfort navigating bureaucracies and working with people holding very different motivations and worldviews.

Other specific forms of career capital

There are other ways to gain useful career capital that could be applied in this career path.

• If you have or gain great communication skills as, say, a journalist or an activist, these skills could be very useful in advocacy and lobbying around AI governance.
  • Especially since advocacy around AI issues is still in its early stages, it will likely need people with experience advocating in other important cause areas to share their knowledge and skills.
• Academics with relevant skill sets are sometimes brought into government for limited stints to serve as advisors in agencies such as the US Office of Science and Technology. This isn’t necessarily the foundation of a longer career in government, though it can be, and it can give an academic deeper insight into policy and politics than they might otherwise gain.
• You can work at an AI lab in non-technical roles, gaining a deeper familiarity with the technology, the business, and the culture. (Read our career review discussing the pros and cons of working at a top AI lab.)
• You could work on political campaigns and get involved in party politics. This is one way to get involved in legislation, learn about policy, and help impactful lawmakers, and you can also potentially help shape the discourse around AI governance. Note, though, the previously mentioned downsides of potentially polarising public opinion around AI policy; and entering party politics may limit your potential for impact whenever the party you’ve joined doesn’t hold power.
• You could even try to become an elected official yourself, though it’s obviously competitive. If you take this route, make sure you find trustworthy and highly informed advisors to rely on to build expertise in AI, since politicians have many other responsibilities and won’t be able to focus as much on any particular issue.
• You can focus on developing specific skill sets that might be valuable in AI governance, such as information security, intelligence work, diplomacy with China, etc.
  • Other skills: Organisational, entrepreneurial, management, diplomatic, and bureaucratic skills will also likely prove highly valuable in this career path. There may be new auditing agencies to set up or policy regimes to implement. Someone who has worked at high levels in other high-stakes industries, started an influential company, or coordinated complicated negotiations between various groups, would bring important skills to the table.

Where can this kind of work be done?

Since successful AI governance will require work from governments, industry, and other parties, there will be many potential jobs and places to work for people in this path. The landscape will likely shift over time, so if you’re just starting out on this path, the places that seem most important might be different by the time you’re pivoting to using your career capital to make progress on the issue.

Within the US government, for instance, it’s not clear which bodies will be most impactful when it comes to AI policy in five years. It will likely depend on choices that are made in the meantime.

That said, it seems useful to give our understanding of which parts of the government are generally influential in technology governance and most involved right now to help orient. Gaining AI-related experience in government right now should still serve you well if you end up wanting to move into a more impactful AI-related role down the line when the highest-impact areas to work in are clearer.

We’ll also give our current sense of important actors outside government where you might be able to build career capital and potentially have a big impact.

Note that this list has by far the most detail about places to work within the US government. We would like to expand it to include more options as we learn more. You can use this form to suggest additional options for us to include. (And the fact that an option isn’t on this list shouldn’t be taken to mean we recommend against it or even that it would necessarily be less impactful than the places listed.)

We have more detail on other options in separate (and older) career reviews, including the following:

• China-related AI safety and governance paths
• Party politics (UK-focused)
• Policy-oriented government jobs (UK-focused)

With that out of the way, here are some of the places where someone could do promising work or gain valuable career capital:

Our job board features opportunities in AI safety and policy:

View all opportunities

How this career path can go wrong

Doing harm

As we discuss in an article on accidental harm, there are many ways to set back a new field that you’re working in when you’re trying to do good, and this could mean your impact is negative rather than positive. (You may also want to read our article on harmful careers.)

It seems likely there’s a lot of potential to inadvertently cause harm in the emerging field of AI governance. We discussed some possibilities in the section on advocacy and lobbying. Some other possibilities include:

• Pushing for a given policy to the detriment of a superior policy
• Communicating about the risks of AI in a way that ratchets up geopolitical tensions
• Enacting a policy that has the opposite impact of its intended effect
• Setting policy precedents that could be exploited by dangerous actors down the line
• Funding projects in AI that turn out to be dangerous
• Sending the message, implicitly or explicitly, that the risks are being managed when they aren’t, or that they’re lower than they in fact are
• Suppressing technology that would actually be extremely beneficial for society

The trouble is that we have to act with incomplete information, so it may never be very clear when or if people in AI governance are falling into these traps. Being aware that they are potential ways of causing harm will help you keep alert for these possibilities, though, and you should remain open to changing course if you find evidence that your actions may be damaging.

And we recommend keeping in mind the following pieces of general guidance from our article on accidental harm:

1. Ideally, eliminate courses of action that might have a big negative impact.
2. Don’t be a naive optimizer.
3. Have a degree of humility.
4. Develop expertise, get trained, build a network, and benefit from your field’s accumulated wisdom.
5. Follow cooperative norms
6. Match your capabilities to your project and influence.
7. Avoid hard-to-reverse actions.

Burning out

We think this work is exceptionally pressing and valuable, so we encourage our readers who might have a strong personal fit for governance work to test it out. But going into government, in particular, can be difficult. Some people we’ve advised have gone into policy roles with the hope of having an impact, only to burn out and move on.

At the same time, many policy practitioners find their work very meaningful, interesting, and varied.

Some roles in government may be especially challenging for the following reasons:

• Some roles can be very fast-paced, involving relatively high stress and long hours. This is particularly true in Congress and senior executive branch positions and much less so in think tanks or junior agency roles.
• It can take a long time to get into positions with much autonomy or decision-making authority.
• Progress on the issues you care about can be slow, and you often have to work on other priorities. Congressional staffers in particular typically have very broad policy portfolios.
• Work within bureaucracies faces many limitations, which can be frustrating.
• It can be demotivating to work with people who don’t share your values. Though note that policy can select for altruistic people — even if they have different beliefs about how to do good.
• The work isn’t typically well paid relative to comparable positions outside of government.

So we recommend speaking to people in the kinds of positions you might aim to have in order to get a sense of whether the career path would be right for you. And if you do choose to pursue it, look out for signs that the work may be having a negative effect on you and seek support from people who understand what you care about.

If you end up wanting or needing to leave and transition into a new path, that’s not necessarily a loss or a reason for regret. You will likely make important connections and learn a lot of useful information and skills. This career capital can be useful as you transition into another role, perhaps pursuing a complementary approach to AI governance and coordination.

What the increased attention on AI means

We’ve been concerned about risks posed by AI for years. Based on the arguments that this technology could potentially cause a global catastrophe, and otherwise have a dramatic impact on future generations, we’ve advised many people to work to mitigate the risks.

The arguments for the risk aren’t completely conclusive, in our view. But the arguments are worth taking seriously, and given the fact that few others in the world seemed to be devoting much time to even figuring out how big the threat was or how to mitigate it (while at the same time progress in making AI systems more powerful was accelerating) we concluded it was worth ranking among our top priorities.

Now that there’s increased attention on AI, some might conclude that it’s less neglected and thus less pressing to work on. However, the increased attention on AI also makes many interventions potentially more tractable than they had been previously, as policymakers and others are more open to the idea of crafting AI regulations.

And while more attention is now being paid to AI, it’s not clear it will be focused on the most important risks. So there’s likely still a lot of room for important and pressing work positively shaping the development of AI policy.

Read next

If you’re interested in this career path, we recommend checking out some of the following articles next.

• 

  US policy master's degrees

  These degrees are highly valuable for those hoping to take on important roles in the US federal government.

  Read more
• 

  Working in US AI policy

  The US government is likely to be a key actor in how advanced AI is developed and used in society, whether directly or indirectly.

  Read more
• 

  Should you work at a leading AI lab?

  Working at a leading AI lab is an important career option to consider, but the impact of any given role is complex to assess.

  Read more
• 

  Preventing an AI-related catastrophe

  AI might bring huge benefits — if we avoid the risks.

  Read more

Learn more

Top recommendations

• AI Governance Course – AGI Safety Fundamentals from BlueDot Impact
• Podcast: Tantum Collins on what he’s learned as an AI policy insider
• A list of AI policy resources to learn about the field and recent development

Further recommendations

Read next:  Learn about other high-impact careers

Want to consider more paths? See our list of the highest-impact career paths according to our research.

Continue →

The post AI governance and coordination appeared first on 80,000 Hours.

But these risks can be tackled.

With further progress in AI safety, we have an opportunity to develop AI for good: systems that are safe, ethical, and beneficial for everyone.

This article explains how you can help.

Thanks to Adam Gleave, Jacob Hilton and Rohin Shah for reviewing this article. And thanks to Charlie Rogers-Smith for his help, and his article on the topic — How to pursue a career in technical AI alignment.

Why AI safety technical research is high impact

As we’ve argued, in the next few decades, we might see the development of hugely powerful machine learning systems with the potential to transform society. This transformation could bring huge benefits — but only if we avoid the risks.

We think that the worst-case risks from AI systems arise in large part because AI systems could be misaligned — that is, they will aim to do things that we don’t want them to do. In particular, we think they could be misaligned in such a way that they develop (and execute) plans that pose risks to humanity’s ability to influence the world, even when we don’t want that influence to be lost.

We think this means that these future systems pose an existential threat to civilisation.

Even if we find a way to avoid this power-seeking behaviour, there are still substantial risks — such as misuse by governments or other actors — which could be existential threats in themselves.

There are many ways in which we could go about reducing the risks that these systems might pose. But one of the most promising may be researching technical solutions that prevent unwanted behaviour — including misaligned behaviour — from AI systems. (Finding a technical way to prevent misalignment in particular is known as the alignment problem.)

In the past few years, we’ve seen more organisations start to take these risks more seriously. Many of the leading industry labs developing AI — including Google DeepMind and OpenAI — have teams dedicated to finding these solutions, alongside academic research groups including at MIT, Oxford, Cambridge, Carnegie Mellon University, and UC Berkeley.

That said, the field is still very new. We think there are only around 300 people working on technical approaches to reducing existential risks from AI systems,¹ which makes this a highly neglected field.

Finding technical ways to reduce this risk could be quite challenging. Any practically helpful solution must retain the usefulness of the systems (remaining economically competitive with less safe systems), and continue to work as systems improve over time (that is, it needs to be ‘scalable’). As we argued in our problem profile, it seems like it might be difficult to find viable solutions, particularly for modern ML (machine learning) systems.

(If you don’t know anything about ML, we’ve written a very very short introduction to ML, and we’ll go into more detail on how to learn about ML later in this article. Alternatively, if you do have ML experience, talk to our team — they can give you personalised career advice, make introductions to others working on these issues, and possibly even help you find jobs or funding opportunities.)

Although it seems hard, there are lots of avenues for more research — and the field really is very young, so there are new promising research directions cropping up all the time. So we think it’s moderately tractable, though we’re highly uncertain.

In fact, we’re uncertain about all of this and have written extensively about reasons we might be wrong about AI risk.

But, overall, we think that — if it’s a good fit for you — going into AI safety technical research may just be the highest-impact thing you can do with your career.

What does this path involve?

AI safety technical research generally involves working as a scientist or engineer at major AI labs, in academia, or in independent nonprofits.

These roles can be very hard to get. You’ll likely need to build up career capital before you end up in a high-impact role (more on this later, in the section on how to enter). That said, you may not need to spend a long time building this career capital — we’ve seen exceptionally talented people move into AI safety from other quantitative fields, sometimes in less than a year.

Most AI safety technical research falls on a spectrum between empirical research (experimenting with current systems as a way of learning more about what will work), and theoretical research (conceptual and mathematical research looking at ways of ensuring that future AI systems are safe).

No matter where on this spectrum you end up working, your career path might look a bit different depending on whether you want to aim at becoming a research lead — proposing projects, managing a team and setting direction — or a contributor — focusing on carrying out the research.

Finally, there are two slightly different roles you might aim for:

• In academia, research is often led by professors — the key distinguishing feature of being a professor is that you’ll also teach classes and mentor grad students (and you’ll definitely need a PhD).
• Many (but not all) contributor roles in empirical research are also engineers, often software engineers. Here, we’re focusing on software roles that directly contribute to AI safety research (and which often require some ML background) — we’ve written about software engineering more generally in a separate career review.

We think that research lead roles are probably higher-impact in general. But overall, the impact you could have in any of these roles is likely primarily determined by your personal fit for the role — see the section on how to predict your fit in advance.

Next, we’ll take a look at what working in each path might involve. Later, we’ll go into how you might enter each path.

What does work in the empirical AI safety path involve?

Empirical AI safety tends to involve teams working directly with ML models to identify any risks and develop ways in which they might be mitigated.

That means the work is focused on current ML techniques and techniques that might be applied in the very near future.

Practically, working on empirical AI safety involves lots of programming and ML engineering. You might, for example, come up with ways you could test the safety of existing systems, and then carry out these empirical tests.

You can find roles in empirical AI safety in industry and academia, as well as some in AI safety-focused nonprofits.

Particularly in academia, lots of relevant work isn’t explicitly labelled as being focused on existential risk — but it can still be highly valuable. For example, work in interpretability, adversarial examples, diagnostics and backdoor learning, among other areas, could be highly relevant to reducing the chance of an AI-related catastrophe.

