In this post I share my career decision-making process. I hope that it will help others in their career decisions, by serving as a detailed case study. I believe that the general framework I used can be useful to a wide audience. Furthermore, in the last section (which comprises more than half of the post), I detail all of the options that I considered, many of which are not frequently discussed in EA, and I believe they will be particularly relevant for people with technological or scientific background. I encourage all community members to share their career decision-making process as well.
About half a year ago, I left my math PhD to pursue a more impactful career. In the past few months, I was working on generating a long-list of options, learning more about them and making a decision. Ultimately, I decided to start a PhD in computational healthcare, aiming to work on neglected areas in the healthcare system (in developed countries) and employ tools that are not commonly used in healthcare.
I tried (and failed) to keep this post short. If you want more details on anything, don't hesitate to reach out (by a comment, private message, email [email@example.com] or any other method). I have also tried (and hopefully succeeded) to write it so that each section can be read independently, according to the reader's interests.
Table of Contents
- Goals of this Post
- My Background
- Preferences and Constraints
- Narrowing Down the Long-List
- Making a Final Decision
- Final Plan
- General Helpful Resources
- Options Considered
It is a pleasure to thank Edo Arad for numerous conversations that were instrumental in my process, for connecting me to my new PhD advisor, and for carefully reading this post. I would also like to thank Nadav Brandes, Omri Sheffer, Shahar Lahad, Sella Nevo and Gidon Kadosh for their valuable feedback on this post.
This post does not necessarily reflect their views, and all mistakes are mine.
1. Goals of this Post
There are several reasons for me writing this post, I wish to emphasize the main one - I want more people to write posts like this one. They don't have to be as detailed, don't have to follow the same format, and don't have to discuss the same aspects. Many of us are making career decisions at one point or another, and I believe that it will be extremely valuable to the community to have many detailed and diverse case studies. In particular, I hope this and other people's posts will serve the following purposes:
- Learn from each other's methodologies. We can learn how others generate ideas and figure out which of them are best for their purposes. In my experience, the material on 80,000 hours' website is very helpful, but I felt that it lacks concrete examples of making career choices. I hope we can fill this gap together.
- Share vague and concrete career options, our take on them, and references to read more about them.
- Get to know more people in the community who are (at least somewhat) interested in career paths we are interested in.
- Share our struggle and frustration in this journey.
2. My Background
tl;dr - I'm Shay Ben Moshe, 26 years old from Israel. I have a BSc and an MSc in mathematics, and I have worked as a programmer and cyber-security researcher for about 10 years. I have been involved in the EA community for the past 4 years, and I recently decided to leave my math PhD after one year in (out of five), to pursue a more EA-aligned career.
I started programming when I was in high school - I worked as a freelance web developer (doing both front- and back-end). At the age of 18 I joined the Israel Defense Forces (mandatory) for 5 years, where for the most part I served as a cyber-security researcher, and in the last year of my service I was an R&D team leader. At the same time, I finished a BSc in math and started an MSc in math as well. I left the army three and a half years ago and started working on my master's thesis (in homotopy theory), finishing about a year and a half ago. After finishing my military service I also started working part-time in cyber-security, which is financially rewarding. I enjoyed both my studies and my work a lot.
After finishing my master's, I debated with myself whether I wanted to continue in math, or pursue a more impactful career. At this point, I was already very familiar with EA and had several good alternatives to starting a PhD in math (most of them described in this post). I decided to start a PhD, with the same advisor at the same university. The main reason being that I felt that I would regret it if I didn't try it (even though I didn't think that this is an impactful career path). It is worth mentioning that in hindsight I think that, at least in my case, it was a very good decision, because now I feel content with leaving math and shifting to a different career path.
I enjoyed my PhD a lot and even started working on a paper with my advisor, which is a very fun and interesting experience. However, around March 2020 I was having second thoughts, and after thinking about it for a month or so, I decided to leave my PhD, to pursue a more impactful career. Since I wanted to leave on good terms and enjoyed my studies, I decided to keep participating and helping with our group organization until the end of the semester, and finish my part in the paper I was (and still am) working on. I then let my advisor know, who was very supportive (though he was personally very sad about my decision). The semester ended at the end of June 2020, and then I started thinking about my next steps.
It is worth mentioning that I have good backup options (due to my background in mathematics, programming and cyber-security, and connections that I have made during my military service), as well as enough financial runway to feel comfortable taking high-risk career paths.
3. Preferences and Constraints
I believe that others in a similar situation, living somewhere else, or having different personal preferences, might make very different decisions. Therefore, I think that it is important to state my preferences and constraints explicitly, to understand the decisions I have made. Note however that this list was not clear to me at the onset, but rather was understood along the way.
- I am not fixed on a specific cause-area, but rather I am interested in improving the lives of human and non-human animals (now and in the future). Some "standard" EA cause areas I am interested in include global health & development, animal welfare, climate change, biorisks, nuclear safety and others.
