Benjamin is a research analyst at 80,000 Hours. Before joining 80,000 Hours, he worked for the UK Government and did some economics and physics research.
To clarify what I mean by unknown unknowns, here's a climate-related example: We're uncertain about the strength of various feedback loops, like how much warming could be produced by cloud feedbacks. We'd then classify "cloud feedbacks" as a known unknown. But we're also uncertain about whether there are feedback loops we haven't identified. Since we don't know what these might be, these loops are unknown unknowns. As you say, the known feedback loops don't seem likely to warm earth enough to cause a complete destruction of civilisation, which means that if climate change were to lead to civilisational collapse, that would probably be because of something we failed to consider.
But here's the thing: generally we do know something about unknown unknowns.[1] In the case of these unknown feedback loops, we can place some constraints on them. For example:
They couldn't cool the Earth past absolute zero, because that's pretty much impossible.[2]
They almost certainly couldn't make the earth hotter than the Sun (because at some point the Earth would start forming a fusing ball of plasma, and the Earth isn't heavy enough to be hotter than the sun if it turned into a star).
In fact, we can gather a broad variety of evidence about these unknown unknowns, using various different lines of evidence. These lines of evidence include:
The physics constraining possible feedback processes
The historical climate record (since 1800)
The paleoclimate record (millions of years into the past)
Accounting for these multiple lines of evidence is exactly what the 6th Assessment Report attempts to do when calculating climate sensitivity (how much Earth's surface will cool or warm after a specified factor causes a change in its climate system):[3]
In AR6 [the 6th Assessment report], the assessments of ECS [equilibrium climate sensitivity] and TCR [transient climate response] are made based on multiple lines of evidence, with ESMs [earth system models] representing only one of several sources of information. The constraints on these climate metrics are based on radiative forcing and climate feedbacks assessed from process understanding (Section 7.5.1), climate change and variability seen within the instrumental record (Section 7.5.2), paleoclimate evidence (Section 7.5.3), emergent constraints (Section 7.5.4), and a synthesis of all lines of evidence (Section 7.5.5). In AR5 [the 5th assessment report], these lines of evidence were not explicitly combined in the assessment of climate sensitivity, but as demonstrated by Sherwood et al. (2020) their combination narrows the uncertainty ranges of ECS compared to that assessed in AR5.
That is, as I mentioned in the main post "the IPCC's Sixth Assessment Report... attempts to account for structural uncertainty and unknown unknowns. Roughly, they find it’s unlikely that all the various lines of evidence are biased in just one direction — for every consideration that could increase warming, there are also considerations that could decrease it."
As a result, even when accounting for unknown unknowns, it looks extremely unlikely that anthropogenic warming could heat the earth enough to cause complete civilisational collapse (for a discussion of how hot that would need to be, see the first section of the main post!).
It's of course true that there are some kinds of unknown unknowns that are impossible to account for — that is, things about which we have no information. But these are rarely particularly important unknown unknowns, in part because of that lack of information: in order to have no information about something, we necessarily can't have any evidence for its existence, so from the perspective of Occam's razor, they're inherently unlikely.
At least, in macroscopic systems. You can have negative absolute temperatures in systems with a population inversion (like a laser while it's lasing), although these systems are generally considered thermodynamically hotter than positive-temperature systems (because heat flows from the negative temperature system to the positive temperature system).
I don't currently have a confident view on this beyond "We’re really not sure. It seems like OpenAI, Google DeepMind, and Anthropic are currently taking existential risk more seriously than other labs."
But I agree that if we could reach a confident position here (or even just a confident list of considerations), that would be useful for people — so thanks, this is a helpful suggestion!
Thanks, this is an interesting heuristic, but I think I don't find it as valuable as you do.
First, while I do think it'd probably be harmful in expectation to work at leading oil companies / at the Manhattan project, I'm not confident in that view — I just haven't thought about this very much.
Second, I think that AI labs are in a pretty different reference class from oil companies and the development of nuclear weapons.
Why? Roughly:
Whether, in a broad sense, capabilities advances are good or bad is pretty unclear. (Note some capabilities advances in particular areas are very clearly harmful.) In comparison, I do think that, in a broad sense, the development of nuclear weapons, and the release of greenhouse gases are harmful.