We’re also excited by experimental work to develop safety standards that AI companies might adhere to in the future — for example, the work being carried out by METR.

To learn more about the sorts of research taking place at labs focused on empirical AI safety, take a look at:

• OpenAI’s approach to alignment research
• Anthropic’s views on AI safety
• Redwood Research’s recent research highlights
• Publications from Google DeepMind’s safety team

While programming is central to all empirical work, generally, research lead roles will be less focused on programming; instead, they need stronger research taste and theoretical understanding. In comparison, research contributors need to be very good at programming and software engineering.

What does work in the theoretical AI safety path involve?

Theoretical AI safety is much more heavily conceptual and mathematical. Often it involves careful reasoning about the hypothetical behaviour of future systems.

Generally, the aim is to come up with properties that it would be useful for safe ML algorithms to have. Once you have some useful properties, you can try to develop algorithms with these properties (bearing in mind that to be practically useful these algorithms will have to end up being adopted by industry). Alternatively, you could develop ways of checking whether systems have these properties. These checks could, for example, help hold future AI products to high safety standards.

Many people working in theoretical AI safety will spend much of their time proving theorems or developing new mathematical frameworks. More conceptual approaches also exist, although they still tend to make heavy use of formal frameworks.

Some examples of research in theoretical AI safety include:

• Risks from learned optimisation in advanced machine learning systems by Hubinger et al.
• Eliciting latent knowledge by Christiano, Cotra and Xu.
• Formalizing the presumption of independence by Christiano, Neyman, and Xu
• Discovering agents by Kenton et al.
• Active reward learning from multiple teachers by Barnett et al.

There are generally fewer roles available in theoretical AI safety work, especially as research contributors. Theoretical research contributor roles exist at nonprofits (primarily the Alignment Research Center), as well as at some labs (for example, Anthropic’s work on conditioning predictive models and the Causal Incentives Working Group at Google DeepMind). Most contributor roles in theoretical AI safety probably exist in academia (for example, PhD students in teams working on projects relevant to theoretical AI safety).

Some exciting approaches to AI safety

There are lots of technical approaches to AI safety currently being pursued. Here are just a few of them:

• Scalably learning from human feedback. Examples include iterated amplification, AI safety via debate, building AI assistants that are uncertain about our goals and learn them by interacting with us, and other ways to get AI systems trained with stochastic gradient descent to report truthfully what they know.
• Threat modelling. An example of this work would be demonstrating the possibility of (allowing us to study) dangerous capabilities, like deceptive or manipulative AI systems. You can read an overview in a recent Google DeepMind paper. This work splits into work that evaluates whether a model has dangerous capabilities (like the work of METR in evaluating GPT-4), and work that evaluates whether a model would cause harm in practice (like Anthropic’s research into the behaviour of large language models and this paper on goal misgeneralisation).
• Interpretability research. This work involves studying why AI systems do what they do and trying to put it into human-understandable terms. For example, this paper examined how AlphaZero learns chess, and this paper looked into finding latent knowledge in language models without supervision. This category also includes mechanistic interpretability — for example, Zoom In: An Introduction to Circuits by Olah et al.). For more, see this survey paper, as well as Hubinger’s a transparency and interpretability tech tree, and Nanda’s A Longlist of Theories of Impact for Interpretability for overviews of of how interpretability research could reduce existential risk from AI.
• Other anti-misuse research to reduce the risks of catastrophe caused by misuse of systems. (We’ve written more on this in our problem profile on AI risk). For example, this work includes training AIs so they’re hard to use for dangerous purposes. (Note there’s lots of overlap with the other work on this list).
• Research to increase the robustness of neural networks. This work involves ensuring that the sorts of behaviour neural networks display when exposed to one set of inputs continues when exposed to inputs they haven’t previously been exposed to, in order to prevent AI systems changing to unsafe behaviour. See section 2 of Unsolved Problems in AI safety for more.
• Work to build cooperative AI. Find ways to ensure that even if individual AI systems seem safe, they don’t produce bad outcomes through interacting with other sociotechnical systems. For more, see Open Problems in Cooperative AI by Dafoe et al. or the Cooperative AI Foundation. This seems particularly relevant for the reduction of ‘s-risks.’
• More generally, there are some unified safety plans. For more, see Hubinger’s 11 possible proposals for building safe advanced AI, or Karnofsky’s How might we align transformative AI if it’s developed very soon.²

It’s worth noting that there are many approaches to AI safety, and people in the field strongly disagree on what will or won’t work.

This means that, once you’re working in the field, it can be worth being charitable and careful not to assume that others’ work is unhelpful just because it seemed so on a quick skim. You should probably be uncertain about your own research agenda as well.

What’s more, as we mentioned earlier, lots of relevant work across all these areas isn’t explicitly labelled ‘safety.’

So it’s important to think carefully about how or whether any particular research helps reduce the risks that AI systems might pose.

What are the downsides of this career path?

AI safety technical research is not the only way to make progress on reducing the risks that future AI systems might pose. Also, there are many other pressing problems in the world that aren’t the possibility of an AI-related catastrophe, and lots of careers that can help with them. If you’d be a better fit working on something else, you should probably do that.

Beyond personal fit, there are a few other downsides to the career path:

• It can be very competitive to enter (although once you’re in, the jobs are well paid, and there are lots of backup options).
• You need quantitative skills — and probably programming skills.
• The work is geographically concentrated in just a few places (mainly the California Bay Area and London, but there are also opportunities in places with top universities such as Oxford, New York, Pittsburgh, and Boston). That said, remote work is increasingly possible at many research labs.
• It might not be very tractable to find good technical ways of reducing the risk. Although assessments of its difficulty vary, and while making progress is almost certainly possible, it may be quite hard to do so. This reduces the impact that you could have working in the field. That said, if you start out in technical work you might be able to transition to governance work, since that often benefits from technical training and experience with the industry, which most people do not have.)
• Relatedly, there’s lots of disagreement in the field about what could work; you’ll probably be able to find at least some people who think what you’re working on is useless, whatever you end up doing.
• Most importantly, there’s some risk of doing harm. While gaining career capital, and while working on the research itself, you’ll have to make difficult decisions and judgement calls about whether you’re working on something beneficial (see our anonymous advice about working in roles that advance AI capabilities). There’s huge disagreement on which technical approaches to AI safety might work — and sometimes this disagreement takes the form of thinking that a strategy will actively increase existential risks from AI.

Finally, we’ve written more about the best arguments against AI being pressing in our problem profile on preventing an AI-related catastrophe. If those are right, maybe you could have more impact working on a different issue.

How much do AI safety technical researchers earn?

Many technical researchers work at companies or small startups that pay wages competitive with the Bay Area and Silicon Valley tech industry, and even smaller organisations and nonprofits will pay competitive wages to attract top talent. The median compensation for a software engineer in the San Francisco Bay area was $222,000 per year in 2020.³ (Read more about software engineering salaries).

This $222,000 median may be an underestimate, as AI roles, especially in top AI labs that are rapidly scaling up their work in AI, often pay better than other tech jobs, and the same applies to safety researchers — even those in nonprofits.

However, academia has lower salaries than industry in general, and we’d guess that AI safety research roles in academia pay less than commercial labs and nonprofits.

Examples of people pursuing this path

How to predict your fit in advance

You’ll generally need a quantitative background (although not necessarily a background in computer science or machine learning) to enter this career path.

There are two main approaches you can take to predict your fit, and it’s helpful to do both:

• Try it out: try out the first few steps in the section below on learning the basics. If you haven’t yet, try learning some python, as well as taking courses in linear algebra, calculus, and probability. And if you’ve done that, try learning a bit about deep learning and AI safety. Finally, the best way to try this out for many people would be to actually get a job as a (non-safety) ML engineer (see more in the section on how to enter).
• Talk to people about whether it would be a good fit for you: If you want to become a technical researcher, our team probably wants to talk to you. We can give you 1-1 advice, for free. If you know anyone working in the area (or something similar), discuss this career path with them and ask for their honest opinion. You may be able to meet people through our community. Our advisors can also help make connections.

It can take some time to build expertise, and enjoyment can follow expertise — so be prepared to take some time to learn and practice before you decide to switch to something else entirely.

If you’re not sure what roles you might aim for longer term, here are a few rough ways you could make a guess about what to aim for, and whether you might be a good fit for various roles on this path:

• Testing your fit as an empirical research contributor: In a blog post about hiring for safety researchers, the Google DeepMind team said “as a rough test for the Research Engineer role, if you can reproduce a typical ML paper in a few hundred hours and your interests align with ours, we’re probably interested in interviewing you.”
  • Looking specifically at software engineering, one hiring manager at Anthropic said that if you could, with a few weeks’ work, write a complex new feature or fix a very serious bug in a major ML library, they’d want to interview you straight away. (Read more.)
• Testing your fit for theoretical research: If you could have got into a top 10 maths or theoretical computer science PhD programme if you’d optimised your undergrad to do so, that’s a decent indication of your fit (and many researchers in fact have these PhDs). The Alignment Research Center (one of the few organisations that hires for theoretical research contributors, as of 2023) said that they were open to hiring people without any research background. They gave four tests of fit: creativity (e.g. you may have ideas for solving open problems in the field, like Eliciting Latent Knowledge); experience designing algorithms, proving theorems, or formalising concepts; broad knowledge of maths and computer science; and having thought a lot about the AI alignment problem in particular.
• Testing your fit as a research lead (or for a PhD): The vast majority of research leads have a PhD. Also, many (but definitely not all) AI safety technical research roles will require a PhD — and if they don’t, having a PhD (or being the sort of person that could get one) would definitely help show that you’re a good fit for the work. To get into a top 20 machine learning PhD programme, you’d probably need to publish something like a first author workshop paper, as well as a third author conference paper at a major ML conference (like NeurIPS or ICML). (Read more about whether you should do a PhD).

Read our article on personal fit to learn more about how to assess your fit for the career paths you want to pursue.

How to enter

You might be able to apply for roles right away — especially if you meet, or are near meeting, the tests we just looked at — but it also might take you some time, possibly several years, to skill up first.

So, in this section, we’ll give you a guide to entering technical AI safety research. We’ll go through four key questions:

1. How to learn the basics
2. Whether you should do a PhD
3. How to get a job in empirical research
4. How to get a job in theoretical research

Hopefully, by the end of the section, you’ll have everything you need to get going.

Learning the basics

To get anywhere in the world of AI safety technical research, you’ll likely need a background knowledge of coding, maths, and deep learning.

You might also want to practice enough to become a decent ML engineer (although this is generally more useful for empirical research), and learn a bit about safety techniques in particular (although this is generally more useful for empirical research leads and theoretical researchers).

We’ll go through each of these in turn.

Learning to program

You’ll probably want to learn to code in python, because it’s the most widely used language in ML engineering.

The first step is probably just trying it out. As a complete beginner, you can write a Python program in less than 20 minutes that reminds you to take a break every two hours. Don’t be discouraged if your code doesn’t work the first time — that’s what normally happens when people code!

Once you’ve done that, you have a few options:

• Teach yourself to program. Try working through a free beginner course like Automate the boring stuff with Python by Al Seigart. There also are many great introductory computer science and programming courses online, including: Udacity’s Intro to Computer Science, MIT’s Introduction to Computer Science and Programming, and Stanford’s Programming Methodology. Then, try finding something you want to build, and building it — or getting involved in an open-source project. For interview practice, try leetcode or TopCoder, or the exercises in Cracking the Coding Interview by Gayle McDowell.
• Take a college course. If you’re in university, this is a great option because it allows you to learn programming while the opportunity cost of your time is lower. You can even consider majoring in computer science (or another subject involving lots of programming).
• Learn on the job. If you can find internships, you’ll gain practical experience and skills you otherwise wouldn’t pick up from academic degrees.
• Go to a bootcamp. Coding bootcamps are focused on taking people with little knowledge of programming to as highly paid a job as possible within a couple of months — though some claim the long-term prospects are not as good because you lack a deep understanding of computer science. Course Report is a great guide to choosing a bootcamp. Be careful to avoid low-quality bootcamps. You can also find online bootcamps — for people completely new to programming — focused on ML, like Udemy’s Python for Data Science and Machine Learning Bootcamp.

You can read more about learning to program — and how to get your first job in software engineering (if that’s the route you want to take) — in our career review on software engineering.

Learning the maths

The maths of deep learning relies heavily on calculus and linear algebra, and statistics can be useful too — although generally learning the maths is much less important than programming and basic, practical ML.

We’d generally recommend studying a quantitative degree (like maths, computer science or engineering), most of which will cover all three areas pretty well.

If you want to actually get good at maths, you have to be solving problems. So, generally, the most useful thing that textbooks and online courses provide isn’t their explanations — it’s a set of exercises to try to solve, in order, with some help if you get stuck.

