The best career path for you will depend on your values, strengths and situation, so the ideal approach is to generate your own list of promising options, rather than use a generic list.
In the first section, we’ll outline a process you can use to do this, focusing on the factors we think are most important. Later in the article we’ll apply the process (based on our list of problems and current knowledge) to come up with our five key categories and list of priority paths.
This process can be applied no matter your career stage – whether you’re an undergraduate or nearing retirement. If you already have significant experience in a skill, then also take a look at our advice by skill type.
The aim of the process is to give you ideas for options to aim for over the medium term. Elsewhere we cover how to further narrow down your shortlist in terms of your specific situation and personal fit. We wouldn’t usually encourage anyone to take a path where they have lower than average chances of success, or are likely to become unhappy.
1. Decide which global problems are most pressing
Should you focus on reducing climate change, improving education, lowering the chance of nuclear conflict, or another area entirely? The first stage is to come up with a short-list of 1-5 areas that you think are especially effective to work on.
This is perhaps the most important step, because we’ve argued some problem areas are 100 or even 1,000 times more pressing than those that people often focus on. By this we mean that we expect each unit of resources will lead to 100 times as much impact.1 So, if you start out focused on the wrong area, you might forgo over 99% of your impact.
We cover how to compare problem areas and why there are such big differences in effectiveness in the main career guide or in this more technical article.
Personally, we think the most important issues relate to improving the long-term future and reducing catastrophic risks, and include AI safety, biorisk, building the effective altruism community and global priorities research.
If you’d like to see more of our views on this question, see:
Why to focus on future generations
Why reducing extinction risk is the key priority.
How our views about the world’s biggest problem have changed over the last 8 years.
A ranked list of areas
2. Identify the key bottlenecks in each area
Each problem has different needs. For instance, we’d argue that developing a cure to cancer is not mainly held back by a lack of awareness, since everyone is already aware that cancer is a problem. Rather, it’s held back by a lack of progress on biomedical research, and that mainly requires talented biologists and funding.
We define the key bottleneck facing an area as the input it most needs to make progress.
More precisely, the input that’s the biggest bottleneck is the input which would yield the most progress towards solving the problem if one more person started working on it on average.
Again, we think there are large differences in the extent to which different inputs are bottlenecks, and so if you focus on the wrong one, you might give up most of your impact.
Some inputs we often consider include:
Funding – additional financial resources from donations or fundraising.
Insights – new ideas about how to solve the problem.
Awareness & support – how many people know and care about the issue, and how influential they are.
Political capital – the amount of political power that’s available for the issue.
Coordination – the extent to which existing resources effectively work together.
Community building – finding other people who want to work on the issue.
Logistics and operations – the extent to which programmes can be delivered at scale.
Leadership and management – the extent to which concrete plans can be formed and executed on using the resources already available.
One complication in what’s most relevant is the bottleneck in the future period while you’re working, so you need to try to predict how the area will unfold.
It’s useful to try to be as specific as possible about the bottleneck. For instance, within AI safety, we think that a key bottleneck right now is more talented researchers. We think this is a key bottleneck because:
In the long-term, to address major risks from AI, we’ll need a flourishing, credible field of AI safety research that can come up with solutions to the alignment problem.
Right now, the field isn’t held back by funding, since there are already several large funders and institutions who are willing to cover almost any good opportunities in the field (e.g. MILA, OpenAI, DeepMind, The Open Philanthropy Project, BERI / CHAI).
We think the field is also not mainly held back by a lack of general awareness, since recent press coverage about AI risks has reached many people and there’s a significant group of people who are concerned by the issues. (Though, further high-quality outreach could be useful, since much of the existing coverage misportrays the issue.) It’s also not held back by a lack of political capital, since we don’t know what policy changes would help.
Instead, we expect that what would most benefit the field is more talented researchers working on the issue, especially those who are able to publish well received papers on the topic.
More researchers would not only directly feed into progress on key research questions, but this could also start a positive feedback loop. If more great academic papers on the topic could be published, it would demonstrate that it’s a credible, tractable field, and attract even more researchers into the field.
This is borne out when we talk to experts in the area. These experts often estimate the value of an additional promising technical AI safety researcher is equivalent to $1-$10m of extra funding per year — much more than most people could donate.
Here are some general guidelines on where the bottlenecks are likely to be depending on the stage:
Defining the field – Early on, what’s often needed is insight. First, insight is needed to work out that it’s an issue worth working on, and then it’s needed again to define the key issues in the field (a form of “disentanglement research”). Insight is needed a third time to work out what the best solutions to the problem are. This is the stage we’re at in AI strategy and policy.