Unlike with oil companies and the Manhattan Project, I think that there's a good chance that a leading, careful AI project could be a huge force for good, substantially reducing existential risk — and so it seems weird not to consider working at what could be one of the world's most (positively) impactful organisations. Of course, you should also consider the chance that the organisation could be one of the world's most negatively impactful organisations.
I think that for many people (but not everyone) and for many roles they might work in (but not all roles), this is a reasonable plan.
Most importantly, I think it's true that working at a top AI lab as an engineer is one of the best ways to build technical skills (see the section above on "it's often excellent career capital").
I'm more sceptical about the ability to push towards safe decisions (see the section above on "you may be able to help labs reduce risks").
The right answer here depends a lot on the specific role. I think it's important to remember than not all AI capabilities work is necessarily harmful (see the section above on "you might advance AI capabilities, which could be (really) harmful"), and that top AI labs could be some of the most positive-impact organisations in the world (see the section above on "labs could be a huge force for good - or harm"). On the other hand, there are roles that seem harmful to me (see "how can you mitigate the downsides of this option").
I'm not sure of the relevance of "having a good understanding of how to do alignment" to your question. I'd guess that lots of knowing "how to do alignment" is being very good at ML engineering or ML research in general, and that working at a top AI lab is one of the best ways to learn those skills.
The Portuguese version at 80000horas.com.br is a project of Altruísmo Eficaz Brasil. We often give people permission to translate our content when they ask - but as to when, that would be up to Altruísmo Eficaz Brasil! Sorry I can't give you a more concrete answer.
There are important reasons to think that the change by the EA community is within the measurement error of these surveys, which makes this less noteworthy.
(Like say you put +/- 10 years and +/- 10% on all these answers - note there are loads of reasons why you wouldn't actually assess the uncertainty like this, (e.g. probabilities can't go below 0 or above 1), but just to get a feel for the uncertainty this helps. Well, then you get something like:
10%-30% chance of TAI by 2026-2046
40%-60% by 2050-2070
and 75%-95% by 2100
Then many many EA timelines and shifts in EA timelines fall within those errors.)
2. Low response rates + selection biases + not knowing the direction of those biases
The surveys plausibly had a bunch of selection biases in various directions.
This means you need a higher sample to converge on the population means, so the surveys probably aren't representative. But we're much less certain in which direction they're biased.
For example, you might think researchers who go to the top AI conferences are more likely to be optimistic about AI, because they have been selected to think that AI research is doing good. Alternatively, you might think that researchers who are already concerned about AI are more likely to respond to a survey asking about these concerns
3. Other problems, like inconsistent answers in the survey itself
AI impacts wrote some interesting caveats here, including:
Asking people about specific jobs massively changes HLMI forecasts. When we asked some people when AI would be able to do several specific human occupations, and then all human occupations (presumably a subset of all tasks), they gave very much later timelines than when we just asked about HLMI straight out. For people asked to give probabilities for certain years, the difference was a factor of a thousand twenty years out! (10% vs. 0.01%) For people asked to give years for certain probabilities, the normal way of asking put 50% chance 40 years out, while the ‘occupations framing’ put it 90 years out. (These are all based on straightforward medians, not the complicated stuff in the paper.)
People consistently give later forecasts if you ask them for the probability in N years instead of the year that the probability is M. We saw this in the straightforward HLMI question, and most of the tasks and occupations, and also in most of these things when we tested them on mturk people earlier. For HLMI for instance, if you ask when there will be a 50% chance of HLMI you get a median answer of 40 years, yet if you ask what the probability of HLMI is in 40 years, you get a median answer of 30%.
Thanks for this thoughtful post! I think I stand by my 1 in 10,000 estimate despite this.
A few short reasons:
Broad things: First, these scenarios and scenarios like them are highly conjunctive (many rare things need to happen), which makes any one scenario unlikely (although of course there may be many such scenarios). Second, I think these and similar scenarios are reason to think there may be a large catastrophe, but large and existential are a long way apart. (I discuss this a bit here but don't come to a strong overall conclusion. More work on this would be great.)