- I am mostly interested in technical work, which for me means researching or applying tools from computer science, mathematics, or physics. I am also interested in team leading and similar roles, as long as they stay relatively close to technical work.
- At this point, I am looking for a good balance between career capital and direct impact. I am still at a relatively early stage of my career, however, I already have experience in math, programming and cyber-security, and have a lot of connections from the army. Therefore, I believe that it makes sense to try to have a high direct impact now, while also trying to build more career capital to some extent.
- I don't want to relocate from Israel for a long period, however, I am open to remote work, traveling once in a while, etc.
- I am willing to take risks with my career if the benefits could be high (this is mainly because I have good backup options). For example, I am ok with taking unconventional or less-paying roles. However, I tend to prefer working on projects with relatively high chances of success.
- At this point in my life, I don't want to be an entrepreneur (including for- and non-profit endeavors), although it is very possible that I will be open to that a few years down the road.
tl;dr - I dedicated 2-3 days a week for a few months to come up with many options, read about them, and talk to experts and people I personally know. Additionally, I had a weekly 2 hours conversation with a friend from EA Israel.
When the semester finished (July 2020), I decided to take a few months to explore as many potential directions as I can, and clarify my preferences. I kept working part-time in cyber-security and still worked on some projects from my math PhD, but I had about 2 or 3 days (net) a week for thinking and exploring.
I should mention that I have read a lot of EA-related material before. In particular, I have read most of the general resources at 80,000 hours' website (including the old career guide, key ideas page, and more), have been following the EA forum (skimming most of the posts and reading older posts), etc. Therefore I decided to start looking for options and read some more material along the way, rather than reading a lot of general career decision-making resources.
At this point, I created a google doc with all of my notes. This document was initially nearly empty, and each section grew naturally as I progressed. Additionally, I created a separate google doc for each meeting I had, summarizing the key takeaways.
My notes contain:
- My preferences and constraints as described above (though in less detail).
- A list of cause areas that I might be interested in and able to work on.
- A long-list of options I was considering. Some of these options are more concrete (e.g. specific companies) while others are more general (such as health or earning to give).
- A list of plans of things that I want to do in my process. This includes the people I want to talk to, the next things I want to read more about, tasks such as applying to jobs or events, etc.
- Notes and links regarding specific options.
I find it extremely helpful to have all of my thoughts and links (at least somewhat) organized in one place. I personally can't remember so many things and manage that many tasks at the same time, without having some sort of organization system. It is certainly possible to use other ways to organize this data (be it task management apps or wikis), but since this is not a huge project, I found the simplicity of a single google doc very suitable for my purposes.
Then, I started exploring the most promising options, by reading more about them, talking to people who might be relevant, and making the options more concrete (e.g. by finding relevant specific sub-areas, finding companies and researchers in academia who work in this domain, etc). At the same time, I was generating more options, by reading more resources and talking to more people.
I have talked with over 25 different people in depth about my career (including experts I reached out to, and people I know personally). This was extremely valuable to my process, both by helping me understand specific options more deeply and eliminating some of them, generating more options, understanding better what I am looking for and uncovering more considerations.
Furthermore, I had a weekly 2 hours conversation with my friend Edo Arad, a fellow member of EA Israel. In these conversations we discussed, among many other things, my process. He helped me generate more options, be more confident about my leading options, pushed me into reaching out to more people, and helped in various other ways. His help was invaluable in my process, and I can't thank him enough for that.
Opportunities and Cause Areas
Another important point that I wish to emphasize is that I was looking for promising options or opportunities, rather than promising cause areas. I believe that this methodology is much better suited when looking at the career options of a single person. That is because while some cause area might rank fairly low in general, specific options which might be a great fit for the person in question could be highly impactful (for example, climate change and healthcare [in the developed world] are considered very non-neglected in EA, while I believe that there are promising opportunities in both areas). That said, it surely is natural to look for specific options within a promising cause area.
5. Narrowing Down the Long-List
After reading most of the material I wanted to read and talking to all of the people I wanted to talk to, I decided that it was time to focus and start converging. At this point, most of my options seemed less promising than the leading ones (for reasons which I tried to detail for each option separately), or I wasn't very excited about pursuing them. This left me with the following options (ordered alphabetically):
- Computational healthcare PhD with Eitan Bachmat. This option made it to the next step.
- Earning to give. Although I could potentially have a fairly large impact in this path, it was eliminated, mostly because I don't think that I will enjoy working in cyber-security for many years (say more than 5 years from now). Furthermore, since I already have plenty of background in this domain, I don't think that working for a few more years would help me build career capital, especially compared to the other options.
- Part-time earning to give. I think that this option could be promising if I found good side projects to run. In a way, this is the situation with the computational healthcare PhD as explained there, so it made it to the next step in that sense.
- Flood forecasting at Google (and other AI for good). It was fairly straightforward to choose Google as the leading option among this career path, and it made it to the next step.