If you want to self-study (especially if you don’t have a quantitative degree) here are some possible resources:

• Calculus: 3blue1brown’s video series on calculus could be a good place to start. You may also be able to follow recorded university courses: MIT’s single variable calculus (which requires only high school algebra and trigonometry) followed by MIT’s course in vector and multivariable calculus.
• Linear algebra: Again, we’d suggest 3blue1brown’s video series on linear algebra as a place to start. In his post about technical alignment careers, Rogers-Smith recommends Linear Algebra Done Right by Sheldon Axler. Finally, if you prefer lectures, try MIT’s undergraduate course in linear algebra (although note that this course assumes knowledge of multivariate calculus).
• Probability: Take a look at MIT’s undergraduate course in probability and random variables.

You might be able to find resources that cover all these areas, like Imperial College’s Mathematics for Machine Learning.

Learning basic machine learning

You’ll likely need to have a decent understanding of how AI systems are currently being developed. This will involve learning about machine learning and neural networks, before diving into any specific subfields of deep learning.

Again, there’s the option of covering this at university. If you’re currently at college, it’s worth checking if you can take an ML course even if you’re not majoring in computer science.

There’s one important caveat here: you’ll learn a huge amount on the job, and the amount you’ll need to know in advance for any role or course will vary hugely! Not even top academics know everything about their fields. It’s worth trying to find out how much you’ll need to know for the role you want to do before you invest hundreds of hours into learning about ML.

With that caveat in mind, here are some suggestions of places you might start if you want to self-study the basics:

• 3blue1brown’s series on neural networks is a really great place to start for beginners.
• When I was learning, I used Neural Networks and Deep Learning — it’s an online textbook, good if you’re familiar with the maths, with some helpful exercises as well.
• Online intro courses like fast.ai (focused on practical applications), Full Stack Deep Learning, and the various courses at deeplearning.ai.
• For more detail, see university courses like MIT’s *Introduction to Machine Learning, NYU’s Deep Learning for even more detail. We’d also recommend Google DeepMind’s lecture series.

PyTorch is a very common package used for implementing neural networks, and probably worth learning! When I was first learning about ML, my first neural network was a 3-layer convolutional neural network with L2 regularisation classifying characters from the MNIST database. This is a pretty common first challenge, and a good way to learn PyTorch.

Learning about AI safety

If you’re going to work as an AI safety researcher, it usually helps to know about AI safety.

This isn’t always true — some engineering roles won’t require much knowledge of AI safety. But even then, knowing the basics will probably help land you a position, and can also help with things like making difficult judgement calls and avoiding doing harm. And if you want to be able to identify and do useful work, you’ll need to learn about the field eventually.

Because the field is still so new, there probably aren’t (yet) university courses you can take. So you’ll need to do some self-study. Here are some places you might start:

• Section 3 of our problem profile about preventing an AI-related catastrophe provides an introduction to the problems that AI safety attempts to solve (with a particular focus on alignment).
• Rob Miles’ YouTube channel is full of popular and well-explained introductory videos that don’t need much background knowledge of ML.
• AXRP – the AI X-risk Research Podcast — is full of in-depth (and enjoyable) conversations with researchers about their research.
• The courses from AGI Safety Fundamentals, in particular the AI Alignment Course, possibly followed by Alignment 201, which provide an introduction to research on the alignment problem.
• Intro to ML Safety, a course from the Center for AI Safety focuses on withstanding hazards (“robustness”), identifying hazards (“monitoring”), and reducing systemic hazards (“systemic safety”), as well as alignment.

For more suggestions — especially when it comes to reading about the nature of the risks we might face from AI systems — take a look at the top resources to learn more from our problem profile.

Should you do a PhD?

Some technical research roles will require a PhD — but many won’t, and PhDs aren’t the best option for everyone.

The main benefit of doing a PhD is probably practising setting and carrying out your own research agenda. As a result, getting a PhD is practically the default if you want to be a research lead.

That said, you can also become a research lead without a PhD — in particular, by transitioning from a role as a research contributor. At some large labs, the boundary between being a contributor and a lead is increasingly blurry.

Many people find PhDs very difficult. They can be isolating and frustrating, and take a very long time (4–6 years). What’s more, both your quality of life and the amount you’ll learn will depend on your supervisor — and it can be really difficult to figure out in advance whether you’re making a good choice.

So, if you’re considering doing a PhD, here are some things to consider:

• Your long-term vision: If you’re aiming to be a research lead, that suggests you might want to do a PhD — the vast majority of research leads have PhDs. If you mainly want to be a contributor (e.g. an ML or software engineer), that suggests you might not. If you’re unsure, you should try doing something to test your fit for each, like trying a project or internship. You might try a pre-doctoral research assistant role — if the research you do is relevant to your future career, these can be good career capital, whether or not you do a PhD.
• The topic of your research: It’s easy to let yourself become tied down to a PhD topic you’re not confident in. If the PhD you’re considering would let you work on something that seems useful for AI safety, it’s probably — all else equal — better for your career, and the research itself might have a positive impact as well.
• Mentorship: What are the supervisors or managers like at the opportunities open to you? You might be able to find ML engineering or research roles in industry where you could learn much more than you would in a PhD — or vice versa. When picking a supervisor, try reaching out to the current or former students of a prospective supervisor and asking them some frank questions. (Also, see this article on how to choose a PhD supervisor.)
• Your fit for the work environment: Doing a PhD means working on your own with very little supervision or feedback for long periods of time. Some people thrive in these conditions! But some really don’t and find PhDs extremely difficult.

Read more in our more detailed (but less up-to-date) review of machine learning PhDs.

It’s worth remembering that most jobs don’t need a PhD. And for some jobs, especially empirical research contributor roles, even if a PhD would be helpful, there are often better ways of getting the career capital you’d need (for example, working as a software or ML engineer). We’ve interviewed two ML engineers who have had hugely successful careers without doing a PhD.

Whether you should do a PhD doesn’t depend (much) on timelines

We think it’s plausible that we will develop AI that could be hugely transformative for society by the end of the 2030s.

All else equal, that possibility could argue for trying to have an impact right away, rather than spending five (or more) years doing a PhD.

Ultimately, though, how well you, in particular, are suited to a particular PhD is probably a much more important factor than when AI will be developed.

That is to say, we think the increase in impact caused by choosing a path that’s a good fit for you is probably larger than any decrease in impact caused by delaying your work. This is in part because the spread in impact caused by the specific roles available to you, as well as your personal fit for them, is usually very large. Some roles (especially research lead roles) will just require having a PhD, and others (especially more engineering-heavy roles) won’t — and people’s fit for these paths varies quite a bit.

We’re also highly uncertain about estimates about when we might develop transformative AI. This uncertainty reduces the expected cost of any delay.

Most importantly, we think PhDs shouldn’t be thought of as a pure delay to your impact. You can do useful work in a PhD, and generally, the first couple of years in any career path will involve a lot of learning the basics and getting up to speed. So if you have a good mentor, work environment, and choice of topic, your PhD work could be as good as, or possibly better than, the work you’d do if you went to work elsewhere early in your career. And if you suddenly receive evidence that we have less time than you thought, it’s relatively easy to drop out.

There are lots of other considerations here — for a rough overview, and some discussion, see this post by 80,000 Hours advisor Alex Lawsen, as well as the comments.

Overall, we’d suggest that instead of worrying about a delay to your impact, think instead about which longer-term path you want to pursue, and how the specific opportunities in front of you will get you there.

How to get into a PhD

ML PhDs can be very competitive. To get in, you’ll probably need a few publications (as we said above, something like a first author workshop paper, as well as a third author conference paper at a major ML conference (like NeurIPS or ICML), and references, probably from ML academics. (Although publications also look good whatever path you end up going down!)

To end up at that stage, you’ll need a fair bit of luck, and you’ll also need to find ways to get some research experience.

One option is to do a master’s degree in ML, although make sure it’s a research masters — most ML master’s degrees primarily focus on preparation for industry.

Even better, try getting an internship in an ML research group. Opportunities include RISS at Carnegie Mellon University, UROP at Imperial College London, the Aalto Science Institute international summer research programme, the Data Science Summer Institute, the Toyota Technological Institute intern programme and MILA. You can also try doing an internship specifically in AI safety, for example at CHAI. However, there are sometimes disadvantages to doing internships specifically in AI safety directly — in general, it may be harder to publish and mentorship might be more limited.

Another way of getting research experience is by asking whether you can work with researchers. If you’re already at a top university, it can be easiest to reach out to people working at the university you’re studying at.

PhD students or post-docs can be more responsive than professors, but eventually, you’ll want a few professors you’ve worked with to provide references, so you’ll need to get in touch. Professors tend to get lots of cold emails, so try to get their attention! You can try:

• Getting an introduction, for example from a professor who’s taught you
• Mentioning things you’ve done (your grades, relevant courses you’ve taken, your GitHub, any ML research papers you’ve attempted to replicate as practice)
• Reading some of their papers and the main papers in the field, and mention them in the email
• Applying for funding that’s available to students who want to work in AI safety, and letting people know you’ve got funding to work with them

Ideally, you’ll find someone who supervises you well and has time to work with you (that doesn’t necessarily mean the most famous professor — although it helps a lot if they’re regularly publishing at top conferences). That way, they’ll get to know you, you can impress them, and they’ll provide an amazing reference when you apply for PhDs.

It’s very possible that, to get the publications and references you’ll need to get into a PhD, you’ll need to spend a year or two working as a research assistant, although these positions can also be quite competitive.

This guide by Adam Gleave also goes into more detail on how to get a PhD, including where to apply and tips on the application process itself. We discuss ML PhDs in more detail in our career review on ML PhDs (though it’s outdated compared to this career review).

Getting a job in empirical AI safety research

Ultimately, the best way of learning to do empirical research — especially in contributor and engineering-focused roles — is to work somewhere that does both high-quality engineering and cutting-edge research.

The top three labs are probably Google DeepMind (who offer internships to students), OpenAI (who have a 6-month residency programme) and Anthropic. (Working at a leading AI lab carries with it some risk of doing harm, so it’s important to think carefully about your options. We’ve written a separate article going through the major relevant considerations.)

To end up working in an empirical research role, you’ll probably need to build some career capital.

Whether you want to be a research lead or a contributor, it’s going to help to become a really good software engineer. The best ways of doing this usually involve getting a job as a software engineer at a big tech company or at a promising startup. (We’ve written an entire article about becoming a software engineer.)

Many roles will require you to be a good ML engineer, which means going further than just the basics we looked at above. The best way to become a good ML engineer is to get a job doing ML engineering — and the best places for that are probably leading AI labs.

For roles as a research lead, you’ll need relatively more research experience. You’ll either want to become a research contributor first, or enter through academia (for example by doing a PhD).

All that said, it’s important to remember that you don’t need to know everything to start applying, as you’ll inevitably learn loads on the job — so do try to find out what you’ll need to learn to land the specific roles you’re considering.

How much experience do you need to get a job? It’s worth reiterating the tests we looked at above for contributor roles:

• In a blog post about hiring for safety researchers, the DeepMind team said “as a rough test for the Research Engineer role, if you can reproduce a typical ML paper in a few hundred hours and your interests align with ours, we’re probably interested in interviewing you.”
• Looking specifically at software engineering, one hiring manager at Anthropic said that if you could, with a few weeks’ work, write a new feature or fix a serious bug in a major ML library, they’d want to interview you straight away. (Read more.)

In the process of getting this experience, you might end up working in roles that advance AI capabilities. There are a variety of views on whether this might be harmful — so we’d suggest reading our article about working at leading AI labs and our article containing anonymous advice from experts about working in roles that advance capabilities. It’s also worth talking to our team about any specific opportunities you have.

If you’re doing another job, or a degree, or think you need to learn some more before trying to change careers, there are a few good ways of getting more experience doing ML engineering that go beyond the basics we’ve already covered:

• Getting some experience in software / ML engineering. For example, if you’re doing a degree, you might try an internship as a software engineer during the summer. DeepMind offer internships for students with at least two years of study in a technical subject,
• Replicating papers. One great way of getting experience doing ML engineering, is to replicate some papers in whatever sub-field you might want to work in. Richard Ngo, an AI governance researcher at OpenAI, has written some advice on replicating papers. But bear in mind that replicating papers can be quite hard — take a look at Amid Fish’s blog on what he learned replicating a deep RL paper. Finally, Rogers-Smith has some suggestions on papers to replicate. If you do spend some time replicating papers, remember that when you get to applying for roles, it will be really useful to be able to prove you’ve done the work. So try uploading your work to GitHub, or writing a blog on your progress. And if you’re thinking about spending a long time on this (say, over 100 hours), try to get some feedback on the papers you might replicate before you start — you could even reach out to a lab you want to work for.
• Taking or following a more in-depth course in empirical AI safety research. Redwood Research ran the MLAB bootcamp, and you can apply for access to their curriculum here. You could also take a look at this Deep Learning Curriculum by Jacob Hilton, a researcher at the Alignment Research Center — although it’s probably very challenging without mentorship.⁴ The Alignment Research Engineer Accelerator is a program that uses this curriculum. Some mentors on the SERI ML Alignment Theory Scholars Program focus on empirical research.
• Learning about a sub-field of deep learning. In particular, we’d suggest natural language processing (in particular transformers — see this lecture as a starting point) and reinforcement learning (take a look at Pong from Pixels by Andrej Karpathy, and OpenAI’s Spinning up in Deep RL). Try to get to the point where you know about the most important recent advances.