On inducing nuclear war:My estimate of the direct risk of nuclear war is 1 in 10,000, and the indirect risk is 1 in 1,000. It seems like the chances that climate change causes a nuclear war, weighted by the extent to which the war was more likely by virtue of climate change and not e.g. geopolitical tensions unrelated to climate change is, while subjective and difficult to judge, probably much less than 10%. If it's say 1%, this gives less than 1 in 100,000 indirect x-risk from climate change. This seems a bit small, but consistent with my 1 in 10,000 estimate. Note this includes inducing nuclear war from ways other than crop failure.
On runaway warming: My understanding is that the main limit here is how many fossil fuels it's possible to recover from the ground - see more here. Even taking into account uncertainty and huge model error, it seems highly unlikely that we'll end up with runaway warming that itself leads to extinction. I'd also add that lots of the reduction in risk occurs because climate change is a gradual catastrophe (unlike a pandemic or nuclear war), which means that, for example, we may find other emissionless technology (e.g. nuclear fusion) or get over our fear of nuclear fission, etc., reducing the risk of resource depletion. Relatedly, unless there is extremely fast runaway warming over only a few years, the gradual nature of climate change increases the chances of successful adaptation to a warmer environment. (Again, I mean adaptation to prevent an existential catastrophe - a large catastrophe that isn't quite existential seems far far more likely.)
On coastal cities: I'd guess the existential risk from war breaking out between great powers is also around 1 in 10,000 (within an order of magnitude or so), although I've thought about this less. So again, while cyanobacteria blooms sounds like a not-impossible way in which climate change could lead to war (personally I'd be more worried about flooding and migration crises in South Asia), I think this is all consistent with my 1 in 10,000 estimate.
If it helps at all, my subjective estimate of the risk from AI is probably around 1%, and approximately none of that comes from worrying about killer nanobots. I wrote about what an AI-caused existential catastrophe might actually look like here.
Hi! Wanted to follow up as the author of the 80ksoftware engineering career review, as I don't think this gives an accurate impression. A few things to say:
I try to have unusually high standards for explaining why I believe the things I write, so I really appreciate people pushing on issues like this.
At the time, when you responded to <the Anthropic person>, you said "I think <the Anthropic person> is probably right" (although you added "I don't think it's a good idea to take this sort of claim on trust for important career prioritisation research").
When I leave claims like this unsourced, it’s usually because I (and my editors) think they’re fairly weak claims, and/or they lack a clear source to reference. That is, the claim is effectively is a piece of research based on general knowledge (e.g. I wouldn't source the claim "Biden is the President of the USA”) and/or interviews with a range of experts, and the claim is weak or unimportant enough not to investigate further. (FWIW I think it’s likely I should have prioritised writing a longer footnote on why I believe this claim.)
The closest data is the three surveys of NeurIPS researchers, but these are imperfect. They ask how long it will take until there is "human-level machine intelligence". The median expert asked thought there was an around 1 in 4 chance of this by 2036. Of course, it's not clear that HLMI and transformative AI are the same thing, or that thinking HLMI being developed soon necessarily means that HLMI will be made by scaling and adapting existing ML methods. In addition, no survey data pre-dates 2016, so it's hard to say that these views have changed based solely on survey data. (I've written more about these surveys and their limitationshere, with lots of detail in footnotes; and I discuss the timelines parts of those surveys in the second paragraphhere.)
As a result, when I made this claim I was relying on three things. First, that there are likely correlations that make the survey data relevant (i.e., that many people answering the survey think that HLMI will be relatively similar to or cause transformative AI, and that many people answering the survey think that if HLMI is developed soon that suggests it will be ML-based). Second, that people did not think that ML could produce HLMI in the past (e.g. because other approaches like symbolic AI were still being worked on, because texts like Superintelligence do not focus on ML and this was not widely remarked upon at the time despite that book’s popularity, etc.). Third, that people in the AI and ML fields who I spoke to had a reasonable idea of what other experts used to think and how that has changed (note I spoke to many more people than the one person who responded to you in the comments on my piece)!
It's true that there may be selection bias on this third point. I'm definitely concerned about selection bias for shorter timelines in general in the community, and plan to publish something about this at some point. But in general I think that the best way, as an outsider, to understand what prevailing opinions are in a field, is to talk to people in that field – rather than relying on your own ability to figure out trends across many papers, many of which are difficult to evaluate, many of which may not replicate. I also think that asking about what others in the field think, rather than what the people you're talking to think, is a decent (if imperfect) way of dealing with that bias.
