(Just my personal, current, non-expert thoughts, as always. Also, Iâm not sure Iâm addressing precisely the question you had in mind.)
A summary of my recommendations in this vicinity:
If people want to do research and want a menu of ideas/âquestions to work on, including ideas/âquestions that seem like they obviously should have a bunch of work on them but donât yet, they could check out this central directory for open research questions, and/âor an overlapping 80,000 Hours post.
If people want to discover ânewâ instances of such ideas/âquestions, one option might be to just try to notice ideas/âvariables/âassumptions that seem important to some peopleâs beliefs, but that seem debatable and vague, have been contested by others, and/âor havenât been stated explicitly and fleshed out.
One way to do this might be to have a go at rigorously, precisely writing out the arguments that people seem to be acting as if they believe, in order to spot the assumptions that seem required but that those people havenât stated/âemphasised.
One could then try to explore those assumptions in detail, either just through more fleshed-out âarmchair reasoningâ, or through looking at relevant empirical evidence and academic work, or through some mixture of those things.
I think this is a big part of what Iâve done this year.
Hereâs one example of a piece of my own work which came from roughly that sort of process.
Iâll add more detailed thoughts below.
---
I interpret this question as being focused on cases in which an idea/âopen question seems like it shouldâve been obvious, or seems obvious in retrospect, yet it has been neglected so far. (Or the many cases we should assume still exist in which the idea/âquestion is still neglected, but wouldâif and when finally tackledâseem obvious.)
It seems to me that there are two major types of such cases:
Unnoticed: Cases in which the ideas/âopen questions havenât even been noticed by almost anyone
Or at least, almost anyone in the relevant community/âfield.
So Iâd still say an idea counts as âunnoticedâ for these purposes even if, for example, a very similar ideas has been explored thoroughly in sociology, but no one in longtermism has noticed that that idea is relevant to some longtermist issue, nor independently arrived at a similar idea.
Noticed yet neglected: Cases in which the ideas/âopen questions have been noticed, but no one has really fleshed them out or tackled them much
E.g., a fair number of longtermists have noticed that the question of how likely various types of recovery are from various types of civilizational collapse. But as far as Iâm aware, there was nothing even approaching a thorough analysis of the question until some recent still-in-progress work, and thereâs still room for much more work here.
Another example is questions related to how likely global, stable totalitarianism is; what factors could increase or decrease the odds of that; and what to do about this. Some people have highlighted such questions (including but not only in the context of advanced AI), but Iâm not aware of any detailed work on them.
This is really more a continuum than a binary distinction. In almost all cases, thereâs probably been someone in a relevant community whoâs at least briefly noticed something relevant. But sometimes itâll just be that something kind-of relevant has been discussed verbally a few times and then forgotten, while other times itâll be that people have prominently highlighted pretty precisely the relevant open question, yet no one has actually worked on it. (And of course thereâll be many cases in between.)
---
For ânoticed yet neglectedâ ideas/âquestions, recommendation 1 from above will be more relevant: people could find many ideas/âquestions of this type in this central directory for open research questions, and just get cracking on them.
That directory is like a map pointing the way to many trees that might be full of low-hanging fruit that wouldâve been plucked by now in a better world. And I really would predict that a lot of EAs could do valuable work by just having a go at those questions. (Iâm less confident that this is the most valuable thing lots of EAs could be doing, and each person would have to think that through for themselves, in light of their specific circumstances. See also.)
So we donât necessarily need all EA-aligned researchers to try to cultivate a skill of ânoticing the ideas that shouldâve been tackled/âfleshed out alreadyâ (though Iâm sure some should). Some could just focus on actually exploring the ideas that have been noticed but still havenât been tackled/âfleshed out.
---
For âunnoticedâ ideas/âquestions, recommendation 2 from above will be more relevant.
I think this dovetails somewhat with Ben Garfinkel calling for[1] more people to just try to rigorously write up more detailed versions of arguments about AI risk that often float around in sketchier or briefer form. (Obviously brevity is better than length, all else held equal, but often a few pages isnât enough to give an idea proper treatment.)
---
There are at least two other approaches for finding âunnoticedâ ideas/âquestions which seem to have sometimes worked for me, but which Iâm less sure would often be useful for many people, and less sure Iâll describe clearly. These are:
Trying to sketch out causal diagrams of the pathway to something (e.g., an existential catastrophe) happening
I think that doing something like this has sometimes helped me notice there there are:
assumptions or steps missing in the standard/âfleshed-out stories of how something might happen,
alternative pathways by which something could happen, and/âor
Trying to define things precisely, and/âor to precisely distinguish concepts from each other, and seeing if anything interesting falls out
Hereâs an abstract example, but one which matches various real examples that have happened for me:
I try to define X, but then notice that that definition would fail tocover some cases of what Iâd usually think of as X, and/âor that it would cover some cases of what Iâd usually think of as Y (which is a distinct concept).
This makes me realise that X and/âor Y might be able to take somewhat different forms or occur via different pathways to what was typically considered, or that thereâs actually an extra requirement for X or Y to happen that was typically ignored.
I feel like itâd be easy to misinterpret my stance here.
I actually think that definitions will never or almost never really be âperfectâ, and I agree with the ideas in this post (see also family resemblance). And I think that many debates over definitions are largely nitpicking and wasting time.
But I also think that, in many case, being clearer about definitions can substantially benefit both thought and communication.
---
I should again mention that Iâm only ~1.5 years into my research career, so maybe Iâll later change my mind about a bunch of those points, and there are probably a lot of useful things that could be said on this that I havenât said.
[1] See the parts of the transcript after Howie asks âDo you know what it would mean for the arguments to be more sussed out?â
10. âObvious questionsâ
(Just my personal, current, non-expert thoughts, as always. Also, Iâm not sure Iâm addressing precisely the question you had in mind.)