Finally, Athena is an AI alignment mentorship program for women with a technical background looking to get jobs in the alignment field

Getting a job in theoretical AI safety research

There are fewer jobs available in theoretical AI safety research, so it’s harder to give concrete advice. Having a maths or theoretical computer science PhD isn’t always necessary, but is fairly common among researchers in industry, and is pretty much required to be an academic.

If you do a PhD, ideally it’d be in an area at least somewhat related to theoretical AI safety research. For example, it could be in probability theory as applied to AI, or in theoretical CS (look for researchers who publish in COLT or FOCS).

Alternatively, one path is to become an empirical research lead before moving into theoretical research.

Compared to empirical research, you’ll need to know relatively less about engineering, and relatively more about AI safety as a field.

Once you’ve done the basics, one possible next step you could try is reading papers from a particular researcher, or on a particular topic, and summarising what you’ve found.

You could also try spending some time (maybe 10–100 hours) reading about a topic and then some more time (maybe another 10–100 hours) trying to come up with some new ideas on that topic. For example, you could try coming up with proposals to solve the problem of eliciting latent knowledge. Alternatively, if you wanted to focus on the more mathematical side, you could try having a go at the assignment at the end of this lecture by Michael Cohen, a grad student at the University of Oxford.

If you want to enter academia, reading a ton of papers seems particularly important. Maybe try writing a survey paper on a certain topic in your spare time. It’s a great way to master a topic, spark new ideas, spot gaps, and come up with research ideas. When applying to grad school or jobs, your paper is a fantastic way to show you love research so much you do it for fun.

There are some research programmes aimed at people new to the field, such as the SERI ML Alignment Theory Scholars Program, to which you could apply.

Other ways to get more concrete experience include doing research internships, working as a research assistant, or doing a PhD, all of which we’ve written about above, in the section on whether and how you can get into a PhD programme.

One note is that a lot of people we talk to try to learn independently. This can be a great idea for some people, but is fairly tough for many, because there’s substantially less structure and mentorship.

Recommended organisations

AI labs in industry that have empirical technical safety teams, or are focused entirely on safety:

• Anthropic is an AI safety company working on building interpretable and safe AI systems. They focus on empirical AI safety research. Anthropic cofounders Daniela and Dario Amodei gave an interview about the lab on the Future of Life Institute podcast. On our podcast, we spoke to Chris Olah, who leads Anthropic’s research into interpretability, and Nova DasSarma, who works on systems infrastructure at Anthropic.
• METR works on assessing whether cutting-edge AI systems could pose catastrophic risks to civilization, including early-stage, experimental work to develop techniques, and evaluating systems produced by Anthropic and OpenAI.
• The Center for AI Safety is a nonprofit that does technical research and promotion of safety in the wider machine learning community.
• FAR AI is a research nonprofit that incubates and accelerates research agendas that are too resource-intensive for academia but not yet ready for commercialisation by industry, including research in adversarial robustness, interpretability and preference learning.
• Google DeepMind is probably the largest and most well-known research group developing general artificial machine intelligence, and is famous for its work creating AlphaGo, AlphaZero, and AlphaFold. It is not principally focused on safety, but has two teams focused on AI safety, with the Scalable Alignment Team focusing on aligning existing state-of-the-art systems, and the Alignment Team focused on research bets for aligning future systems.
• OpenAI, founded in 2015, is a lab that is trying to build artificial general intelligence that is safe and benefits all of humanity. OpenAI is well known for its language models like GPT-4. Like DeepMind, it is not principally focused on safety, but has a safety team and a governance team. Jan Leike (co-lead of the superalignment team) has some blog posts on how he thinks about AI alignment, and has spoken on our podcast about the sorts of people he’d like to hire for his team.
• Ought is a machine learning lab building Elicit, an AI research assistant. Their aim is to align open-ended reasoning by learning human reasoning steps, and to direct AI progress towards helping with evaluating evidence and arguments.
• Redwood Research is an AI safety research organisation, whose first big project attempted to make sure language models (like GPT-3) produce output following certain rules with very high probability, in order to address failure modes too rare to show up in standard training.

Theoretical / conceptual AI safety labs:

• The Alignment Research Center (ARC) is attempting to produce alignment strategies that could be adopted in industry today while also being able to scale to future systems. They focus on conceptual work, developing strategies that could work for alignment and which may be promising directions for empirical work, rather than doing empirical AI work themselves. Their first project was releasing a report on Eliciting Latent Knowledge, the problem of getting advanced AI systems to honestly tell you what they believe (or ‘believe’) about the world. On our podcast, we interviewed ARC founder Paul Christiano about his research (before he founded ARC).
• The Center on Long-Term Risk works to address worst-case risks from advanced AI. They focus on conflict between AI systems.
• The Machine Intelligence Research Institute was one of the first groups to become concerned about the risks from machine intelligence in the early 2000s, and its team has published a number of papers on safety issues and how to resolve them.
• Some teams in commercial labs also do some more theoretical and conceptual work on alignment, such as Anthropic’s work on conditioning predictive models and the Causal Incentives Working Group at Google DeepMind.

AI safety in academia (a very non-comprehensive list; while the number of academics explicitly and publicly focused on AI safety is small, it’s possible to do relevant work at a much wider set of places):

• The Algorithmic Alignment Group in the Computer Science and Artificial Intelligence Laboratory at MIT, led by Dylan Hadfield-Menell
• The Center for Human-Compatible AI at UC Berkeley, led by Stuart Russell, focuses on academic research to ensure AI is safe and beneficial to humans. (Our podcast with Stuart Russell examines his approach to provably beneficial AI.)
• Jacob Steinhardt’s research group in the Department of Statistics at UC Berkeley
• The NYU Alignment research Group led by Sam Bowman
• David Krueger’s research group at the Computational and Biological Learning Laboratory at the University of Cambridge
• The Foundations of Cooperative AI Lab at Carnegie Mellon University
• The Future of Humanity Institute at the University of Oxford has an AI safety research group
• The Alignment of Complex Systems research group at Charles University, Prague

Find a job in this path

If you think you might be a good fit for this path and you’re ready to start looking at job opportunities that are currently accepting applications, see our curated list of opportunities for this path:

View all opportunities

Learn more about AI safety technical research

Top recommendations

• The AGI safety fundamentals technical alignment curriculum
• The 80,000 Hours Podcast on Artificial Intelligence (a collection of 10 key AI episodes from our podcast)
• Charlie Rogers-Smith’s step-by-step guide to AI safety careers (which this article is in large part based on) provides some helpful concrete advice, including ways you might get some funding to help you move into an AI safety technical research career.

Further recommendations

Read next:  Learn about other high-impact careers

Want to consider more paths? See our list of the highest-impact career paths according to our research.

Continue →

The post AI safety technical research appeared first on 80,000 Hours.

Note from the author: At its core, this problem profile tries to predict the future of technology. This is a notoriously difficult thing to do. In addition, there has been much less rigorous research into the risks from AI than into the other risks 80,000 Hours writes about (like pandemics or climate change).¹ That said, there is a growing field of research into the topic, which I’ve tried to reflect. For this article I’ve leaned especially on this draft report by Joseph Carlsmith at Open Philanthropy (also available as a narration), as it’s the most rigorous overview of the risk that I could find. I’ve also had the article reviewed by over 30 people with different expertise and opinions on the topic. (Almost all are concerned about advanced AI’s potential impact.)

If you have any feedback on this article — whether there’s something technical we’ve got wrong, some wording we could improve, or just that you did or didn’t like reading it — we’d really appreciate it if you could tell us what you think using this form.

Why do we think that reducing risks from AI is one of the most pressing issues of our time? In short, our reasons are:

1. Even before getting into the actual arguments, we can see some cause for concern — as many AI experts think there’s a small but non-negligible chance that AI will lead to outcomes as bad as human extinction.
2. We’re making advances in AI extremely quickly — which suggests that AI systems could have a significant influence on society, soon.
3. There are strong arguments that “power-seeking” AI could pose an existential threat to humanity² — which we’ll go through below.
4. Even if we find a way to avoid power-seeking, there are still other risks.
5. We think we can tackle these risks.
6. This work is neglected.

We’re going to cover each of these in turn, then consider some of the best counterarguments, explain concrete things you can do to help, and finally outline some of the best resources for learning more about this area.

1. Many AI experts think there’s a non-negligible chance AI will lead to outcomes as bad as extinction

In May 2023, hundreds of AI prominent scientists — and other notable figures — signed a statement saying that mitigating the risk of extinction from AI should be a global priority.

So it’s pretty clear that at least some experts are concerned.

But how concerned are they? And is this just a fringe view?

We looked at three surveys of AI researchers who published at NeurIPS and ICML (two of the most prestigious machine learning conferences) — one in 2016, one in 2019, and one in 2022.³

It’s important to note that there could be considerable selection bias on surveys like this. For example, you might think researchers who go to the top AI conferences are more likely to be optimistic about AI, because they have been selected to think that AI research is doing good. Alternatively, you might think that researchers who are already concerned about AI are more likely to respond to a survey asking about these concerns.⁴

All that said, here’s what we found:

In all three surveys, the median researcher thought that the chances that AI would be “extremely good” was reasonably high: 20% in the 2016 survey, 20% in 2019, and 10% in 2022.⁵

Indeed, AI systems are already having substantial positive effects — for example, in medical care or academic research.

But in all three surveys, the median researcher also estimated small — and certainly not negligible — chances that AI would be “extremely bad (e.g. human extinction)”: a 5% chance of extremely bad outcomes in the 2016 survey, 2% in 2019, and 5% in 2022.^{6 7}

In the 2022 survey, participants were specifically asked about the chances of existential catastrophe caused by future AI advances — and again, over half of researchers thought the chances of an existential catastrophe was greater than 5%.⁸

So experts disagree on the degree to which AI poses an existential risk — a kind of threat we’ve argued deserves serious moral weight.

This fits with our understanding of the state of the research field. Three of the leading labs developing AI — DeepMind, Anthropic and OpenAI — also have teams dedicated to figuring out how to solve technical safety issues that we believe could, for reasons we discuss at length below, lead to an existential threat to humanity.⁹

There are also several academic research groups (including at MIT, Oxford, Cambridge, Carnegie Mellon University, and UC Berkeley) focusing on these same technical AI safety problems.¹⁰

It’s hard to know exactly what to take from all this, but we’re confident that it’s not a fringe position in the field to think that there is a material risk of outcomes as bad as an existential catastrophe. Some experts in the field maintain, though, that the risks are overblown.

Still, why do we side with those who are more concerned? In short, it’s because there are arguments we’ve found persuasive that AI could pose such an existential threat — arguments we will go through step by step below.

It’s important to recognise that the fact that many experts recognise there’s a problem doesn’t mean that everything’s OK, the experts have got it covered. Overall, we think this problem remains highly neglected, with only around 400 people working directly on the issue worldwide (more on this below).

Meanwhile, there are billions of dollars a year going into making AI more advanced.¹¹

2. We’re making advances in AI extremely quickly

Before we try to figure out what the future of AI might look like, it’s helpful to take a look at what AI can already do.

Modern AI techniques involve machine learning (ML): models that improve automatically through data input. The most common form of this technique used today is known as deep learning.

ML systems today can only perform a very small portion of tasks that humans can do, and (with a few exceptions) only within narrow specialties (like playing one particular game or generating one particular kind of image).

That said, since the increasingly widespread use of deep learning in the mid-2010s, there has been huge progress in what can be achieved with ML. Here’s a brief timeline of only some of the advances we saw from 2019 to 2022:

• AlphaStar, which can beat top professional players at StarCraft II (January 2019)
• MuZero, a single system that learned to win games of chess, shogi, and Go — without ever being told the rules (November 2019)
• GPT-3, a natural language model capable of producing high-quality text (May 2020)
• GPT-f, which can solve some Maths Olympiad problems (September 2020)
• AlphaFold 2, a huge step forward in solving the long-perplexing protein-folding problem (July 2021)
• Codex, which can produce code for programs from natural language instructions (August 2021)
• PaLM, a language model which has shown impressive capabilities to reason about things like cause and effect or explaining jokes (April 2022)
• DALL-E 2 (April 2022) and Imagen (May 2022), which are both capable of generating high-quality images from written descriptions
• SayCan, which takes natural language instructions and uses them to operate a robot (April 2022)
• Gato, a single ML model capable of doing a huge number of different things (including playing Atari, captioning images, chatting, and stacking blocks with a real robot arm), deciding based on its context what it should output (May 2022)
• Minerva can solve complex maths problems — fairly well at college level, and even better at high school maths competition level. (Minerva is far more successful than forecasters predicted in 2021.)

If you’re anything like us, you found the complexity and breadth of the tasks these systems can carry out surprising.