Community and field building – After it’s clearer what the solutions are, you can start to build a community focused on the issue. Building a community is often more effective than trying to solve the issue directly because it lets you mobilise others, and achieve a “multiplier” on your efforts. This is usually best achieved through targeted advocacy early on — it’s best to avoid broad advocacy until you’ve better worked out the message, because it’s easy to get wrong and hard to unwind. You may also need some work to make concrete progress on the problem to show that progress is possible, helping to mobilise others. This is closer to the stage that AI technical safety is in.
Scaling up the best solutions – Once the low hanging fruit in community building has been taken, then it’s time to solve the problem through whichever means seem most effective. This might mean launching a research programme, advocacy campaign, or the scale up of a promising intervention. At this point the bottleneck will depend on what the best solution is. If the best solution is research, the bottleneck will eventually be insight; if it’s policy change then it will be political capital; if it’s rolling out an intervention it might be logistics, funding or entrepreneurship. Global health seems to have a logistics bottleneck: one major challenge is to scale up existing evidence-backed treatments such as malaria nets which mainly requires money, logistics and management. An area can also get held up at an earlier stage. For instance, a field might ultimately need insight, but if it’s unable to attract researchers due to a lack of funding, then the funding bottleneck has to be overcome first.
Besides looking at the stage, you can also try to determine the key bottleneck by directly estimating which of the listed inputs are most needed at the margin. For this, it’s often useful to investigate which are most relatively neglected.
Before moving to the next step, aim to identify the 1-3 key bottlenecks in each of the top global problems you want to focus on. You can find some of our own assessments within our problem profiles.
3. Identify the career paths that best address these bottlenecks
In this stage, the aim is to generate specific career paths that help to resolve these bottlenecks.
Below, we list some general categories to consider. Try to think of at least one interesting concrete career option within each.
Direct work — find the best organisations addressing the problem, and work there. These are usually non-profits, but could also be for-profit organisations with a social mission. This could mean working in management, operations, outreach and using many other skills.
Entrepreneurship — help found a new organisation addressing a key bottleneck in the area.
Research — try to make progress on whichever research questions are most important in the area. This usually means seeking a PhD and aiming to work in academia, but there are also research options in non-profits, think tanks and companies.
Mass advocacy — make people more aware of the issue and how it can be solved, or encourage them to take action. This often means working in the media or a campaigning non-profit.
Targeted advocacy — as above, but focused on a niche audience. This could be done alongside another job, especially one that gives you access to influential people.
Government and policy — take a job in a relevant area of government / think tanks / party politics, then try to improve policy, or otherwise enable government to better manage the issue.
Earning to give — take a job where you can earn more, then donate to the best organisations in the area.
Don’t forget you can address bottlenecks either directly or indirectly. For instance, if insights are the key bottleneck, you could try to contribute directly by becoming a researcher, but you could also earn to give and fund researchers or do targeted advocacy to recruit more researchers.
Rather than contribute right away, another option is to invest in yourself (or a community) and explore in order to have a greater impact in the future. We will discuss whether or not to do this in a future article (for now, see our old article).
We’d roughly estimate that by focusing on the career paths that effectively address the key bottlenecks in an area, you can increase your impact about 2-5 times (compared to choosing randomly).2 This makes this stage significantly less important than your choice of area, but still important.
4. Focus on the career options with the best personal fit
As we argue in our main career guide, the most productive contributors in a complex job have 10 or even 100 times as much impact as the median. Not all of this difference is predictable, but even if it only partially is, then “personal fit” is still a key consideration. Among options that you might seriously consider, we’d roughly estimate that the best in terms of personal fit are 2-10 times better than the median.2
This means that although your choice of problem area is likely overall a more important factor than personal fit, once you have a list of options that are plausibly high-impact, then personal fit becomes the key consideration.
More technically, we define personal fit as:
Personal fit: how productive you expect to be in the job in the long-term compared to the average of others who typically take that job.
With this definition, then the total expected immediate impact of a role is:
Expected impact = (average impact of role) x (personal fit)
The “impact of role” depends on the other two factors — how pressing the problem and whether the role makes a large contribution to it on average. If we think of these as “effectiveness factors”, then we can write:
Expected impact = (avg pressingness of problem) x (avg effectiveness of method) x (personal fit)
This formula is the first element in our career framework. However, note that to make an all-considered comparison of options, you also need to consider career capital, job satisfaction, coordination with a community, and other personal factors. We’ll cover these factors in later articles.
We can now see that because the factors roughly multiply, balance is key. If an option is terrible on any dimension then you should probably eliminate it.
We can also see that if we can achieve an increase on each factor, the increases will multiply producing a very large total increase. For instance, if you can find a problem that’s 100-times more pressing, a method that’s 5-times more effective and an option that’s a 10-times better personal fit, then the total increase in impact would be 5000-times. However, in practice the factors will conflict. For instance, your best option for personal fit might not be in the best problem area. Still, we often think it’s possible to achieve a 10 to 100 times increase in impact depending on where you start.