Overall, I thought the claim I made was weak enough (e.g. "many experts" not "most experts" or "all experts") that I didn't feel the need to evaluate this further.
It's likely, given you’ve raised this, that I should have put this all in a footnote. The only reason I didn't is that I try to prioritise, and I thought this claim was weak enough to not need much substantiation. I may go back and change that now (depending on how I prioritise this against other work).
By "engagement time" I mean exactly "time spent on the website".
Thanks for this comment Tyler!
To clarify what I mean by unknown unknowns, here's a climate-related example: We're uncertain about the strength of various feedback loops, like how much warming could be produced by cloud feedbacks. We'd then classify "cloud feedbacks" as a known unknown. But we're also uncertain about whether there are feedback loops we haven't identified. Since we don't know what these might be, these loops are unknown unknowns. As you say, the known feedback loops don't seem likely to warm earth enough to cause a complete destruction of civilisation, which means that if climate change were to lead to civilisational collapse, that would probably be because of something we failed to consider.
But here's the thing: generally we do know something about unknown unknowns.[1] In the case of these unknown feedback loops, we can place some constraints on them. For example:
In fact, we can gather a broad variety of evidence about these unknown unknowns, using various different lines of evidence. These lines of evidence include:
Accounting for these multiple lines of evidence is exactly what the 6th Assessment Report attempts to do when calculating climate sensitivity (how much Earth's surface will cool or warm after a specified factor causes a change in its climate system):[3]
That is, as I mentioned in the main post "the IPCC's Sixth Assessment Report... attempts to account for structural uncertainty and unknown unknowns. Roughly, they find it’s unlikely that all the various lines of evidence are biased in just one direction — for every consideration that could increase warming, there are also considerations that could decrease it."
As a result, even when accounting for unknown unknowns, it looks extremely unlikely that anthropogenic warming could heat the earth enough to cause complete civilisational collapse (for a discussion of how hot that would need to be, see the first section of the main post!).
If you're interested in diving into this further, I'd suggest taking a look at the original paper "An Assessment of Earth's Climate Sensitivity Using Multiple Lines of Evidence" by Sherwood et al., or Why low-end 'climate sensitivity' can now be ruled out, a popular summary by the paper's authors.
It's of course true that there are some kinds of unknown unknowns that are impossible to account for — that is, things about which we have no information. But these are rarely particularly important unknown unknowns, in part because of that lack of information: in order to have no information about something, we necessarily can't have any evidence for its existence, so from the perspective of Occam's razor, they're inherently unlikely.
At least, in macroscopic systems. You can have negative absolute temperatures in systems with a population inversion (like a laser while it's lasing), although these systems are generally considered thermodynamically hotter than positive-temperature systems (because heat flows from the negative temperature system to the positive temperature system).
From the introduction to section 7.5 of the Working Group I contribution to the Sixth Assessment Report (p.993).
I don't currently have a confident view on this beyond "We’re really not sure. It seems like OpenAI, Google DeepMind, and Anthropic are currently taking existential risk more seriously than other labs."
But I agree that if we could reach a confident position here (or even just a confident list of considerations), that would be useful for people — so thanks, this is a helpful suggestion!
Thanks, this is an interesting heuristic, but I think I don't find it as valuable as you do.
First, while I do think it'd probably be harmful in expectation to work at leading oil companies / at the Manhattan project, I'm not confident in that view — I just haven't thought about this very much.
Second, I think that AI labs are in a pretty different reference class from oil companies and the development of nuclear weapons.
Why? Roughly:
Because these issues are difficult and we don’t think we have all the answers, I also published a range of opinions about a related question in our anonymous advice series. Some of the respondents took a very sceptical view of any work that advances capabilities, but others disagreed.
Hi Yonatan,
I think that for many people (but not everyone) and for many roles they might work in (but not all roles), this is a reasonable plan.
Most importantly, I think it's true that working at a top AI lab as an engineer is one of the best ways to build technical skills (see the section above on "it's often excellent career capital").
I'm more sceptical about the ability to push towards safe decisions (see the section above on "you may be able to help labs reduce risks").