And if the technology keeps advancing at this pace, it seems clear there will be major effects on society. At the very least, automating tasks makes carrying out those tasks cheaper. As a result, we may see rapid increases in economic growth (perhaps even to the level we saw during the Industrial Revolution).

If we’re able to partially or fully automate scientific advancement we may see more transformative changes to society and technology.¹³

That could be just the beginning. We may be able to get computers to eventually automate anything humans can do. This seems like it has to be possible — at least in principle. This is because it seems that, with enough power and complexity, a computer should be able to simulate the human brain. This would itself be a way of automating anything humans can do (if not the most efficient method of doing so).

And as we’ll see in the next section, there are some indications that extensive automation may well be possible through scaling up existing techniques.

Current trends show rapid progress in the capabilities of ML systems

There are three things that are crucial to building AI through machine learning:

1. Good algorithms (e.g. more efficient algorithms are better)
2. Data to train an algorithm
3. Enough computational power (known as compute) to do this training

We spoke to Danny Hernandez, who (at the time) was a research scientist on the Foresight team at OpenAI. Hernandez and his team looked at how two of these inputs (compute and algorithm efficiency) are changing over time.

They found that, since 2012, the amount of compute used for training the largest AI models has been rising exponentially — doubling every 3.4 months.

That is to say, since 2012, the amount of computational power used to train our largest machine learning models has grown by over 1 billion times.

Hernandez and his team also looked at how much compute has been needed to train a neural network to have the same performance as AlexNet (an early image classification algorithm).

They found that the amount of compute required for the same performance has been falling exponentially — halving every 16 months.

So since 2012, the amount of compute required for the same level of performance has fallen by over 100 times. Combined with the increased compute used, that’s a lot of growth.¹⁶

It’s hard to say whether these trends will continue, but they speak to incredible gains over the past decade in what it’s possible to do with machine learning.

Indeed, it looks like increasing the size of models (and the amount of compute used to train them) introduces ever more sophisticated behaviour. This is how things like GPT-3 are able to perform tasks they weren’t specifically trained for.

These observations have led to the scaling hypothesis: that we can simply build bigger and bigger neural networks, and as a result we will end up with more and more powerful artificial intelligence, and that this trend of increasing capabilities may increase to human-level AI and beyond.

If this is true, we can attempt to predict how the capabilities of AI technology will increase over time simply by looking at how quickly we are increasing the amount of compute available to train models.

But as we’ll see, it’s not just the scaling hypothesis that suggests we could end up with extremely powerful AI relatively soon — other methods of predicting AI progress come to similar conclusions.

When can we expect transformative AI?

It’s difficult to predict exactly when we will develop AI that we expect to be hugely transformative for society (for better or for worse) — for example, by automating all human work or drastically changing the structure of society.¹⁷ But here we’ll go through a few approaches.

One option is to survey experts. Data from the 2019 survey of 300 AI experts implies that there is 20% probability of human-level machine intelligence (which would plausibly be transformative in this sense) by 2036, 50% probability by 2060, and 85% by 2100.¹⁸ There are a lot of reasons to be suspicious of these estimates,⁴ but we take it as one data point.

Ajeya Cotra (a researcher at Open Philanthropy) attempted to forecast transformative AI by comparing modern deep learning to the human brain. Deep learning involves using a huge amount of compute to train a model, before that model is able to perform some task. There’s also a relationship between the amount of compute used to train a model and the amount used by the model when it’s run. And — if the scaling hypothesis is true — we should expect the performance of a model to predictably improve as the computational power used increases. So Cotra used a variety of approaches (including, for example, estimating how much compute the human brain uses on a variety of tasks) to estimate how much compute might be needed to train a model that, when run, could carry out the hardest tasks humans can do. She then estimated when using that much compute would be affordable.

Cotra’s 2022 update on her report’s conclusions estimates that there is a 35% probability of transformative AI by 2036, 50% by 2040, and 60% by 2050 — noting that these guesses are not stable.¹⁹

Tom Davidson (also a researcher at Open Philanthropy) wrote a report to complement Cotra’s work. He attempted to figure out when we might expect to see transformative AI based only on looking at various types of research that transformative AI might be like (e.g. developing technology that’s the ultimate goal of a STEM field, or proving difficult mathematical conjectures), and how long it’s taken for each of these kinds of research to be completed in the past, given some quantity of research funding and effort.

Davidson’s report estimates that, solely on this information, you’d think that there was an 8% chance of transformative AI by 2036, 13% by 2060, and 20% by 2100. However, Davidson doesn’t consider the actual ways in which AI has progressed since research started in the 1950s, and notes that it seems likely that the amount of effort we put into AI research will increase as AI becomes increasingly relevant to our economy. As a result, Davidson expects these numbers to be underestimates.

Holden Karnofsky, co-CEO of Open Philanthropy, attempted to sum up the findings of all of the approaches above. He guesses there is more than a 10% chance we’ll see transformative AI by 2036(!), 50% by 2060, and 66% by 2100. And these guesses might be conservative, since they didn’t incorporate what we see as faster-than-expected progress since the estimates were made.

All in all, AI seems to be advancing rapidly. More money and talent is going into the field every year, and models are getting bigger and more efficient.

Even if AI were advancing more slowly, we’d be concerned about it — most of the arguments about the risks from AI (that we’ll get to below) do not depend on this rapid progress.

However, the speed of these recent advances increases the urgency of the issue.

(It’s totally possible that these estimates are wrong – below, we discuss how the possibility that we might have a lot of time to work on this problem is one of the best arguments against this problem being pressing).

3. Power-seeking AI could pose an existential threat to humanity

We’ve argued so far that we expect AI to be an important — and potentially transformative — new technology.

We’ve also seen reason to think that such transformative AI systems could be built this century.

Now we’ll turn to the core question: why do we think this matters so much?

There could be a lot of reasons. If advanced AI is as transformative as it seems like it’ll be, there will be many important consequences. But here we are going to explain the issue that seems most concerning to us: AI systems could pose risks by seeking and gaining power.

We’ll argue that:

1. It’s likely that we’ll build AI systems that can make and execute plans to achieve goals
2. Advanced planning systems could easily be ‘misaligned’ — in a way that could lead them to make plans that involve disempowering humanity
3. Disempowerment by AI systems would be an existential catastrophe
4. People might deploy AI systems that are misaligned, despite this risk

Thinking through each step, I think there’s something like a 1% chance of an existential catastrophe resulting from power-seeking AI systems this century. This is my all things considered guess at the risk incorporating considerations of the argument in favour of the risk (which is itself probabilistic), as well as reasons why this argument might be wrong (some of which I discuss below). This puts me on the less worried end of 80,000 Hours staff, whose views on our last staff survey ranged from 1–55%, with a median of 15%.

It’s likely we’ll build advanced planning systems

We’re going to argue that future systems with the following three properties might pose a particularly important threat to humanity:²⁰

1. They have goals and are good at making plans.

   Not all AI systems have goals or make plans to achieve those goals. But some systems (like some chess-playing AI systems) can be thought of in this way. When discussing power-seeking AI, we’re considering planning systems that are relatively advanced, with plans that are in pursuit of some goal(s), and that are capable of carrying out those plans.

2. They have excellent strategic awareness.

   A particularly good planning system would have a good enough understanding of the world to notice obstacles and opportunities that may help or hinder its plans, and respond to these accordingly. Following Carlsmith, we’ll call this strategic awareness, since it allows systems to strategise in a more sophisticated way.

3. They have highly advanced capabilities relative to today’s systems.

   For these systems to actually affect the world, we need them to not just make plans, but also be good at all the specific tasks required to execute those plans.

   Since we’re worried about systems attempting to take power from humanity, we are particularly concerned about AI systems that might be better than humans on one or more tasks that grant people significant power when carried out well in today’s world.

   For example, people who are very good at persuasion and/or manipulation are often able to gain power — so an AI being good at these things might also be able to gain power. Other examples might include hacking into other systems, tasks within scientific and engineering research, as well as business, military, or political strategy.

These systems seem technically possible and we’ll have strong incentives to build them

As we saw above, we’ve already produced systems that are very good at carrying out specific tasks.

We’ve also already produced rudimentary planning systems, like AlphaStar, which skilfully plays the strategy game Starcraft, and MuZero, which plays chess, shogi, and Go.²¹

We’re not sure whether these systems are producing plans in pursuit of goals per se, because we’re not sure exactly what it means to “have goals.” However, since they consistently plan in ways that achieve goals, it seems like they have goals in some sense.

Moreover, some existing systems seem to actually represent goals as part of their neural networks.²²

That said, planning in the real world (instead of games) is much more complex, and to date we’re not aware of any unambiguous examples of goal-directed planning systems, or systems that exhibit high degrees of strategic awareness.

But as we’ve discussed, we expect to see further advances within this century. And we think these advances are likely to produce systems with all three of the above properties.

That’s because we think that there are particularly strong incentives (like profit) to develop these kinds of systems. In short: because being able to plan to achieve a goal, and execute that plan, seems like a particularly powerful and general way of affecting the world.

Getting things done — whether that’s a company selling products, a person buying a house, or a government developing policy — almost always seems to require these skills. One example would be assigning a powerful system a goal and expecting the system to achieve it — rather than having to guide it every step of the way. So planning systems seem likely to be (economically and politically) extremely useful.²³

And if systems are extremely useful, there are likely to be big incentives to build them. For example, an AI that could plan the actions of a company by being given the goal to increase its profits (that is, an AI CEO) would likely provide significant wealth for the people involved — a direct incentive to produce such an AI.

As a result, if we can build systems with these properties (and from what we know, it seems like we will be able to), it seems like we are likely to do so.²⁴

Advanced planning systems could easily be dangerously ‘misaligned’

There are reasons to think that these kinds of advanced planning AI systems will be misaligned. That is, they will aim to do things that we don’t want them to do.²⁵

There are many reasons why systems might not be aiming to do exactly what we want them to do. For one thing, we don’t know how, using modern ML techniques, to give systems the precise goals we want (more here).²⁶

We’re going to focus specifically on some reasons why systems might by default be misaligned in such a way that they develop plans that pose risks to humanity’s ability to influence the world — even when we don’t want that influence to be lost.²⁷

What do we mean by “by default”? Essentially, unless we actively find solutions to some (potentially quite difficult) problems, then it seems like we’ll create dangerously misaligned AI. (There are reasons this might be wrong — which we discuss later.)

Three examples of “misalignment” in a variety of systems

It’s worth noting that misalignment isn’t a purely theoretical possibility (or specific to AI) — we see misaligned goals in humans and institutions all the time, and have also seen examples of misalignment in AI systems.²⁸

Example 1: Winning elections

The democratic political framework is intended to ensure that politicians make decisions that benefit society. But what political systems actually reward is winning elections, so that’s what many politicians end up aiming for.

This is a decent proxy goal — if you have a plan to improve people’s lives, they’re probably more likely to vote for you — but it isn’t perfect. As a result, politicians do things that aren’t clearly the best way of running a country, like raising taxes at the start of their term and cutting them right before elections.

That is to say, the things the system does are at least a little different from what we would, in a perfect world, want it to do: the system is misaligned.

Example 2: The profit incentive

Companies have profit-making incentives. By producing more, and therefore helping people obtain goods and services at cheaper prices, companies make more money.

This is sometimes a decent proxy for making the world better, but profit isn’t actually the same as the good of all of humanity (bold claim, we know). As a result, there are negative externalities: for example, companies will pollute to make money despite this being worse for society overall.

Again, we have a misaligned system, where the things the system does are at least a little different from what we would want it to do.

Example 3: Specification gaming in existing AI systems

DeepMind has documented examples of specification gaming: an AI doing well according to its specified reward function (which encodes our intentions for the system), but not doing what researchers intended.

In one example, a robot arm was asked to grasp a ball. But the reward was specified in terms of whether humans thought the robot had been successful. As a result, the arm learned to hover between the ball and the camera, fooling the humans into thinking that it had grasped the ball.²⁹

So we know it’s possible to create a misaligned AI system.

Why these systems could (by default) be dangerously misaligned

Here’s the core argument of this article. We’ll use all three properties from earlier: planning ability, strategic awareness, and advanced capabilities.

To start, we should realise that a planning system that has a goal will also develop ‘instrumental goals’: things that, if they occur, will make it easier to achieve an overall goal.

We use instrumental goals in plans all the time. For example, a high schooler planning their career might think that getting into university will be helpful for their future job prospects. In this case, “getting into university” would be an instrumental goal.

A sufficiently advanced AI planning system would also include instrumental goals in its overall plans.

If a planning AI system also has enough strategic awareness, it will be able to identify facts about the real world (including potential things that would be obstacles to any plans), and plan in light of them. Crucially, these facts would include that access to resources (e.g. money, compute, influence) and greater capabilities — that is, forms of power — open up new, more effective ways of achieving goals.

