Competition is a sign of neglect in important causes with long time horizons for impact.
Primary Audience:
Those considering an EA-inspired career change, or people working in competitive, high-impact cause areas.
Summary:
Teaching or advocacy to increase teaching capacity in a highly competitive, high-impact cause area affords a chance to make more than a mere marginal difference, particularly if the field is immature.
General Argument:
In investigating AI safety as a potential career, I learned that the field is bottlenecked by limited mentorship capacity, as is the field of machine learning in general. From an importance, neglectedness, tractability standpoint, this makes the field seem oversubscribed, less of a good choice for an EA hoping to making a marginal impact through direct work. But I want to argue the opposite conclusion.
Instead, if we take for granted that the field is high-impact, an intensely competitive environment affords a successful applicant extra opportunities to do good over the long term—either by doing direct work, or by teaching (or advocating for growth in teaching capacity). This holds as long as additional capacity for direct work will remain very useful roughly 20 years into the future, when your own students would be coming into their own. If you keep your options open and preserve flexibility as much as possible throughout your career, you can respond to the relative needs for direct work or teaching capacity when you come into your own in the field.
This insight might make a difference in some people’s thinking, because on its face, a field that’s very important but very competitive seems like it’s not neglected at all. Certainly it does not seem neglected by potential applicants, who I envision as the primary intended audience of this post. But the existence of the competition is proof that the field is neglected: it’s the bottleneck itself that’s producing the neglect! Neglect, after all, doesn’t only mean lack of interest. It means lack of supply relative to demand.
Of course, the existence of competition is not evidence that a certain field or cause is high-impact. People compete for respect, money, security, and all sorts of understandable selfish interests. But if you’re comparing two high-impact causes that both seem like great personal fits, choosing the one that is more competitive affords you the chance to both do important direct work, or to teach and increase capacity. The less competitive one still affords high-impact direct work, but less opportunity to improve the field by expanding capacity. Even if you let in more students, they may tend to be of lower quality, simply because there are fewer applicants to choose from.
Evaluating A Counterargument Based On The 80⁄20 Rule:
It’s often said that the top 20% of workers do 80% of the work. Increasing capacity in a field, in theory, means that you’re letting in more people who are probably not going to end up in the top 20%, at least if admissions is doing their job. But I think this doesn’t weaken my argument for choosing the more competitive field.
After all, if you’re in the top 20% of workers in your field, you can simply do direct work. Good thing you picked this field to work in, since you’re so great at it!
By contrast, if you’re not in the top 20%, you can now choose between teaching or doing direct work. Your direct work may not be as potent as the top 20%, but it’s probably still helpful, especially since it’s in such an important cause area. And if you choose to teach, increasing capacity for more students, the bottom-80% people you let in will likewise make a smaller but meaningful difference when you add it all up.
Furthermore, if admissions isn’t doing their job perfectly (and there’s every reason to think that they aren’t, since hiring is known to be a hard problem), then when you increase capacity, there’s a chance one of your students will prove to be a “diamond in the rough” who counterfactually would never have had a chance to shine.
Isn’t Lack of Competition Also A Sign Of Neglect?
If there aren’t enough people applying to work in a high-impact cause area, that also suggests that it’s neglected. But that’s something that’s easy for us to perceive, because it matches up with our preconceived notion of “neglect.” That’s one reason this scenario is not the focus of this post.
The other reason is that if a cause is known but not receiving enough applicants to work on it, you’d need to find a way to increase demand, rather than provide supply. I believe the latter is more tractable than the former.
Finally, it may seem like lack of competition gives you a better chance of being in the top 20% of your profession, and thus being more impactful. However, this is wrong thinking. It’s not being in the top 20% of the profession that causes you to be extraordinarily impactful, but the reverse. Being in a field that doesn’t attract the same volume of applicants doesn’t seem likely to make you any more effective of a worker than you’d otherwise have been.
Application AI Safety Specifically:
One argument against pursuing AI safety research is that we may not be able to do productive work on the problem, given that we’re far from sure about how AI technology will take shape as it matures. This fact adds to the value of becoming an AI safety teacher. Turning students onto the problem and giving them the tools to start will put them in a better position to do this research than you can ever hope to be in, simply because their prime working years will be more likely to intersect with the appearance of more productive methods for working on the problem.
It’s certainly true that fields bottlenecked on mentors could make use of more mentors, right now. If you’re already skilled in the area, you can therefore have very high impact by joining/staying in the field.
However, when young people are considering whether they should join in order to become mentors, as you suggest, they should consider whether the field will be bottlenecked on mentors at the time when they would become one, in 10 years time or so. Since there are lots of junior applicants right now, the seniority bottleneck will presumably be smaller, then.
Moreover, insofar as the present lack of mentors is the main bottleneck preventing junior applicants from eventually becoming senior, adding an extra person to the pool of applicants (yourself) will create fewer counterfactual future mentors than if you were in a field that was less mentorship-constrained. (This doesn’t mean it isn’t worth doing, though. You adding yourself to the pool will still increase its value.)
It also implies that it can be extra valuable to move into the field if you’re able to learn relevant skills without making use of present mentors (e.g. by being in a good and relevant PhD-program, or by doing focused studying that few others are doing).
in the absence of other empirical information, I think it’s a safe assumption that present bottlenecks correlate with future bottlenecks, though your first point is well taken.
I’m not quite following your second argument. It seems to say that the same level of applicant pool growth produces fewer mentors in mentorship-bottlenecked fields than in less mentorship-bottlenecked fields, but I don’t understand why. Enlighten me?
Your third point is also correct. Stated generally, finding ways to increase the availability of the primary bottlenecked resource, or accomplish the same goal while using less of it, is how we can get the most leverage.
If a field is bottlenecked on mentors, it has too few mentors per applicants, or put differently, more applicants than the mentors can accept. Assuming that each applicant needs some fixed amount of time with a mentor before becoming senior themselves, increasing the size of the applicant-pool doesn’t increase the number of future senior people, because the present mentors won’t be able to accept more people just because the applicant-pool is bigger.
Caveats:
More people in the applicant-pool may lead to future senior people being better (because the best people in a larger pool are probably better).
It’s not actually true that a fixed amount of mentor-input makes someone senior. With a larger applicantpool, you might be able to select for people who requires less mentor-input, or who has a larger probability of staying in the field, which will translate to more future senior people (but still significantly less than in applicant-bottlenecked fields).
My third point above: some people might be able to circumvent applying to the mentor-constrained positions altogether, and still become senior.
In my OP, I just meant that if the applicant gets in, they can teach. Too many applicants doesn’t necessarily indicate that the field is oversubscribed, it just means that there’s a mentorship bottleneck. One possible reason is that senior people in the field simply enjoy direct work more than teaching and choose not to focus on it. Insofar as that’s the case, candidates are especially suitable if they’re willing to focus more on providing mentorship if they get in and a bottleneck remains by the time they become senior.
Thanks for the feedback, it helps me understand that my original post may not have been as clear as I thought.