- Physics PhD with Ido Kaminer. This option looks very promising, and I believe that I will highly enjoy a PhD in physics. However, I don't think that I want to be a physicist in the long run, and given that I already have a master's in math (and some background in physics), it is not clear to me that investing 5 years in this direction will improve my career capital significantly.
6. Making a Final Decision
My next step was applying to Google, to ensure that this is really an option. I prepared (a lot) for the interviews, and passed them. At this point, I discussed with Sella Nevo, who is leading this team, the specific role I was supposed to take and the long(er) term options in his teams as discussed below.
Now that both options were a possibility, I had to choose between them. Generally speaking, I think that the computational healthcare option is high-risk high-reward, compared to Google. For the reasons detailed below, I ended up choosing the computational healthcare PhD, although both options seem very promising.
In terms of direct impact, I think that the impact at Google is clearer and more certain, and that the project itself is extremely impactful. However, there are many great programmers and researchers who want to work on this project (who are not necessarily part of the EA community). This makes my possible counterfactual impact significantly lower, but probably still fairly high.
As I said, I discussed with Sella the longer term options in his teams. In particular, he assured me that he is interested in founding additional high impact humanitarian and environmental projects at Google. I believe that founding or leading such projects is more counterfactually impactful, as it requires a more unique skill set (e.g. research skills as well project and team management) than joining an existing one.
In contrast, the impact at the computational healthcare option is less clear, and as an outsider, it is hard to estimate how much impact one can have there. In this case however, the key bottleneck of Eitan Bachmat (my advisor) is people with a strong background in computer science or mathematics who want to work on these problems. In particular, he has several ideas that are waiting to be implemented which could have a high impact. Furthermore, I think that there aren't many people in this area who are trying to maximize their impact from an EA perspective.
I tried to quantitatively estimate the direct impact of both options (as well as several other options), so that I can directly compare them. Some options' impact are much harder to model and estimate than others - usually because they are more open-ended and varied. In my case, it was much easier to estimate the impact of the role at Google than that of the computational healthcare PhD. I found out that having a quantitative estimate was more useful to get a sense of the orders of magnitude, rather than having a heads-to-heads comparison.
Apart from the direct impact, a Google salary will allow me to donate a fair amount (although probably less than my top path for earning to give). In the PhD option, since I won't get a scholarship from the university nor a salary from the research institute, I will keep working part-time, and I believe that I will still be able to donate a fair amount. Furthermore, the added flexibility will allow me pursue side projects, which could potentially have high impact as well.
I believe that both options are great in terms of building career capital, but for different reasons.
At Google, I will probably make a lot of connections with people with a strong quantitative background (I should mention that I already have a lot of such connections from my military service, so it is unclear that this is such a high benefit). Furthermore, I will learn machine learning, opening up other ML positions more easily in the future (leading to the AI for good career path).
Pursuing the computational healthcare PhD will earn me a PhD, which is a good credential on its own. Furthermore, it opens a path to continue in academia. Lastly, and probably most importantly, it will significantly broaden my skill set and domain of expertise. This potentially opens a lot of options in the future (e.g. entrepreneurship in healthcare, academic work, working at the public sector, etc).
Google will probably be better socially since I will work closely with nice people of (roughly) my age on a daily basis (at least once the COVID-19 situation ends).
On the other hand, the computational healthcare PhD will probably be much more diverse since I will be able to work on different ideas using multiple tools. Furthermore, I will be working about 2 days a week elsewhere on completely different things. Having done so in the past, I find that I enjoy working on multiple projects at the same time. In addition, both the PhD and the side work are much more flexible than working full-time, which is another advantage.
I believe that I will enjoy both options professionally, but having more diverse work is a big advantage of the computational healthcare PhD.
7. Final Plan
My final decision is to start the computational healthcare PhD. However, as I said, it is a somewhat risky option and it is not clear that it will be as impactful as I hope. Therefore, this is my plan:
- Start the computational healthcare PhD ("plan A").
- After about half a year in (hopefully having done at least one project), re-evaluate this option and try to understand the potential impact better.
- If it turns out this option is not as good as I expected, try to move to Google ("plan B").
If both options don't work for some reason, I will have to go back and look for the leading options available to me at that point.
I also have a fairly straightforward backup plan ("plan Z") in case everything goes wrong, which is going back to working as a cyber-security researcher or as a programmer.
8. General Helpful Resources
In the next section, I added links to resources relevant to specific career options where I describe them. However, some resources do not fit into a specific option (though some of them could fit specific cause areas). Here I wanted to share some of the resources I read, which were relevant broadly and I found helpful.
- 80,000 Hours (Old) Career Guide. I read their guide several years ago, and I found it helpful to read (most of) it again to gain more perspective and ideas. I should mention that they have a newer page of key ideas, which I find less helpful (but this is a topic for a different post).
- Effective Environmentalism Resources contains a lot of resources on different aspects of climate change, as well as links to the community's facebook group and slack channel.