This means that, by default, advanced planning AI systems would have some worrying instrumental goals:

• Self-preservation — because a system is more likely to achieve its goals if it is still around to pursue them (in Stuart Russell’s memorable phrase, “You can’t fetch the coffee if you’re dead”).
• Preventing any changes to the AI system’s goals — since changing its goals would lead to outcomes that are different from those it would achieve with its current goals.
• Gaining power — for example, by getting more resources and greater capabilities.

Crucially, one clear way in which the AI can ensure that it will continue to exist (and not be turned off), and that its objectives will never be changed, would be to gain power over the humans who might affect it (we talk here about how AI systems might actually be able to do that).

What’s more, the AI systems we’re considering have advanced capabilities — meaning they can do one or more tasks that grant people significant power when carried out well in today’s world. With such advanced capabilities, these instrumental goals will not be out of reach, and as a result, it seems like the AI system would use its advanced capabilities to get power as part of the plan’s execution. If we don’t want the AI systems we create to take power away from us this would be a particularly dangerous form of misalignment.

In the most extreme scenarios, a planning AI system with sufficiently advanced capabilities could successfully disempower us completely.

As a (very non-rigorous) intuitive check on this argument, let’s try to apply it to humans.

Humans have a variety of goals. For many of these goals, some form of power-seeking is advantageous: though not everyone seeks power, many people do (in the form of wealth or social or political status), because it’s useful for getting what they want. This is not catastrophic (usually!) because, as human beings:

• We generally feel bound by human norms and morality (even people who really want wealth usually aren’t willing to kill to get it).
• We aren’t that much more capable or intelligent than one another. So even in cases where people aren’t held back by morality, they’re not able to take over the world.

(We discuss whether humans are truly power-seeking later.)

A sufficiently advanced AI wouldn’t have those limitations.

It might be hard to find ways to prevent this sort of misalignment

The point of all this isn’t to say that any advanced planning AI system will necessarily attempt to seek power. Instead, it’s to point out that, unless we find a way to design systems that don’t have this flaw, we’ll face significant risk.

It seems more than plausible that we could create an AI system that isn’t misaligned in this way, and thereby prevent any disempowerment. Here are some strategies we might take (plus, unfortunately, some reasons why they might be difficult in practice):³⁰

• Control the objectives of the AI system. We may be able to design systems that simply don’t have objectives to which the above argument applies — and thus don’t incentivise power-seeking behaviour. For example, we could find ways to explicitly instruct AI systems not to harm humans, or find ways to reward AI systems (in training environments) for not engaging in specific kinds of power-seeking behaviour (and also find ways to ensure that this behaviour continues outside the training environment).

  Carlsmith gives two reasons why doing this seems particularly hard.

  First, for modern ML systems, we don’t get to explicitly state a system’s objectives — instead we reward (or punish) a system in a training environment so that it learns on its own. This raises a number of difficulties, one of which is goal misgeneralisation. Researchers have uncovered real examples of systems that appear to have learned to pursue a goal in the training environment, but then fail to generalise that goal when they operate in a new environment. This raises the possibility that we could think we’ve successfully trained an AI system not to seek power — but that the system would seek power anyway when deployed in the real world.³¹

  Second, when we specify a goal to an AI system (or, when we can’t explicitly do that, when we find ways to reward or punish a system during training), we usually do this by giving the system a proxy by which outcomes can be measured (e.g. positive human feedback on a system’s achievement). But often those proxies don’t quite work.³² In general, we might expect that even if a proxy appears to correlate well with successful outcomes, it might not do so when that proxy is optimised for. (The examples above of politicians, companies, and the robot arm failing to grasp a ball are illustrations of this.) We’ll look at a more specific example of how problems with proxies could lead to an existential catastrophe here.

  For more on the specific difficulty of controlling the objectives given to deep neural networks trained using self-supervised learning and reinforcement learning, we recommend OpenAI governance researcher Richard Ngo’s discussion of how realistic training processes lead to the development of misaligned goals.

• Control the inputs into the AI system. AI systems will only develop plans to seek power if they have enough information about the world to realise that seeking power is indeed a way to achieve its goals.

• Control the capabilities of the AI system. AI systems will likely only be able to carry out plans to seek power if they have sufficiently advanced capabilities in skills that grant people significant power in today’s world.

But to make any strategy work, it will need to both:

• Retain the usefulness of the AI systems — and so remain economically competitive with less safe systems. Controlling the inputs and capabilities of AI systems will clearly have costs, so it seems hard to ensure that these controls, even if they’re developed, are actually used. But this is also a problem for controlling a system’s objectives. For example, we may be able to prevent power-seeking behaviour by ensuring that AI systems stop to check in with humans about any decisions they make. But these systems might be significantly slower and less immediately useful to people than systems that don’t stop to carry out these checks. As a result, there might still be incentives to use a faster, more initially effective misaligned system (we’ll look at incentives more in the next section).

• Continue to work as the planning ability and strategic awareness of systems improve over time. Some seemingly simple solutions (for example, trying to give a system a long list of things it isn’t allowed to do, like stealing money or physically harming humans) break down as the planning abilities of the systems increase. This is because, the more capable a system is at developing plans, the more likely it is to identify loopholes or failures in the safety strategy — and as a result, the more likely the system is to develop a plan that involves power-seeking.

Ultimately, by looking at the state of the research on this topic, and speaking to experts in the field, we think that there are currently no known ways of building aligned AI systems that seem likely to fulfil both these criteria.

So: that’s the core argument. There are many variants of this argument. Some have argued that AI systems might gradually shape our future via subtler forms of influence that nonetheless could amount to an existential catastrophe; others argue that the most likely form of disempowerment is in fact just killing everyone. We’re not sure how a catastrophe would be most likely to play out, but have tried to articulate the heart of the argument, as we see it: that AI presents an existential risk.

There are definitely reasons this argument might not be right! We go through some of the reasons that seem strongest to us below. But overall it seems possible that, for at least some kinds of advanced planning AI systems, it will be harder to build systems that don’t seek power in this dangerous way than to build systems that do.

Disempowerment by AI systems would be an existential catastrophe

When we say we’re concerned about existential catastrophes, we’re not just concerned about risks of extinction. This is because the source of our concern is rooted in longtermism: the idea that the lives of all future generations matter, and so it’s extremely important to protect their interests.

This means that any event that could prevent all future generations from living lives full of whatever you think makes life valuable (whether that’s happiness, justice, beauty, or general flourishing) counts as an existential catastrophe.

It seems extremely unlikely that we’d be able to regain power over a system that successfully disempowers humanity. And as a result, the entirety of the future — everything that happens for Earth-originating life, for the rest of time — would be determined by the goals of systems that, although built by us, are not aligned with us. Perhaps those goals will create a long and flourishing future, but we see little reason for confidence.³³

This isn’t to say that we don’t think AI also poses a risk of human extinction. Indeed, we think making humans extinct is one highly plausible way in which an AI system could completely and permanently ensure that we are never able to regain power.

People might deploy misaligned AI systems despite the risk

Surely no one would actually build or use a misaligned AI if they knew it could have such terrible consequences, right?

Unfortunately, there are at least two reasons people might create and then deploy misaligned AI — which we’ll go through one at a time:³⁴

1. People might think it’s aligned when it’s not

Imagine there’s a group of researchers trying to tell, in a test environment, whether a system they’ve built is aligned. We’ve argued that an intelligent planning AI will want to improve its abilities to effect changes in pursuit of its objective, and it’s almost always easier to do that if it’s deployed in the real world, where a much wider range of actions are available. As a result, any misaligned AI that’s sophisticated enough will try to understand what the researchers want it to do and at least pretend to be doing that, deceiving the researchers into thinking it’s aligned. (For example, a reinforcement learning system might be rewarded for certain apparent behaviour during training, regardless of what it’s actually doing.)

Hopefully, we’ll be aware of this sort of behaviour and be able to detect it. But catching a sufficiently advanced AI in deception seems potentially harder than catching a human in a lie, which isn’t always easy. For example, a sufficiently intelligent deceptive AI system may be able to deceive us into thinking we’ve solved the problem of AI deception, even if we haven’t.

If AI systems are good at deception, and have sufficiently advanced capabilities, a reasonable strategy for such a system could be to deceive humans completely until the system has a way to guarantee it can overcome any resistance to its goals.

2. There are incentives to deploy systems sooner rather than later

We might also expect some people with the ability to deploy a misaligned AI to charge ahead despite any warning signs of misalignment that do come up, because of race dynamics — where people developing AI want to do so before anyone else.

For example, if you’re developing an AI to improve military or political strategy, it’s much more useful if none of your rivals have a similarly powerful AI.

These incentives apply even to people attempting to build an AI in the hopes of using it to make the world a better place.

For example, say you’ve spent years and years researching and developing a powerful AI system, and all you want is to use it to make the world a better place. Simplifying things a lot, say there are two possibilities:

1. This powerful AI will be aligned with your beneficent aims, and you’ll transform society in a potentially radically positive way.
2. The AI will be sufficiently misaligned that it’ll take power and permanently end humanity’s control over the future.

Let’s say you think there’s a 90% chance that you’ve succeeded in building an aligned AI. But technology often develops at similar speeds across society, so there’s a good chance that someone else will soon also develop a powerful AI. And you think they’re less cautious, or less altruistic, so you think their AI will only have an 80% chance of being aligned with good goals, and pose a 20% chance of existential catastrophe. And only if you get there first can your more beneficial AI be dominant. As a result, you might decide to go ahead with deploying your AI, accepting the 10% risk.

4. Even if we find a way to avoid power-seeking, there are still risks

So far we’ve described what a large proportion of researchers in the field² think is the major existential risk from potential advances in AI, which depends crucially on an AI seeking power to achieve its goals.

If we can prevent power-seeking behaviour, we will have reduced existential risk substantially.

But even if we succeed, there are still existential risks that AI could pose.

AI could worsen war

We’re concerned that great power conflict could also pose a substantial threat to our world, and advances in AI seem likely to change the nature of war — through lethal autonomous weapons³⁵ or through automated decision making.³⁶

In some cases, great power war could pose an existential threat — for example, if the conflict is nuclear. It’s possible that AI could exacerbate risks of nuclear escalation, although there are also reasons to think AI could decrease this risk.³⁷

Finally, if a single actor produces particularly powerful AI systems, this could be seen as giving them a decisive strategic advantage. For example, the US may produce a planning AI that’s intelligent enough to ensure that Russia or China could never successfully launch another nuclear weapon. This could incentivise a first strike from the actor’s rivals before these AI-developed plans can ever be put into action.

AI could be used to develop dangerous new technology

We expect that AI systems will help increase the rate of scientific progress.³⁸

While there would be clear benefits to this automation — the rapid development of new medicine, for example — some forms of technological development can pose threats, including existential threats, to humanity. This could be through biotechnology³⁹ (see our article on preventing catastrophic pandemics for more) or through some other form of currently unknown but dangerous technology.⁴⁰

AI could empower totalitarian governments

An AI-enabled authoritarian government could completely automate the monitoring and repression of its citizens, as well as significantly influence the information people see, perhaps making it impossible to coordinate action against such a regime.⁴¹

If this became a form of truly stable totalitarianism, this could make people’s lives far worse for extremely long periods of time, making it a particularly scary possible scenario resulting from AI.

Other risks from AI

We’re also concerned about the following issues, though we know less about them:

• Existential threats that result not from the power-seeking behaviour of AI systems, but as a result of the interaction between AI systems. (In order to pose a risk, these systems would still need to be, to some extent, misaligned.)
• Other ways we haven’t thought of in which AI systems could be misused — especially ones that might significantly affect future generations.
• Other moral mistakes made in the design and use of AI systems, particularly if future AI systems are themselves deserving of moral consideration. For example, perhaps we will (inadvertently) create conscious AI systems, which could then suffer in huge numbers. We think this could be extremely important, so we’ve written about it in a separate problem profile.

So, how likely is an AI-related catastrophe?

This is a really difficult question to answer.

There are no past examples we can use to determine the frequency of AI-related catastrophes.

All we have to go off are arguments (like the ones we’ve given above), and less relevant data like the history of technological advances. And we’re definitely not certain that the arguments we’ve presented are completely correct.

Consider the argument we gave earlier about the dangers of power-seeking AI in particular, based off Carlsmith’s report. At the end of his report, Carlsmith gives some rough guesses of the chances that each stage of his argument is correct (conditional on the previous stage being correct):

1. By 2070 it will be possible and financially feasible to build strategically aware systems that can outperform humans on many power-granting tasks, and that can successfully make and carry out plans: Carlsmith guesses there’s a 65% chance of this being true.
2. Given this feasibility, there will be strong incentives to build such systems: 80%.
3. Given both the feasibility and incentives to build such systems, it will be much harder to develop aligned systems that don’t seek power than to develop misaligned systems that do, but which are at least superficially attractive to deploy: 40%.
4. Given all of this, some deployed systems will seek power in a misaligned way that causes over $1 trillion (in 2021 dollars) of damage: 65%.
5. Given all the previous premises, misaligned power-seeking AI systems will end up disempowering basically all of humanity: 40%.
6. Given all the previous premises, this disempowerment will constitute an existential catastrophe: 95%.

