Put another way: comparative advantage is hard. It is slightly easier in the charity world because the charity world is less competitive overall, but it is still hard. EA is relatively new and thus has found some underfunded areas, but we should expect the orgs that do well by current EA heuristics to eventually become saturated with funds because there will be a lot of people using EA heuristics to inform their giving. The truly best giving opportunities are expected to be at the frontier: causes where the skills needed to evaluate them are rare. This could be due to technical background, professional connections, etc.
I would propose that something that at least helps alleviate this is EAs pursuing higher variance causes. Low variance causes get snapped up pretty quickly, but there is remarkably little stomach among large donors for funding 20 causes and having 19 of them fail, even if the one that succeeds “pays for” the entire experiment in terms of impact.
Cause selection should help here too. If you find a cause that others don’t think clearly about, like Open Borders, or identify an area that is neglected because of newness—synthetic biology biosecurity, then you’re more likely to have a comparative advantage.
I’d argue that even within these areas, it would be hard to find low-hanging research fruit. It seems to me that even new technologies get saturated really quickly. In the bioinformatics research I mentioned, I was doing a fairly uncommon type of data analysis to answer a non-obvious question from a 6-year-old technology—and I still got scooped. My experience with the social sciences is more limited, but I once tried to write a paper exploring the contribution of increasing agricultural productivity to Thai GDP, only to find that the exact topic already had a paper.
So one of my major points is: EA organizations such as GiveWell may do too much original research and writing, when advocacy backed by literature reviews might be better.
If you want to find a comparative advantage in research, maybe breakthrough research, translational research (including efforts for commercialisation), literature reviews are more neglected?
If you want to take a hard line against the usefulness of research among moderately high-talent researchers, then you might argue that it’s better to leave academia for business or policy? Would you try to make that case? The thing is that we’re looking for a comparative advantage, we don’t necessarily need an absolute one (e.g. to be in the top X% at Y task).
I’m not sure it’s right to criticise GiveWell on this sort of topic, when GiveWell mostly focus on reviewing, summarising, integrating and applying existing literature more than creating new experiments and models as academics would focus on.
Yeah I’ll concede that GiveWell is fairly good in terms of not reinventing the wheel, though I think that they could be taking even more shortcuts.
It’s not that I think research is useless—I actually take a very positive view of scientific research as a whole, but this means that an individual is constrained to small marginal returns. Maybe we’re just going to have to accept that, instead of saying “scientific research is fundamentally broken and I’m gonna come in and change everything” the way I used to think. I still think scientific research is a good EA option, though I don’t think translational research is systematically underfunded. There are many incentives to have biomedical applications. In fact, some people argue basic research is underfunded compared to applied research. I’m not sure what you mean by “breakthrough research”.
If one does want to make a more radical difference, one needs to identify systems of incentives that do result in brokenness, like government spending as a whole.
Incentives causing brokenness—We’re going well beyond the subject of your post now, but science still seems to be broken in many ways. As you say, people are incentivised badly—to publish, rather than for the social good. People are incentivised to fund projects with sure incremental progress, much of which is not on the biggest problems facing humanity. Fields are fairly insular and inwards-looking. It takes too long for people to recognise paradigm-shifts. Replication is poor. Science infrastructure is neglected (e.g. LaTeX is old and has lots of room for improvement). Things like cognitive genomics are neglected because of political reasons. So there are a lot of different problems here, although they’re not obviously fixable by an individual motivated researcher. I guess we agree that a person in science might be well-served to try to combat these head-on, rather than just performing research that might be replaceable. I don’t agree that one has to zoom out all the way to government-spending to see problems to be fixed. That would seem to be an overcorrection.
The most prestigious publications like Nature and Science love to publish breakthroughs, but this also leads to sloppiness, like the paper on arsenic-based life that was retracted and the paper on induced stem cells that was retracted. When we see really bold publications like that, we always ask, “How long do you think till this is retracted?” On the other hand, there are also incentives to produce research with social benefit. It’s odd to complain about both breakthrough (i.e. basic, high-risk/high-reward) research and translational (i.e. applied, incremental) research being underfunded. You can’t have it both ways here.
There’s a similar contradiction in this: “It takes too long for people to recognise paradigm-shifts. Replication is poor.” The reason science is slow to change theories is because replication is poor. Individual studies have an inherent stochasticity, so one has to consider a body of work as a whole before being willing to shift the paradigm. (Note that Galileo’s and Mendel’s studies didn’t replicate initially.)
What I think are neglected are science that is ambitious, with low probability of success (e.g. Hsu’s cognitive genomics work, Ioannidis’ statistics work), and work bridging new research to humanitarian applications (e.g. using machine learning to classify medical images, or to detect online fraud or risks to security). These are overlapping sets.
In paradigm shifts, I mean adaptation to different ways of doing things. e.g. working to develop BCI and brain implants, or to apply deep learning in machine learning. These things have occurred too slowly. The replication problems are mostly in soft sciences like psychology, and arise from systemic problems with study pre-registration. The causation of these problems is somewhat entangled, but they’re not the exact same problem. Both should be fixed.
My point is: there are a range of important structural changes that effective altruists might want to make in science.