- How to make tough career decisions by 80,000 Hours describes a methodology to career decision-making, which I roughly followed. It is worth mentioning that they have a newer and much more comprehensive article called How to plan your career, which I view as the advanced version of the first one.
- Problem areas beyond 80,000 Hours' current priorities and Some promising career ideas beyond 80,000 Hours' priority paths by Arden Koehler. I found these posts helpful in expanding my long-list of options (naturally, most of the ideas were not relevant for me directly, but some were, or gave me new directions to look at).
- Thoughts on 80,000 Hours' research that might help with job-search frustrations by Arden Koehler. I find this post mostly reassuring (which is very important when making big changes), rather than directly helpful to figuring out my top options.
9. Options Considered
In this section, I will share my takeaways about most of the options I considered. Here are some important remarks:
- The notes here are taken from my perspective. I tried to separate the things which I think are facts, from my subjective thoughts on them (e.g. am I a good fit for this position), but take everything with a grain of salt.
- I decided to breakdown this section by specific options rather than cause areas more broadly. My reasoning for this is described in the paragraph about opportunities and cause areas.
- At this point I am not trying to compare my options, thus I have ordered them alphabetically.
- I will not be able to give all of the details here, mostly because I want to keep this post in a reasonable size. If you are interested in more details, please let me know.
AI for Good
tl;dr - I believe it is possible to have a very large impact by employing AI to solve real-world problems. However, it is fairly easy to fall into the trap of a lower counterfactual impact.
By "AI for good" I mean a very wide set of different things, whose common thread is employing AI and ML techniques to solve real-world problems (as opposed to doing academic work on making general progress in AI and ML). The reasons I am framing it in this way are:
- This skill-set is highly transferable between different cause areas and problems within. I view this career path as a long-term career plan, in which one would probably work at different companies and institutions, on different kinds of problems.
- I believe that the characteristics of this kind of work (such as the salary, what the day-to-day looks like, the kinds of tools you use, etc) are fairly similar across different roles.
Some examples for this kind of work include advancing science (e.g. at the Allen Institute for AI, which has an Israeli branch), application to healthcare (such as computational healthcare PhD or the Israeli startup Aidoc, K Health and many others), working on natural disasters (e.g. flood forecasting at Google or at the Israeli startup SeismicAi) and so on. (Note that I am not familiar with some of these organizations, and in particular can not vouch for their impact.)
I believe that one can have a very large impact on this career path, especially when they gain enough expertise, by working on the right problems. In particular, I believe that it is possible to find important and neglected problems which can be solved more efficiently using ML compared to standard methods. My main concern is that the counterfactual impact might not be as high. This is because there are so many AI and ML researchers, data scientists and entrepreneurs looking for opportunities in this field, and many of them do want to work on important problems. Therefore, I believe that to have a large counterfactual impact, one needs to find a niche or an alternative approach that others do not take, which can happen for several reasons (e.g. they are too risky for a startup, or not as profitable or prestigious as other approaches).
My other concern is that I am not entirely sure that I will personally enjoy this kind of work, though I believe it is very probable that I will.
One particular example which I separated to give more details about, is flood forecasting at Google.
tl;dr - I am not convinced by the arguments for working on AI safety.
Technical AI safety is regarded by many EAs as a top priority. I engaged with some of the ideas and arguments for working on AI safety even before starting this process. Furthermore, it seems like I am personally a very good fit for technical AI safety work, given my mathematics and programming background.
However, I am not convinced by the arguments. As part of my process, I decided to engage a little bit more with the arguments, and still was not convinced. I will not spell out my reasoning in this post (and frankly, I am not sure that I have novel counterarguments). I wanted to participate in MIRI's AIRCS, but unfortunately due to the COVID-19 situation, no event is planned any time soon.
tl;dr - Alternative proteins is a very promising opportunity to have a large impact. However, I didn't find any especially promising option relevant for me.
Alternative proteins refers to (at least) two different approaches - plant-based food and cellular agriculture. These alternatives have many advantages over traditional food production systems - they benefit animal-welfare, have a much smaller impact on climate change, and offer higher food security. In recent years, the alternative proteins industry, market, and academic work have grown significantly (see reports by the Good Food Institute).
Despite that, I believe that alternative proteins are still somewhat neglected and that there are good opportunities to have a high impact in this area.
Luckily, Israel is one of the leading countries in alternative proteins. Unfortunately, though, I couldn't find any company in Israel where my skill set seems very relevant (though that might change in the coming years, as these small companies mature and face new challenges).
At some point, I found out that Vow (an Australian company) was looking for a software engineer, and I applied for a remote position. I had a short interview with their Head of Engineering, during which we both understood that it will be hard to make this position work remotely for several reasons. Alternatively, he suggested that I could do contract work, on some more self-contained research projects. I still consider this as a potentially good option, which I could pursue as a side project along with part-time earning to give.