Multiplying these numbers together, Carlsmith estimated that there’s a 5% chance that his argument is right and there will be an existential catastrophe from misaligned power-seeking AI by 2070. When we spoke to Carlsmith, he noted that in the year between the writing of his report and the publication of this article, his overall guess at the chance of an existential catastrophe from power-seeking AI by 2070 had increased to >10%.⁴²

The overall probability of existential catastrophe from AI would, in Carlsmith’s view, be higher than this, because there are other routes to possible catastrophe — like those discussed in the previous section — although our guess is that these other routes are probably a lot less likely to lead to existential catastrophe.

For another estimate, in The Precipice, philosopher and advisor to 80,000 Hours Toby Ord estimated a 1-in-6 risk of existential catastrophe by 2120 (from any cause), and that 60% of this risk comes from misaligned AI — giving a total of a 10% risk of existential catastrophe from misaligned AI by 2120.

A 2021 survey of 44 researchers working on reducing existential risks from AI found the median risk estimate was 32.5% — the highest answer given was 98%, and the lowest was 2%.⁴³ There’s obviously a lot of selection bias here: people choose to work on reducing risks from AI because they think this is unusually important, so we should expect estimates from this survey to be substantially higher than estimates from other sources. But there’s clearly significant uncertainty about how big this risk is, and huge variation in answers.

All these numbers are shockingly, disturbingly high. We’re far from certain that all the arguments are correct. But these are generally the highest guesses for the level of existential risk of any of the issues we’ve examined (like engineered pandemics, great power conflict, climate change, or nuclear war).

That said, I think there are reasons why it’s harder to make guesses about the risks from AI than other risks – and possibly reasons to think that the estimates we’ve quoted above are systematically too high.

If I was forced to put a number on it, I’d say something like 1%. This number includes considerations both in favour and against the argument. I’m less worried than other 80,000 Hours staff — our position as an organisation is that the risk is between 3% and 50%.

All this said, the arguments for such high estimates of the existential risk posed by AI are persuasive — making risks from AI a top contender for the most pressing problem facing humanity.

5. We can tackle these risks

We think one of the most important things you can do would be to help reduce the gravest risks that AI poses.

This isn’t just because we think these risks are high — it’s also because we think there are real things we can do to reduce these risks.

We know of two broad approaches:

1. Technical AI safety research
2. AI governance research and implementation

For both of these, there are lots of ways to contribute. We’ll go through them in more detail below, but in this section we want to illustrate the point that there are things we can do to address these risks.

Technical AI safety research

The benefits of transformative AI could be huge, and there are many different actors involved (operating in different countries), which means it will likely be really hard to prevent its development altogether.

(It’s also possible that it wouldn’t even be a good idea if we could — after all, that would mean forgoing the benefits as well as preventing the risks.)

As a result, we think it makes more sense to focus on making sure that this development is safe — meaning that it has a high probability of avoiding all the catastrophic failures listed above.

One way to do this is to try to develop technical solutions to prevent the kind of power-seeking behaviour we discussed earlier — this is generally known as working on technical AI safety, sometimes called just “AI safety” for short.

Read more about technical AI safety research below.

AI governance research and implementation

A second strategy for reducing risks from AI is to shape its development through policy, norms-building, and other governance mechanisms.

Good AI governance can help technical safety work, for example by producing safety agreements between corporations, or helping talented safety researchers from around the world move to where they can be most effective. AI governance could also help with other problems that lead to risks, like race dynamics.

But also, as we’ve discussed, even if we successfully manage to make AI do what we want (i.e. we ‘align’ it), we might still end up choosing something bad for it to do! So we need to worry about the incentives not just of the AI systems, but of the human actors using them.

Read more about AI governance research and implementation below.

6. This work is neglected

We estimate there are around 400 people around the world working directly on reducing the chances of an AI-related existential catastrophe (with a 90% confidence interval ranging between 200 and 1,000). Of these, about three quarters are working on technical AI safety research, with the rest split between strategy (and other governance) research and advocacy.⁴⁴ We think there are around 800 people working in complementary roles, but we’re highly uncertain about this estimate.⁴⁵

In The Precipice, Ord estimated that there was between $10 million and $50 million spent on reducing AI risk in 2020.

That might sound like a lot of money, but we’re spending something like 1,000 times that amount¹¹ on speeding up the development of transformative AI via commercial capabilities research and engineering at large AI labs.

To compare the $50 million spent on AI safety in 2020 to other well-known risks, we’re currently spending several hundreds of billions per year on tackling climate change.

Because this field is so neglected and has such high stakes, we think your impact working on risks from AI could be much higher than working on many other areas — which is why our top two recommended career paths for making a big positive difference in the world are technical AI safety and AI policy research and implementation.

What do we think are the best arguments against this problem being pressing?

As we said above, we’re not totally sure the arguments we’ve presented for AI representing an existential threat are right. Though we do still think that the chance of catastrophe from AI is high enough to warrant many more people pursuing careers to try to prevent such an outcome, we also want to be honest about the arguments against doing so, so you can more easily make your own call on the question.

Here we’ll cover the strongest reasons (in our opinion) to think this problem isn’t particularly pressing. In the next section we’ll cover some common objections that (in our opinion) hold up less well, and explain why.

Arguments against working on AI risk to which we think there are strong responses

We’ve just discussed the major objections to working on AI risk that we think are most persuasive. In this section, we’ll look at objections that we think are less persuasive, and give some reasons why.

What you can do concretely to help

As we mentioned above, we know of two main ways to help reduce existential risks from AI:

1. Technical AI safety research
2. AI strategy/policy research and implementation

The biggest way you could help would be to pursue a career in either one of these areas, or in a supporting area.

The first step is learning a lot more about the technologies, problems, and possible solutions. We’ve collated some lists of our favourite resources here, and our top recommendation is to take a look at the technical alignment curriculum from AGI Safety Fundamentals.

If you decide to pursue a career in this area, we’d generally recommend working at an organisation focused on specifically addressing this problem (though there are other ways to help besides working at existing organisations, as we discuss briefly below).

Technical AI safety

Approaches

There are lots of approaches to technical AI safety, including:

• Scalably learning from human feedback. Examples include iterated amplification, AI safety via debate, building AI assistants that are uncertain about our goals and learn them by interacting with us, and other ways to get AI systems trained with stochastic gradient descent to report truthfully what they know.
• Threat modelling. An example of this work would be demonstrating the possibility of (allowing us to study) dangerous capabilities, like deceptive or manipulative AI systems. You can read an overview in a recent Google DeepMind paper. This work splits into work that evaluates whether a model has dangerous capabilities (like the work of METR in evaluating GPT-4), and work that evaluates whether a model would cause harm in practice (like Anthropic’s research into the behaviour of large language models and this paper on goal misgeneralisation).
• Interpretability research. This work involves studying why AI systems do what they do and trying to put it into human-understandable terms. For example, this paper examined how AlphaZero learns chess, and this paper looked into finding latent knowledge in language models without supervision. This category also includes mechanistic interpretability — for example, Zoom In: An Introduction to Circuits by Olah et al.). For more, see this survey paper, as well as Hubinger’s a transparency and interpretability tech tree, and Nanda’s A Longlist of Theories of Impact for Interpretability for overviews of of how interpretability research could reduce existential risk from AI.
• Other anti-misuse research to reduce the risks of catastrophe caused by misuse of systems. (We’ve written more on this in our problem profile on AI risk. For example, this work includes training AIs so they’re hard to use for dangerous purposes. (Note there’s lots of overlap with the other work on this list).
• Research to increase the robustness of neural networks. This work involves ensuring that the sorts of behaviour neural networks display when exposed to one set of inputs continues when exposed to inputs they haven’t previously been exposed to, in order to prevent AI systems changing to unsafe behaviour. See section 2 of Unsolved Problems in AI safety for more.
• Work to build cooperative AI. Find ways to ensure that even if individual AI systems seem safe, they don’t produce bad outcomes through interacting with other sociotechnical systems. For more, see Open Problems in Cooperative AI by Dafoe et al. or the Cooperative AI Foundation. This seems particularly relevant for the reduction of ‘s-risks.’
• More generally, there are some unified safety plans. For more, see Hubinger’s 11 possible proposals for building safe advanced AI, or Karnofsky’s How might we align transformative AI if it’s developed very soon.⁴⁷

See Neel Nanda’s overview of the AI alignment landscape for more details.

Key organisations

AI labs in industry that have empirical technical safety teams, or are focused entirely on safety:

• Anthropic is an AI safety company working on building interpretable and safe AI systems. They focus on empirical AI safety research. Anthropic cofounders Daniela and Dario Amodei gave an interview about the lab on the Future of Life Institute podcast. On our podcast, we spoke to Chris Olah, who leads Anthropic’s research into interpretability, and Nova DasSarma, who works on systems infrastructure at Anthropic.
• Model Evaluation and Threat Research works on assessing whether cutting-edge AI systems could pose catastrophic risks to civilization, including early-stage, experimental work to develop techniques, and evaluating systems produced by Anthropic and OpenAI.
• The Center for AI Safety is a nonprofit that does technical research and promotion of safety in the wider machine learning community.
• FAR AI is a research nonprofit that incubates and accelerates research agendas that are too resource-intensive for academia but not yet ready for commercialisation by industry, including research in adversarial robustness, interpretability and preference learning.
• Google DeepMind is probably the largest and most well-known research group developing general artificial machine intelligence, and is famous for its work creating AlphaGo, AlphaZero, and AlphaFold. It is not principally focused on safety, but has two teams focused on AI safety, with the Scalable Alignment Team focusing on aligning existing state-of-the-art systems, and the Alignment Team focused on research bets for aligning future systems.
• OpenAI, founded in 2015, is a lab that is trying to build artificial general intelligence that is safe and benefits all of humanity. OpenAI is well known for its language models like GPT-4. Like DeepMind, it is not principally focused on safety, but has a safety team and a governance team. Jan Leike (head of the alignment team) has some blog posts on how he thinks about AI alignment.
• Ought is a machine learning lab building Elicit, an AI research assistant. Their aim is to align open-ended reasoning by learning human reasoning steps, and to direct AI progress towards helping with evaluating evidence and arguments.
• Redwood Research is an AI safety research organisation, whose first big project attempted to make sure language models (like GPT-3) produce output following certain rules with very high probability, in order to address failure modes too rare to show up in standard training.

Theoretical / conceptual AI safety labs:

• The Alignment Research Center (ARC) is attempting to produce alignment strategies that could be adopted in industry today while also being able to scale to future systems. They focus on conceptual work, developing strategies that could work for alignment and which may be promising directions for empirical work, rather than doing empirical AI work themselves. Their first project was releasing a report on Eliciting Latent Knowledge, the problem of getting advanced AI systems to honestly tell you what they believe (or ‘believe’) about the world. On our podcast, we interviewed ARC founder Paul Christiano about his research (before he founded ARC).
• The Center on Long-Term Risk works to address worst-case risks from advanced AI. They focus on conflict between AI systems.
• The Machine Intelligence Research Institute was one of the first groups to become concerned about the risks from machine intelligence in the early 2000s, and its team has published a number of papers on safety issues and how to resolve them.
• Some teams in commercial labs also do some more theoretical and conceptual work on alignment, such as Anthropic’s work on conditioning predictive models and the Causal Incentives Working Group at Google DeepMind.

AI safety in academia (a very non-comprehensive list; while the number of academics explicitly and publicly focused on AI safety is small, it’s possible to do relevant work at a much wider set of places):

• The Algorithmic Alignment Group in the Computer Science and Artificial Intelligence Laboratory at MIT, led by Dylan Hadfield-Menell
• The Center for Human-Compatible AI at UC Berkeley, led by Stuart Russell, focuses on academic research to ensure AI is safe and beneficial to humans. (Our podcast with Stuart Russell examines his approach to provably beneficial AI.)
• Jacob Steinhardt’s research group in the Department of Statistics at UC Berkeley
• The NYU Alignment research Group led by Sam Bowman
• David Krueger’s research group at the Computational and Biological Learning Laboratory at the University of Cambridge
• The Foundations of Cooperative AI Lab at Carnegie Mellon University
• The Future of Humanity Institute at the University of Oxford has an AI safety research group
• The Alignment of Complex Systems research group at Charles University, Prague

If you’re interested in learning more about technical AI safety as an area — e.g. the different techniques, schools of thought, and threat models — our top recommendation is to take a look at the technical alignment curriculum from AGI Safety Fundamentals.

We discuss this path in more detail here:

Career review of technical AI safety research

Alternatively, if you’re looking for something more concrete and step-by-step (with very little in the way of introduction), check out this detailed guide to pursuing a career in AI alignment.

It’s important to note that you don’t have to be an academic or an expert in AI or AI safety to contribute to AI safety research. For example, software engineers are needed at many places conducting technical safety research, and we also highlight more roles below.