Put another way: comparative advantage is hard. It is slightly easier in the charity world because the charity world is less competitive overall, but it is still hard. EA is relatively new and thus has found some underfunded areas, but we should expect the orgs that do well by current EA heuristics to eventually become saturated with funds because there will be a lot of people using EA heuristics to inform their giving. The truly best giving opportunities are expected to be at the frontier: causes where the skills needed to evaluate them are rare. This could be due to technical background, professional connections, etc.
I would propose that something that at least helps alleviate this is EAs pursuing higher variance causes. Low variance causes get snapped up pretty quickly, but there is remarkably little stomach among large donors for funding 20 causes and having 19 of them fail, even if the one that succeeds “pays for” the entire experiment in terms of impact.
Cause selection should help here too. If you find a cause that others don’t think clearly about, like Open Borders, or identify an area that is neglected because of newness—synthetic biology biosecurity, then you’re more likely to have a comparative advantage.
I’d argue that even within these areas, it would be hard to find low-hanging research fruit. It seems to me that even new technologies get saturated really quickly. In the bioinformatics research I mentioned, I was doing a fairly uncommon type of data analysis to answer a non-obvious question from a 6-year-old technology—and I still got scooped. My experience with the social sciences is more limited, but I once tried to write a paper exploring the contribution of increasing agricultural productivity to Thai GDP, only to find that the exact topic already had a paper.
So one of my major points is: EA organizations such as GiveWell may do too much original research and writing, when advocacy backed by literature reviews might be better.
If you want to find a comparative advantage in research, maybe breakthrough research, translational research (including efforts for commercialisation), literature reviews are more neglected?
If you want to take a hard line against the usefulness of research among moderately high-talent researchers, then you might argue that it’s better to leave academia for business or policy? Would you try to make that case? The thing is that we’re looking for a comparative advantage, we don’t necessarily need an absolute one (e.g. to be in the top X% at Y task).
I’m not sure it’s right to criticise GiveWell on this sort of topic, when GiveWell mostly focus on reviewing, summarising, integrating and applying existing literature more than creating new experiments and models as academics would focus on.
Yeah I’ll concede that GiveWell is fairly good in terms of not reinventing the wheel, though I think that they could be taking even more shortcuts.
It’s not that I think research is useless—I actually take a very positive view of scientific research as a whole, but this means that an individual is constrained to small marginal returns. Maybe we’re just going to have to accept that, instead of saying “scientific research is fundamentally broken and I’m gonna come in and change everything” the way I used to think. I still think scientific research is a good EA option, though I don’t think translational research is systematically underfunded. There are many incentives to have biomedical applications. In fact, some people argue basic research is underfunded compared to applied research. I’m not sure what you mean by “breakthrough research”.
If one does want to make a more radical difference, one needs to identify systems of incentives that do result in brokenness, like government spending as a whole.
Breakthrough research—this kind of thing: http://blog.givewell.org/2015/04/14/breakthrough-fundamental-science/.
Incentives causing brokenness—We’re going well beyond the subject of your post now, but science still seems to be broken in many ways. As you say, people are incentivised badly—to publish, rather than for the social good. People are incentivised to fund projects with sure incremental progress, much of which is not on the biggest problems facing humanity. Fields are fairly insular and inwards-looking. It takes too long for people to recognise paradigm-shifts. Replication is poor. Science infrastructure is neglected (e.g. LaTeX is old and has lots of room for improvement). Things like cognitive genomics are neglected because of political reasons. So there are a lot of different problems here, although they’re not obviously fixable by an individual motivated researcher. I guess we agree that a person in science might be well-served to try to combat these head-on, rather than just performing research that might be replaceable. I don’t agree that one has to zoom out all the way to government-spending to see problems to be fixed. That would seem to be an overcorrection.
The most prestigious publications like Nature and Science love to publish breakthroughs, but this also leads to sloppiness, like the paper on arsenic-based life that was retracted and the paper on induced stem cells that was retracted. When we see really bold publications like that, we always ask, “How long do you think till this is retracted?” On the other hand, there are also incentives to produce research with social benefit. It’s odd to complain about both breakthrough (i.e. basic, high-risk/high-reward) research and translational (i.e. applied, incremental) research being underfunded. You can’t have it both ways here.
There’s a similar contradiction in this: “It takes too long for people to recognise paradigm-shifts. Replication is poor.” The reason science is slow to change theories is because replication is poor. Individual studies have an inherent stochasticity, so one has to consider a body of work as a whole before being willing to shift the paradigm. (Note that Galileo’s and Mendel’s studies didn’t replicate initially.)
What I think are neglected are science that is ambitious, with low probability of success (e.g. Hsu’s cognitive genomics work, Ioannidis’ statistics work), and work bridging new research to humanitarian applications (e.g. using machine learning to classify medical images, or to detect online fraud or risks to security). These are overlapping sets.
In paradigm shifts, I mean adaptation to different ways of doing things. e.g. working to develop BCI and brain implants, or to apply deep learning in machine learning. These things have occurred too slowly. The replication problems are mostly in soft sciences like psychology, and arise from systemic problems with study pre-registration. The causation of these problems is somewhat entangled, but they’re not the exact same problem. Both should be fixed.
My point is: there are a range of important structural changes that effective altruists might want to make in science.