I also reached out to a contact at the Israeli branch of the Good Food Institute. He suggested that I might be able to work on cultivated meat modeling, and referred me to the CMMC and a whitepaper they wrote. This direction doesn't seem very mature, and didn't get me too excited, so I didn't dig deeper.
tl;dr - Biorisks pose a significant x-risk. However, I couldn't find any opportunities to work on reducing them using my background. Furthermore, I believe that the COVID-19 pandemic might significantly reduce the counterfactual impact of this career path.
Biorisks are discussed extensively in EA, so I will not repeat the standard arguments.
I find it plausible that there are good opportunities to work on reducing biorisks, coming from my background. However, I was looking for such opportunities in Israel (though not extensively) and found none.
Furthermore, I think that the COVID-19 pandemic might lead to many more resources being directed to biorisks reductions. This is of course good news for the world, but might make this career path less attractive if one is trying to maximize their counterfactual impact.
Climate Venture Capitals
tl;dr - Working at a climate VC could potentially be highly impactful, by allocating money more effectively. It doesn't look like the kind of work I would personally enjoy.
Many VCs are investing in climate-related companies and startups. I believe that by working in one of the bigger VCs, one could potentially affect large amounts of money. It is reasonable that allocating money more effectively could have a large impact on climate change mitigation.
I tried to apply to the Head of Science position at Lowercarbon Capital, after seeing this position listed on climate.careers and thinking that I could be a good fit for this if a remote position is possible. I wish to emphasize that I don't think that this VC is necessarily the best option to have impact, and that the main reason I applied was that I happened to see a position that looked like a plausibly good opportunity. I didn't get any response from them. Most of the work in this area doesn't look like something I would enjoy much, and I couldn't find many technical roles.
Computational Healthcare PhD
tl;dr - Computational healthcare is an interdisciplinary effort to apply techniques from computer science (and related fields) to solve problems in health. I found a promising opportunity to have an impact via a PhD in this field, and this is the option I ended up choosing.
At some point, my friend Edo Arad suggested that I'd talk to Eitan Bachmat, a computer scientist at Ben-Gurion University. He works in many areas, one of which is computational healthcare. When we talked, it became clear that, although he is not a part of the effective altruism community, he definitely shares a lot of our perspectives on impact. He works directly with the two largest HMOs (Health Maintenance Organizations) in Israel (Clalit and Maccabi) on real-world problems, and optimizes for making a significant impact rather than publishing papers. In particular, he works on neglected problems in under-funded areas and tries to apply innovative ideas from computer science to solve them. He has been working in this area for several years, and has many good ideas and successful past results, and his main bottleneck is talented students.
Israel is in a unique situation with regards to its healthcare data. First and foremost, the medical records are highly digitized and comprehensive. Furthermore, every resident is required by law to join one of four HMOs, and people tend to not move between them a lot. This means that the HMOs in Israel have gathered medical data about many patients consistently over multiple decades. In this option I will work at KSM Research and Innovation Institute (the research institute of Israel's second-largest HMO, Maccabi), along with Eitan, on several projects.
It is important to mention that neither the university nor the research institute will pay me any salary or scholarship. I will commit to work on my PhD only about 3 days a week, and I will also work part-time somewhere else. Not being paid is a disadvantage on the one hand, but on the other hand, it allows me to pursue other projects or part-time earning to give at the same time.
As an outsider of this field, being far from an expert, it is hard to estimate the possible direct impact in this path. However, after numerous conversations with Eitan and other people (both working in this domain and others), it seems plausible that this is a very high impact path, if one optimizes for it. In that sense, this path is somewhat high-risk high-reward.
See also the comparison to my other leading options.
Earning to Give and Part-time EtG
tl;dr - Earning to give is discussed a lot in EA. I propose a hybrid approach of part-time earning to give while simultaneously running side projects, which I find very compelling.
Earning to give is a career path, fairly popular in EA, where one tries to work in high-earning jobs, to give a substantial percentage of their income. The arguments for and against earning to give were discussed a lot in EA in recent years, so I will not repeat the standard arguments.
I myself have a high-earning potential in cyber-security, and I considered taking this approach. My rough estimates suggested that this is not the most impactful career path among my options, but it is fairly high. Therefore, I used it as a baseline, and compared my other options to it.
Personally, I am mostly concerned that this approach will build almost no career capital for me, as I will keep on working in the same small industry that I have been working in in the past 7 years. Moreover, although I enjoy my part-time position, I don't think that I will enjoy it full-time for several more years.
However, an interesting alternative, which I didn't see discussed in EA, emerged. I can keep working part-time in a high-earning job, while running other projects (such as working with Vow, contract work for EA organization or even starting my own EA-aligned research projects). From my perspective, this approach has several advantages:
- Working part-time might be more profitable than one might intuitively think. This is due to two factors: the first is that, at least in Israel, the income tax is progressive, making the marginal earning smaller; the second is that in some circumstances a freelancer or a consultant can earn even more per hour than a full-time employee.