AI governance and strategy

Approaches

Quite apart from the technical problems, we face a host of governance issues, which include:

• Coordination problems that are increasing the risks from AI (e.g. there could be incentives to use AI for personal gain in ways that can cause harm, or race dynamics that reduce incentives for careful and safe AI development).
• Risks from accidents or misuse of AI that would be dangerous even if we are able to prevent power-seeking behaviour (as discussed above).
• A lack of clarity on how and when exactly risks from AI (particularly power-seeking AI) might play out.
• A lack of clarity on which intermediate goals we could pursue that, if achieved, would reduce existential risk from AI.

To tackle these, we need a combination of research and policy.⁴⁸

We are in the early stages of figuring out the shape of this problem and the most effective ways to tackle it. So it’s crucial that we do more research. This includes forecasting research into what we should expect to happen, and strategy and policy research into the best ways of acting to reduce the risks.

But also, as AI begins to impact our society more and more, it’ll be crucial that governments and corporations have the best policies in place to shape its development. For example, governments might be able to enforce agreements not to cut corners on safety, further the work of researchers who are less likely to cause harm, or cause the benefits of AI to be distributed more evenly. So there eventually might be a key role to be played in advocacy and lobbying for appropriate AI policy — though we’re not yet at the point of knowing what policies would be useful to implement.

Key organisations

AI strategy and policy organisations:

• AI Impacts attempts to find answers to all sorts of relevant questions about the future of AI, like “How likely is a sudden jump in AI progress at around human-level performance?”
• The AI Security Initiative at UC Berkeley’s Center for Long-Term Cybersecurity.
• The Centre for the Governance of AI (GovAI) aims to build a global research community, dedicated to helping humanity navigate the transition to a world with advanced AI. On our podcast we’ve spoken to Ben Garfinkel, acting director of GovAI, about some weaknesses of classic AI risk arguments, as well as Allan Dafoe, president of GovAI and leader of DeepMind’s Long-Term Strategy and Governance team, about the destabilising effects of AI.
• The Centre for Long-Term Resilience is a UK think tank focused on existential threats, including those from AI.
• The Center for Security and Emerging Technology at Georgetown researches the foundations of AI (talent, data, and computational power). It focuses on how AI can be used in national security. Listen to our podcast with Helen Toner, their Director of Strategy, for more.
• The Centre for the Study of Existential Risk at the University of Cambridge has a group considering the governance of AI.
• DeepMind and OpenAI both have policy teams (listen to our podcast with members of the OpenAI policy team and our podcast with the head of DeepMind’s governance team, Allan Dafoe).
• The Future of Life Insitute advocates for awareness of AI risk within the academic community and gives out grants for work focused on AI safety.
• The Future of Humanity Institute at the University of Oxford has a macrostrategy research group that considers the future of AI and its contribution to existential risk.
• The Leverhulme Centre for the Future of Intelligence is an interdisciplinary research centre at the University of Cambridge focusing on the impacts of AI on humanity.
• Open Philanthropy provides grants to organisations working on altruistic issues. As a result they have research teams looking at the issues they focus on, including a team looking at potential risks from advanced AI. On our podcast, we spoke to Holden Karnofsky, co-CEO of Open Philanthropy, about his views on risks from AI. (Note: Open Philanthropy is 80,000 Hours’ biggest funder.)
• The Institute for AI Policy and Strategy is focused on AI governance and strategy.

If you’re interested in learning more about AI governance, our top recommendation is to take a look at the governance curriculum from AGI safety fundamentals.

We discuss this path in more detail here:

Career review of AI strategy and policy careers

Also note: it could be particularly important for people with the right personal fit to work on AI strategy and governance in China.

Complementary (yet crucial) roles

Even in a research organisation, around half of the staff will be doing other tasks essential for the organisation to perform at its best and have an impact. Having high-performing people in these roles is crucial.

We think the importance of these roles is often underrated because the work is less visible. So we’ve written several career reviews on these areas to help more people enter these careers and succeed, including:

• Operations management to help impactful organisations grow and function as effectively as possible.
• Research management at an AI safety research organisation.
• Being an executive assistant to someone who’s doing really important work on safety and governance.
• Other non-technical roles in leading AI labs.

Other ways to help

AI safety is a big problem and it needs help from people doing a lot of different kinds of work.

One major way to help is to work in a role that directs funding or people towards AI risk, rather than working on the problem directly. We’ve reviewed a few career paths along these lines, including:

• Founding new projects — in this case, starting new initiatives aimed at reducing risks from advanced AI.
• Being a grantmaker to fund promising projects focused on reducing catastrophic AI risk.
• Working in communication roles
• Helping to build communities of people working on this problem. The most relevant community is the AI safety community itself, but it could also be impactful to help build the community of people working on the world’s most pressing problems (including risks from AI).

There are ways all of these could go wrong, so the first step is to become well-informed about the issue.

There are also other technical roles besides safety research that could help contribute, like:

• Working in information security to protect AI (or the results of key experiments) from misuse, theft, or tampering.
• Becoming an expert in AI hardware as a way of steering AI progress in safer directions.

You can read about all these careers — why we think they’re helpful, how to enter them, and how you can predict whether they’re a good fit for you — on our career reviews page.

Find vacancies on our job board

Our job board features opportunities in AI technical safety and governance:

View all opportunities

Read next:  Explore other pressing world problems

Want to learn more about global issues we think are especially pressing? See our list of issues that are large in scale, solvable, and neglected, according to our research.

Continue →

Acknowledgements

Huge thanks to Joel Becker, Tamay Besiroglu, Jungwon Byun, Joseph Carlsmith, Jesse Clifton, Emery Cooper, Ajeya Cotra, Andrew Critch, Anthony DiGiovanni, Noemi Dreksler, Ben Edelman, Lukas Finnveden, Emily Frizell, Ben Garfinkel, Katja Grace, Lewis Hammond, Jacob Hilton, Samuel Hilton, Michelle Hutchinson, Caroline Jeanmaire, Kuhan Jeyapragasan, Arden Koehler, Daniel Kokotajlo, Victoria Krakovna, Alex Lawsen, Howie Lempel, Eli Lifland, Katy Moore, Luke Muehlhauser, Neel Nanda, Linh Chi Nguyen, Luisa Rodriguez, Caspar Oesterheld, Ethan Perez, Charlie Rogers-Smith, Jack Ryan, Rohin Shah, Buck Shlegeris, Marlene Staib, Andreas Stuhlmüller, Luke Stebbing, Nate Thomas, Benjamin Todd, Stefan Torges, Michael Townsend, Chris van Merwijk, Hjalmar Wijk, and Mark Xu for either reviewing this article or their extremely thoughtful and helpful comments and conversations. (This isn’t to say that they would all agree with everything we’ve said here — in fact, we’ve had many spirited disagreements in the comments on this article!)

The post Preventing an AI-related catastrophe appeared first on 80,000 Hours.

In a nutshell:

To reduce the risks posed by the rise of artificial intelligence, we need to figure out how to make sure that powerful AI systems do what we want. Many potential solutions to this problem will require a lot of high-quality data from humans to train machine learning models. Building excellent pipelines so that this data can be collected more easily could be an important way to support technical research into AI alignment, as well as lay the foundation for actually building aligned AIs in the future. If not handled correctly, this work risks making things worse, so this path needs people who can and will change directions if needed.

Why might becoming an expert in data collection for AI alignment be high impact?

We think it’s crucial that we work to positively shape the development of AI, including through technical research on how to ensure that any potentially transformative AI we develop does what we want it to do (known as the alignment problem). If we don’t find ways to align AI with our values and goals — or worse, don’t find ways to prevent AI from actively harming us or otherwise working against our values — the development of AI could pose an existential threat to humanity.

There are lots of different proposals for building aligned AI, and it’s unclear which (if any) of these approaches will work. A sizeable subset of these approaches require humans to give data to machine learning models, including include AI safety via debate, microscope AI, and iterated amplification.

These proposals involve collecting human data on tasks like:

• Evaluating whether a critique of an argument was good
• Breaking a difficult question into easier subquestions
• Examining the outputs of tools that interpret deep neural networks
• Using one model as a tool to make a judgement on how good or bad the outputs of another model are
• Finding ways to make models behave badly (e.g. generating adversarial examples by hand)

Collecting this data — ideally by setting up scalable systems to both contract people to carry out these sorts of tasks as well as collect and communicate the results — could be a valuable way to support alignment researchers who use it in their experiments.

But also, once we have good alignment techniques, we may need AI companies around the world to have the capacity to implement them. That means developing systems and pipelines for the collection of this data now could make it easier to implement alignment solutions that require this data in the future. And if it’s easier, it’s more likely to actually happen.

What does this path involve?

Human data collection mostly involves hiring contractors to answer relevant questions and then creating well-designed systems to collect high-quality data from them.

This includes:

• Figuring out who will be good at actually generating this data (i.e. doing the sorts of tasks that we listed earlier, like evaluating arguments), as well as how to find and hire these people
• Designing training materials, processes, pay levels, and incentivisation structures for contractors
• Ensuring good communication between researchers and contractors, for example by translating researcher needs into clear instructions for contractors (as well as being able to predict and prevent people misinterpreting these instructions)
• Designing user interfaces to make it easy for contractors to complete their tasks as well as for alignment researchers to design and update tasks for contractors to carry out
• Scheduling workloads among contractors, for example making sure that when data needs to be moved in sequence among contractors, the entire data collection can happen reasonably quickly
• Assessing data quality, including developing ways of rapidly detecting problems with your data or using hierarchical schemes of more and less trusted contractors

Being able to do all these things well is a pretty unique and rare skill set (similar to entrepreneurship or operations), so if you’re a good fit for this type of work, it could be the most impactful thing you could do.

Avoiding harm

If you follow this path, it’s particularly important to make sure that you are able to exercise excellent judgement about when not to provide these services.

We think it’s extremely difficult to make accurate calls about when research into AI capabilities could be harmful.

For example, it sounds pretty likely to us that work that helps make current AI systems safe and useful will be fairly different from work that is useful for making transformative AI (when we’re able to build it) safe and useful. You’ll need to be able to make judgements about whether the work you are doing is good for this future task.

We’ve written an article about whether working at a leading AI lab might cause harm, and how to avoid it.

If you think you might be a good fit for this career path, but aren’t sure how to avoid doing harm, our advising team may be able to help you decide what to do.

Example people

How to predict your fit in advance

The best experts at human data collection will have:

• Experience designing surveys and social science experiments
• Ability to analyse the data collected from experiments
• Some familiarity with the field of AI alignment
• Enough knowledge about machine learning to understand what sorts of data are useful to collect and the machine learning research process
• At least some front-end software engineering knowledge
• Some aptitude for entrepreneurship or operations

Data collection is often considered somewhat less glamorous than research, making it especially hard to find good people. So if you have three or more of these skills, you’re likely a better candidate than most!

How to enter

If you already have experience in this area, there are two main ways you might get a job as a human data expert:

• Find jobs at organisations working on alignment, particularly those doing empirical alignment research. For example, OpenAI, DeepMind, Anthropic, Redwood Research, and Ought are all good choices. (As March 2022, Anthropic is hiring for these sorts of roles.) Surge AI is a startup that is also carrying out this sort of work.
• Consider founding an organisation to do this work, as suggested by alignment researcher Beth Barnes. However, in 2022, Matt Putz and Rudolf Laine tried to start an organisation working on this but thought there wasn’t sufficient demand. They wrote about why they didn’t found a human data for alignment organistion. If you are interested in founding an organisation, contact our team.

If you don’t have enough experience to work directly on this now, you can gain experience in a few ways:

• Do academic research, for example in psychology, sociology, economics, or another social science.
• Work in human-computer interaction or software crowdsourcing.
• Work for machine learning companies in labelling teams — and because these roles are less popular, they can be a great way to rapidly gain experience and promotions in machine learning organisations.

The Effective Altruism Long-Term Future Fund and the Survival and Flourishing Fund may provide funding for promising individuals to learn skills relevant to helping future generations — including human data collection. As a way of learning the necessary skills (and directly helping at the same time), you could apply for a grant to build a dataset that you think could be useful for AI alignment. The Machine Intelligence Research Institute has put up a bounty for such a dataset.

Find a job in this path

If you think you might be a good fit for this path and you’re ready to start looking at job opportunities, you may find relevant roles on our job board:

View all opportunities

Want one-on-one advice on pursuing this path?

If you think this path might be a great option for you, but you need help deciding or thinking about what to do next, our team might be able to help.

We can help you compare options, make connections, and possibly even help you find jobs or funding opportunities.

APPLY TO SPEAK WITH OUR TEAM

Learn more about data collection for AI alignment

• Why we’re not founding a human-data-for-alignment org
• Our problem profile on positively shaping the development of AI
• The 80,000 Hours Podcast on Artificial Intelligence (a collection of 10 key AI episodes from our podcast)
• Our career review of AI safety technical research
• AI safety needs social scientists by Geoffrey Irving and Amanda Askell

Read next:  Learn about other high-impact careers

Want to consider more paths? See our list of the highest-impact career paths according to our research.

Continue →

The post Data collection for AI alignment appeared first on 80,000 Hours.