A summary of my recommendations in this vicinity:
If people want to do research and want a menu of ideas/âquestions to work on, including ideas/âquestions that seem like they obviously should have a bunch of work on them but donât yet, they could check out this central directory for open research questions, and/âor an overlapping 80,000 Hours post.
If people want to discover ânewâ instances of such ideas/âquestions, one option might be to just try to notice ideas/âvariables/âassumptions that seem important to some peopleâs beliefs, but that seem debatable and vague, have been contested by others, and/âor havenât been stated explicitly and fleshed out.
One way to do this might be to have a go at rigorously, precisely writing out the arguments that people seem to be acting as if they believe, in order to spot the assumptions that seem required but that those people havenât stated/âemphasised.
One could then try to explore those assumptions in detail, either just through more fleshed-out âarmchair reasoningâ, or through looking at relevant empirical evidence and academic work, or through some mixture of those things.
I think this is a big part of what Iâve done this year.
Hereâs one example of a piece of my own work which came from roughly that sort of process.
Iâll add more detailed thoughts below.
---
I interpret this question as being focused on cases in which an idea/âopen question seems like it shouldâve been obvious, or seems obvious in retrospect, yet it has been neglected so far. (Or the many cases we should assume still exist in which the idea/âquestion is still neglected, but wouldâif and when finally tackledâseem obvious.)
It seems to me that there are two major types of such cases:
Unnoticed: Cases in which the ideas/âopen questions havenât even been noticed by almost anyone
Or at least, almost anyone in the relevant community/âfield.
So Iâd still say an idea counts as âunnoticedâ for these purposes even if, for example, a very similar ideas has been explored thoroughly in sociology, but no one in longtermism has noticed that that idea is relevant to some longtermist issue, nor independently arrived at a similar idea.
Noticed yet neglected: Cases in which the ideas/âopen questions have been noticed, but no one has really fleshed them out or tackled them much
E.g., a fair number of longtermists have noticed that the question of how likely various types of recovery are from various types of civilizational collapse. But as far as Iâm aware, there was nothing even approaching a thorough analysis of the question until some recent still-in-progress work, and thereâs still room for much more work here.
More thoughts and notes on this here and here.
Another example is questions related to how likely global, stable totalitarianism is; what factors could increase or decrease the odds of that; and what to do about this. Some people have highlighted such questions (including but not only in the context of advanced AI), but Iâm not aware of any detailed work on them.
This is really more a continuum than a binary distinction. In almost all cases, thereâs probably been someone in a relevant community whoâs at least briefly noticed something relevant. But sometimes itâll just be that something kind-of relevant has been discussed verbally a few times and then forgotten, while other times itâll be that people have prominently highlighted pretty precisely the relevant open question, yet no one has actually worked on it. (And of course thereâll be many cases in between.)
---
For ânoticed yet neglectedâ ideas/âquestions, recommendation 1 from above will be more relevant: people could find many ideas/âquestions of this type in this central directory for open research questions, and just get cracking on them.
That directory is like a map pointing the way to many trees that might be full of low-hanging fruit that wouldâve been plucked by now in a better world. And I really would predict that a lot of EAs could do valuable work by just having a go at those questions. (Iâm less confident that this is the most valuable thing lots of EAs could be doing, and each person would have to think that through for themselves, in light of their specific circumstances. See also.)
So we donât necessarily need all EA-aligned researchers to try to cultivate a skill of ânoticing the ideas that shouldâve been tackled/âfleshed out alreadyâ (though Iâm sure some should). Some could just focus on actually exploring the ideas that have been noticed but still havenât been tackled/âfleshed out.
---
For âunnoticedâ ideas/âquestions, recommendation 2 from above will be more relevant.
I think this dovetails somewhat with Ben Garfinkel calling for[1] more people to just try to rigorously write up more detailed versions of arguments about AI risk that often float around in sketchier or briefer form. (Obviously brevity is better than length, all else held equal, but often a few pages isnât enough to give an idea proper treatment.)
---
There are at least two other approaches for finding âunnoticedâ ideas/âquestions which seem to have sometimes worked for me, but which Iâm less sure would often be useful for many people, and less sure Iâll describe clearly. These are:
Trying to sketch out causal diagrams of the pathway to something (e.g., an existential catastrophe) happening
I think that doing something like this has sometimes helped me notice there there are:
assumptions or steps missing in the standard/âfleshed-out stories of how something might happen,
alternative pathways by which something could happen, and/âor
alternative/âadditional outcomes that may occur
See also
Trying to define things precisely, and/âor to precisely distinguish concepts from each other, and seeing if anything interesting falls out
Hereâs an abstract example, but one which matches various real examples that have happened for me:
I try to define X, but then notice that that definition would fail to cover some cases of what Iâd usually think of as X, and/âor that it would cover some cases of what Iâd usually think of as Y (which is a distinct concept).
This makes me realise that X and/âor Y might be able to take somewhat different forms or occur via different pathways to what was typically considered, or that thereâs actually an extra requirement for X or Y to happen that was typically ignored.
I feel like itâd be easy to misinterpret my stance here.
I actually think that definitions will never or almost never really be âperfectâ, and I agree with the ideas in this post (see also family resemblance). And I think that many debates over definitions are largely nitpicking and wasting time.
But I also think that, in many case, being clearer about definitions can substantially benefit both thought and communication.
---
I should again mention that Iâm only ~1.5 years into my research career, so maybe Iâll later change my mind about a bunch of those points, and there are probably a lot of useful things that could be said on this that I havenât said.
[1] See the parts of the transcript after Howie asks âDo you know what it would mean for the arguments to be more sussed out?â