- Given the previous point, one can still donate a fairly large amount (though admittedly smaller than full-time earning to give).
- Having relatively high financial freedom allows one to take higher-risk side projects.
- One can work on multiple projects at the same time (which I personally enjoy a lot).
- One can build career capital by gaining knowledge in many different fields as well as forming connections in multiple organizations.
The main drawback in my opinion is that in doing so, one is not fully invested in any project. This could potentially make it harder socially and requires some level of self-discipline to actually start and finish side projects.
I find this option very compelling and ended up doing something along these lines.
tl;dr - I think that for- and non-profit entrepreneurship are some of the best ways to have an unusually large impact. I decided that at this point in my life, I don't want this lifestyle..
Entrepreneurship is clearly an extremely large path. First, it is useful to divide it into for- and non-profit entrepreneurship. I was considering both.
I don't have much to say about entrepreneurship, and I don't feel qualified to do so. I do want to mention that I think that it is possible to have a very large impact by founding a for-profit organization, and I think this is not discussed enough in EA. In particular, I couldn't find a good analysis of for-profit entrepreneurship as a way to make an impact from an EA point of view.
Ultimately, after reading about different options, and discussing this with several friends that founded startups in the past few years, I realized that at this stage in my life I don't want to be an entrepreneur. From my understanding, being an entrepreneur is mentally and emotionally hard, and makes it very hard to find a good life-work balance. Furthermore, it seems that most entrepreneurs have a strong desire to start their own projects, which I lack. Despite that, I believe that this is one of the best ways to have an unusually large impact, if it is a good personal fit.
More resources - I don't feel qualified to recommend resources in this area. Some resources that I did find useful for thinking about non-profits are Charity Entrepreneurship's Handbook and charity ideas, and Y Combinator's Non-profit Program as well as What Y Combinator Looks for in Nonprofits and this post in the EA forum.
Flood Forecasting at Google
tl;dr - Joining Google's flood forecasting team looks very promising and was one of my leading options.
(This option is part of the AI for good option, but I decided to write it separately, because it is much more specific while being one of my leading options.)
Sella Nevo, who also founded EA Israel, is leading Google's flood forecasting efforts in developing countries (Talking Machines podcast hosted an episode with Sella on this topic). This project involves creating new real-world models to predict floods more accurately (both in time and place) and as early on as possible, as well as working with developing countries to deploy this system. As is described in the link, the project already has a lot of traction and covers more than 250 million people in India and Bangladesh. Sella chose to lead this project a few years ago, specifically because he believed that if successful, it will have a very large impact (and that the probability that it does succeed, is high). I too believe that working on this project is very impactful.
Furthermore, this option has many of the other advantages that the general AI for good option has, while directly having a large impact. On the technical side, there is a fairly large team of ML researchers at Google Israel, working on different aspects of this problem, and I considered joining this team. I was very excited about this option, and ended up applying and being accepted to it.
Geoengineering and Carbon Removal
tl;dr - This is a neglected field that could potentially have a large impact. I couldn't identify any specific highly promising opportunities available to me.
Geoengineering and carbon removal are a set of interventions that seek to directly affect Earth's climate system. The set of possible interventions is very large, and they range from concrete ideas already implemented in the industry to academic work. Despite that, there are relatively few academics and companies pursuing such solutions, making it fairly neglected.
Currently, it seems to me that the technology that has the highest potential to actually be deployed on a huge scale, while still being neglected, is direct air capture. However, I am skeptical about this too, since the current projections are that they can achieve capture at around $100 per ton CO2 (see for example this report), which is still fairly high compared to other climate change related interventions.
Unfortunately, because of these reasons, I concluded that I couldn't find any highly promising viable options in this area.
I do believe that this direction is worth more consideration, mostly because of the sheer size of the problem. It will be great if someone in the EA community will research it further and write about it.
More resources - We Need To Take CO2 Out Of The Sky is a very good introduction. Air Miners is a community of people working on carbon removal, see in particular their 101 guide. This report by the IEA contains a lot of data about direct air capture. Some organizations working in the field are Global CCS Institute, Carbon Capture Coalition and Circular Carbon Network. Stripe's work in this area is also very insightful. Y Combinator looks to invest in carbon removal technologies. Some of the leading companies working on direct air capture are Carbon Engineering, Climeworks, and Global Thermostat.
tl;dr - Health is a vast subject, generally non-neglected. However, specific areas within it might be very neglected and impactful.
Health is such a vast subject, which makes it hard to describe the leading opportunities. Furthermore, it is very non-neglected as a whole. Nevertheless, I believe that there are specific areas within health, which are far more neglected than others. The standard example in EA is of course global health, which I haven't found a way to directly contribute to (besides donation, that is). I will try to explain a few other areas that I believe are potentially impactful.