The right answer here depends a lot on the specific role. I think it's important to remember than not all AI capabilities work is necessarily harmful (see the section above on "you might advance AI capabilities, which could be (really) harmful"), and that top AI labs could be some of the most positive-impact organisations in the world (see the section above on "labs could be a huge force for good - or harm"). On the other hand, there are roles that seem harmful to me (see "how can you mitigate the downsides of this option").
I'm not sure of the relevance of "having a good understanding of how to do alignment" to your question. I'd guess that lots of knowing "how to do alignment" is being very good at ML engineering or ML research in general, and that working at a top AI lab is one of the best ways to learn those skills.
The Portuguese version at 80000horas.com.br is a project of Altruísmo Eficaz Brasil. We often give people permission to translate our content when they ask - but as to when, that would be up to Altruísmo Eficaz Brasil! Sorry I can't give you a more concrete answer.
There are important reasons to think that the change by the EA community is within the measurement error of these surveys, which makes this less noteworthy.
(Like say you put +/- 10 years and +/- 10% on all these answers - note there are loads of reasons why you wouldn't actually assess the uncertainty like this, (e.g. probabilities can't go below 0 or above 1), but just to get a feel for the uncertainty this helps. Well, then you get something like:
Then many many EA timelines and shifts in EA timelines fall within those errors.)
Reasons why these surveys have huge error
1. Low response rates.
The response rates were really quite low.
2. Low response rates + selection biases + not knowing the direction of those biases
The surveys plausibly had a bunch of selection biases in various directions.
This means you need a higher sample to converge on the population means, so the surveys probably aren't representative. But we're much less certain in which direction they're biased.
Quoting me:
3. Other problems, like inconsistent answers in the survey itself
AI impacts wrote some interesting caveats here, including:
The 80k podcast on the 2016 survey goes into this too.
Thanks for this! Looks like we actually roughly agree overall :)
Thanks for this thoughtful post! I think I stand by my 1 in 10,000 estimate despite this.
A few short reasons:
If it helps at all, my subjective estimate of the risk from AI is probably around 1%, and approximately none of that comes from worrying about killer nanobots. I wrote about what an AI-caused existential catastrophe might actually look like here.
Hi! Wanted to follow up as the author of the 80k software engineering career review, as I don't think this gives an accurate impression. A few things to say:
The closest data is the three surveys of NeurIPS researchers, but these are imperfect. They ask how long it will take until there is "human-level machine intelligence". The median expert asked thought there was an around 1 in 4 chance of this by 2036. Of course, it's not clear that HLMI and transformative AI are the same thing, or that thinking HLMI being developed soon necessarily means that HLMI will be made by scaling and adapting existing ML methods. In addition, no survey data pre-dates 2016, so it's hard to say that these views have changed based solely on survey data. (I've written more about these surveys and their limitations here, with lots of detail in footnotes; and I discuss the timelines parts of those surveys in the second paragraph here.)
As a result, when I made this claim I was relying on three things. First, that there are likely correlations that make the survey data relevant (i.e., that many people answering the survey think that HLMI will be relatively similar to or cause transformative AI, and that many people answering the survey think that if HLMI is developed soon that suggests it will be ML-based). Second, that people did not think that ML could produce HLMI in the past (e.g. because other approaches like symbolic AI were still being worked on, because texts like Superintelligence do not focus on ML and this was not widely remarked upon at the time despite that book’s popularity, etc.). Third, that people in the AI and ML fields who I spoke to had a reasonable idea of what other experts used to think and how that has changed (note I spoke to many more people than the one person who responded to you in the comments on my piece)!
It's true that there may be selection bias on this third point. I'm definitely concerned about selection bias for shorter timelines in general in the community, and plan to publish something about this at some point. But in general I think that the best way, as an outsider, to understand what prevailing opinions are in a field, is to talk to people in that field – rather than relying on your own ability to figure out trends across many papers, many of which are difficult to evaluate, many of which may not replicate. I also think that asking about what others in the field think, rather than what the people you're talking to think, is a decent (if imperfect) way of dealing with that bias.
Overall, I thought the claim I made was weak enough (e.g. "many experts" not "most experts" or "all experts") that I didn't feel the need to evaluate this further.