5. Consider making quantitative estimates of the value of different career options
The process we’ve covered so far is mainly based on a broad qualitative analysis, but it’s also useful to compare your outputs to some quantitative estimates, and this section will cover one way of doing that.
The first point is that we can think of different career paths as contributing different resources to global problems. We can then try to compare the value of these different contributions in a standard unit.
One unit we’ve used is dollars of donations to the problem.
You can measure the value of different contributions in dollars of donations by considering trade offs like the following:
Which makes a bigger contribution?
An additional researcher willing to work on the area (at a specific level of personal fit).
$X of donations per year.
The value of “X” at which you’re indifferent between the two is an estimate of the dollar value of the contributions of this research.
If you do this for each career option, then you’ll have a quantitative estimate of the value of your contribution to each area.
If you ask the question for a specific person (e.g. consider Jane working as a researcher), then the estimate should consider all of the factors we’ve listed above (key bottleneck, appropriate career path, personal fit).
Note as well that if you phrase the question appropriately, then it should also consider the “replaceability” of different staff, since we’re asking about the value of an additional researcher considering who else the organisation could hire instead.
3. That said, the difficulty of taking account of all of these factors means that people’s estimates will be highly uncertain, so should be used with caution.
We’ve tried to perform this kind of analysis for the problem areas we’ve reviewed in-depth. We usually do this by asking experts in the problem areas to make these kinds of tradeoffs.
4. For instance, we did this for the effective altruism community as a whole in our 2017 talent survey, which found results like the following:
Unfortunately, these estimates will involve a huge amount of uncertainty and disagreement. But the presence of uncertainty doesn’t mean we shouldn’t at least try to make estimates. Rather, we should make the best estimates we can, while also bearing in mind that they could easily change, and combine quantitative analysis with (hopefully) more robust qualitative arguments. The process of making a quantitative estimate also helps to clarify and improve our reasoning.
Once you have the estimates for the value of your contribution to different areas, you need to combine them with your estimates of the relative effectiveness of working on different problem areas (or at different organisations).
Then, the multiple of the two gives you the expected value of the career paths i.e.:
Expected impact = (effectiveness of area) * (dollar value of contribution to area)
For instance, suppose you’re comparing two options:
Earning to give supporting factory farming charities.
Working in a non-profit focused on global health.
You determine the following (these are entirely hypothetical figures):
Each dollar contributed to global health produces 1 “unit” of “good done” or “value”.
Each dollar contributed to opposing factory farming produces 3 units of value.
If you earn to give, you can donate $10,000 per year.
If you work at a global health non-profit, they’d value having you working there as equivalent to their next best hire at an additional $20,000 of donations each year.
Then, the value of the earning to give for anti-factory farming work is:
3 * 10,000 = 30,000 units
The expected value of the global health non-profit option is:
1 * 20,000 = 20,000 units
So, with these figures, earning to give to support factory farming charities comes out ahead.
However, given the huge uncertainties in these kinds of estimates, the factory farming option is only narrowly ahead so this is close to a draw. Usually, we’d look for one option to be several times better before putting much weight on the estimate. When the differences are more narrow, then we’d mainly focus on more qualitative analysis (e.g. where you’ll have the best personal fit), or other factors that we will cover in later articles (e.g. career capital, value of information).
6. Start to narrow down, or go back to the start
Summing up what we’ve covered, the highest-impact career option for you is the one that does best based on a combination of whether it:
Problem: Contributes to a pressing global problem (perhaps 100-fold increase in impact).
Method: Makes a large contribution to a key bottleneck in one of these problems (perhaps 2-5 fold increase).
Fit: Is an excellent personal fit (2-10 fold increase among reasonable options, maybe 100-fold increase among a wide sample of options).
You can analyse these factors both qualitatively and quantitatively.
Having applied the process, if you don’t have any plausible options at this point, then go back to the start and widen your search. In general, you can either focus on the same problem area, but consider less pressing bottlenecks (e.g. earn to give); or you can consider a wider range of problem areas. We don’t recommend aiming for something with below average personal fit, but you could aim for “good” rather than “excellent”.
If you have a reasonable short-list, then you could start to narrow down and make your plan.
In brief, this will involve the following steps, which we cover in more depth in our career guide:
Decide how highly to weigh gaining career capital and flexibility compared to immediate impact (read more).
Assess your long-term options in terms of impact, career capital, personal fit and comparative advantage, to work out which is best (read more about our framework and assessing your options).
Decide whether to commit to entering the option that seems best or doing more to learn about which option is best. It may even be worth spending several years trying out different paths before revisiting your list (read more about how to try out your options)
If you decide to commit, work out the most effective next step to enter your top option.
Whether you explore or commit, make sure you have some nearby back-up options and a plan Z (read more about ABZ plans).
Now we’ll cover what options we think are best if we apply the steps we’ve just covered.