One impactful area I considered was working on antimicrobial resistance (AMR), which is the phenomenon where bacteria develop immunity to antibiotics (and more generally, other types of microbes and medicine). This means that over time, we might not have any treatment for conditions caused by such microbes. This was recognized by the World Health Organization as "one of the top 10 global public health threats facing humanity", and they say that "the 60 products in development ... bring little benefit over existing treatments and very few target the most critical resistant bacteria". Furthermore, the latest class of antibiotics reached the market in 1987. This suggests that this problem is at least very large at scale, and fairly neglected. Fortunately, it looks like some new funds and other kinds of organizations are trying to change the situation. As for myself, I was excited to see a paper using ML to find new antibiotics, and I believe that this is a potentially impactful path for me to pursue.
I believe that there are many other big problems in health, similar to AMR, which are worth considering.
Another path I considered was applying AI techniques to solve problems in health in the industry. I know quite a few startups in Israel that are doing just that. Furthermore, I believe that their work can indeed be impactful (though the scale is usually smaller than, say, AMR, they tend to choose more tractable low-hanging fruits in the field). As described at AI for good, I think that my counterfactual impact in these kinds of companies will be lower than one might intuitively think, because so many great people in the industry are attracted to this area, making me much more replaceable. Therefore, I think that to justify this path, you either need to find a very neglected niche, or be an entrepreneur yourself.
The last path I considered, which is the one I ended up choosing, is doing a PhD in computational healthcare, see more details there.
Information Security and Formal Verification
tl;dr - I don't believe there are concrete opportunities to have a high impact on these domains.
Important Edit: Everything I wrote below refers only to technical cyber-security and formal verification roles. I don't have strong views on whether governance, advocacy or other types of work related to those fields could be impactful. My intuition is that these are indeed more promising than technical roles.
I have plenty of background in information security, and I am interested in formal verification.
I was trying to come up with ideas for highly impactful work I could in information security. In particular, I read the forum post Information security careers for GCR reduction and posted in the infosec EA group. Unfortunately, I couldn't find any promising opportunities in this area today. It looks like most people who advocate for this career path think that the EA community will need experts in this topic in the future, and for now we need people to gain expertise in this area. I feel that I am experienced enough in this area broadly, and very experienced in a specific sub-topic, so that gaining more experience in other specific sub-topics is not a good strategy. Furthermore, I believe it will not be too hard to hire experts in infosec for EA goals.
The idea of working on formal verification came up from this post. I am naturally inclined to this kind of work (as it combines several subjects that I enjoy working on), and I do think that it is promising in advancing parts of mathematics and the software industry. However I was not convinced there are, or will be, impactful opportunities in this field from an EA perspective.
tl;dr - Working on meta-science could potentially be a way to make a significant impact on certain fields, and I believe that I can pursue this option in several ways.
Meta-science is a field of science that aims to study and improve scientific research in general.
I became aware of this idea during my conversations with Edo Arad. After discussing this with him, I became convinced that work in this field could make science progress faster and towards better goals (from an EA perspective). One example of potentially impactful work is prioritization in science, which is assisting (basic and applied) science disciplines prioritize their goals and forming a long-term strategy for solving important problems in their field. Another idea is using techniques from AI and ML to make it easier for scientists to explore large bodies of literature, such as SciSight (see also AI for good).
My main concern with this approach is that it is very hard for me to approximate or predict the impact (both its sign and magnitude) of this kind of work, because this is a very indirect way to make an impact (e.g. creating tools that enable scientists to be more productive, which will make more progress possible, which will in turn hopefully have a positive large impact when these methods are employed). Nevertheless, I am fairly certain that accelerating progress in some fields (e.g. cultivated meat) could have a large positive impact.
As for myself, there are several ways for me to work on this topic in academia and other institutions (such as the Israeli branch of the Allen Institute for Artificial Intelligence), as well as by working on such projects independently.
Nuclear Fission Energy
tl;dr - I believe advocacy and policy work for fission energy could be very impactful. I believe technical work on fission energy could also be impactful, but not as much. Further, I found no opportunities to work on this in Israel.
Nuclear fission (not to be confused with fusion) is a physical process already being used to generate energy. Funding for fission R&D has been consistently decreasing (see in particular the section on corporate spending), and is fairly low nowadays.
The debate about fission energy is too long for this short sub-section, so I will not go down this rabbit hole. Personally, I believe fission energy should be used much more, and we should fund much more R&D work on fission energy. Furthermore, Advanced Nuclear Reactors (also called Generation IV reactors) promise many advantages over existing fission reactors.
As I see it, there are two completely different ways to work on nuclear fission:
The first is advocacy and policy work. I believe that this is the bigger constraint, and that this kind of work could have a high impact although I didn't look at it very seriously. However, this kind of work is very far from what I am looking for.
The second is advancing nuclear fission technology. From my understanding, there aren't many open research problems in this area but rather engineering problems. Furthermore, it looks like the biggest problem with fission reactors is their very high capital cost rather than their operation costs. In fact, there are multiple proposals for advanced nuclear reactor designs which solve many of the safety problems and are cheaper and easier to build compared to existing reactors. Since many people have worked on these problems in the past decades and there are multiple private companies working on them today, I believe that this path could be somewhat impactful but not as much as my other alternatives. Moreover I am personally less attracted to working on engineering problems. Lastly, there isn't much work in this domain in Israel.
Nuclear Fusion Energy
tl;dr - Work on fusion energy seems to be extremely impactful. However, there are almost no opportunities to do it in Israel, so I ruled it out.
Nuclear fusion (not to be confused with fission) is a physical process, suggested as a means to generate energy on a large scale. Fusion reactors could potentially be a game-changer for the industry - in short, they promise cheap, reliable and clean energy. The problem is that although fusion processes are not too hard to create in a lab, it is much harder to achieve net-positive, let alone commercially viable, fusion reactors. However, in recent years there was a lot of progress in the area, and the number of private fusion companies grew significantly (data and explanation can be found here and here). Certain startups, such as Commonwealth Fusion Systems, which recently published their status, seem to indicate that fusion energy is possible in the near future.
Using the data here and ITER's budget, I estimate the funding for nuclear fusion R&D in recent years to be on the order of magnitude of $200-1000M per year (including academic and commercial work). This is a very small share of the world's total spending on energy R&D, which is around ~$100B per year (including public and private spending). This indicates that work on fusion is highly neglected.
These facts together suggest that working on fusion energy could have a very high impact, by mitigating climate change and air pollution, as well as making energy cheaper worldwide.
Furthermore, there are several ways to contribute to fusion work, including academic (e.g. PhD in plasma physics) and work in industry (e.g. as a plasma physicist, writing simulations, and many more).
Personally, I found this path very appealing, both from an interest in the topics and enthusiasm about its potential impact. However, there seem to be no companies or academic work on fusion in Israel. Since I don't want to relocate, I had to rule out this option.
More resources - Fusion Energy Base is an extremely useful website that maps existing organizations and their funding. The chapter Nuclear Fusion Power Plants gives a very good and fairly short introduction to the topic. The book The Future Of Fusion Energy is a quite technical book on the current state of fusion energy, which tries to also explain the challenges that we need to overcome to achieve commercially viable fusion. On the academic side, MIT's and Princeton's centers for fusion and the European EP-Fusion are valuable.
tl;dr - A PhD in physics could potentially be very impactful, both by working on important problems at the intersection with other fields, and to build career capital.
A PhD in physics can serve as a means to have a direct impact during the PhD program, or to build career capital. Physics is so vast, and has so many applications, and there are definitely various ways to make impactful work, but I would like to discuss a few specific ideas that I had.
The first is doing a PhD in nuclear fusion or plasma physics, to work directly on nuclear fusion. See there for more information.
The second is specializing in applications in a specific field outside of physics, in which there is a lot of room for impact (examples include medical imaging, microscopy, geoengineering and many others [though I am not sure that these are good examples]). The reasoning for this is that becoming an expert in traditional areas of physics (such as high energy physics, solid-state physics, etc) is extremely hard and competitive, while becoming an expert in applications to another field is much less competitive since there are significantly fewer physicists working in these areas (note that those other areas don't necessarily have to be neglected per se, but lacking physicists).
At some point, my friend Edo Arad suggested that I'd talk to Ido Kaminer, a physics professor in the Technion, after learning that Kaminer has several ideas that sound impactful and is interested in applying his work to important real-world problems. Kaminer works in some areas in physics, mostly in quantum physics and lasers (both from a theoretical and experimental perspective), which I find very interesting (although probably not very impactful from an EA perspective). Furthermore, he also has several new ideas for physical application to fields outside of physics (e.g. improving x-rays, UV applications, and microscopy). I believe that his ideas are indeed innovative and can make a meaningful impact. Furthermore, it is likely that specialists in those other fields won't work on these ideas (mostly because they require a very strong academic background in physics). However, it isn't clear to me that he thinks about impact in the same way I do, and if he really has the time to work on these kinds of projects which are somewhat tangential to his work (although it looks like he recently started working on one of these projects, so I might be wrong about that).
Research at EA Organizations
tl;dr - There are many EA organizations that do very impactful work. It looks like I don't have a good personal fit for these organizations.
By research at EA organizations I mean being a researcher at organizations such as GiveWell, Rethink Priorities, Global Priorities Institute, Future of Humanity Institute, Charity Entrepreneurship, etc. This is of course extremely broad and entails different kinds of research areas. Let me say clearly that I do believe many of these organizations are doing great work, and that it is possible to have a very large impact working for them.
For me personally, most of these organizations are not relevant at the onset, either because I lack the qualifications, or because they would require me to relocate from Israel. In addition, I was uncertain that I would enjoy this kind of research (which is very different from the research I am used to from math and cyber-security).
I did apply to a remote research position at Rethink Priorities, from which I got rejected. I learned a lot from the application process and tasks. In particular, I became more convinced that I personally wouldn't enjoy this kind of research.