I think it would be really bad for humanity to rush to build superintelligence before we solve the difficult problem of how to make it safe. But also I think it would be a horrible tragedy if humanity never ever built superintelligence. I hope we figure out how to thread this needle with wisdom.
I agree with this fwiw. Currently I think we are in way way more danger of rushing to build it too fast than of never building it at all, but if e.g. all the nations of the world had agreed to ban it, and in fact were banning AI research more generally, and the ban had held stable for decades and basically strangled the field, I’d be advocating for judicious relaxation of the regulations (same thing I advocate for nuclear power basically).
I am not really clear that I should be worried on the scale of decades? If we’re doing a calculation of expected future years of a flourishing technologically mature civilization, slowing down for 1,000 years here in order to increase the chance of success by like 1 percentage point is totally worth it in expectation.
Given this, it seems plausible to me that one should rather spend 200 years trying to improve civilizational wisdom and decision-making rather than instead attempt to specifically just unlock regulation on AI (of course the specifics here are cruxy).
I agree that 200 years would be worth it if we actually thought that it would work. My concern is that it’s not clear civilization would get better/moresane/etc. over the next century vs. worse. And relatedly, every decade that goes by, we eat another percentage point or three of x-risk from miscellaneous other sources (nuclear war, pandemics, etc.) which basically impose a time-discount factor on our calculations large enough to make a 200 year pause seem really dangerous and bad to me.
while I agree for smaller numbers like a few decades, I don’t think I agree with a 1000 year pause.
I think (a) it’s perfectly reasonable for people to be selfish and care about superintelligence happening during their lifetime (forget future people and discount factors thereof—almost every single person alive today cares ooms more about themselves than about some random person on the other side of the planet), (b) it’s easy for “delay forever” people to basically pascal’s mug you this way, as in nuclear power (c) it’s unclear that humanity becomes monotonically more wise over time (as an unrealistic example, consider a world where we successfully create an international treaty to ensure ASI is safe, and then for some reason the entire world modern order collapses and the only actors left are random post-collapse states racing to build ASI. then it would have been better to build ASI in a functional pre-collapse world order than to delay. one could reasonably (though i personally don’t) believe that the current world order is likely to fail in the coming decades and ASI is best built now than in the ensuing chaos)
it’s perfectly reasonable for people to be selfish and care about superintelligence happening during their lifetime
Yes people are selfish, that is why you should sometimes be ready to fight against them. Point a is not a disagreement with Ben.
then for some reason the entire world modern order collapses
This is low probability on time scale of decades but is an argument people can use to justify their self-serving desires for immortality as somehow altruistic.
If I understood Eliezers argument correctly we can shorten those timescales buy improving human intelligences through methods like genetic engineering. Once majority of humans have Von Neumann level IQ I think its fine to let them decide how to proceed on AI research. Question is, how fast can this happen, and it probably would take a century or 2 at least.
>slowing down for 1,000 years here in order to increase the chance of success by like 1 percentage point is totally worth it in expectation.
Is it? What meaning of worth it is used here? If you put it on a vote, as an option, I expect it would lose. People don’t care that much about happiness of distant future people.
Given this, it seems plausible to me that one should rather spend 200 years trying to improve civilizational wisdom and decision-making
Make it a thousand, or two thousand. To Daniel’s point, societal change is not always positive on the scale of centuries. But on the scale of millennia it is. At least the last few ones.
SI can come when we say so. On the human evolutionary timescale, 10K years is short. On a cosmic timescale, it is nothing.
i think it’s plausible humans/humanity should be carefully becoming ever more intelligent ≈forever and not ever create any highly non-[human-descended] top thinker[1]
If a superintelligence governs the world, preventing extinction or permanent disempowerment for the future of humanity, without itself posing these dangers, then it could be very useful. It’s unclear how feasible setting up something like this is, before originally-humans can be uplifted to a similar level of competence. But also, uplifting humans to that level of competence doesn’t necessarily guard (the others) against permanent disempowerment or some other wasteful breakdowns of coordination, so a governance-establishing superintelligence could still be useful.
Superintelligence works as a threshold-concept for a phase change compared to the modern world. Non-superintelligent AGIs are still just an alien civilization that remains in principle similar in the kinds of things it can do to humanity (even if they reproduce to immediately fill all available compute, and think 10,000x faster). While superintelligence is something at the next level, even if it only takes non-superintelligent AGIs to transition to superintelligence a very short time (if they decide to do that, rather than to not do that).
Apart from superintelligence being a threshold-concept, there is technological maturity, the kinds of things that can’t be significantly improved upon in another 1e10 years of study, but that maybe only take 1-1000 years to figure out for the first time. And one of those things is plausibly efficient use of compute for figuring things out, which gives superintelligence at a given scale of compute. This is in particular the reason to give some credence to software-only singularity, where first AGIs quickly learn to make a shockingly better use of existing compute, so that their capabilities improve much faster than it would take them to build new computing hardware. I think the most likely reason for software-only singularity to not happen is that it’s intentionally delayed (by AGIs themselves) because of the danger it creates, rather than because it’s technologically impossible.
Every year we don’t build superintelligence is a worse tragedy than any historical tragedy you can name. Tens of millions dead, hundreds of millions suffering, etc.
That doesn’t mean we should rush ahead, because rushing ahead is most likely far worse. But we should be aware of the cost.
I am undecided as to whether superintelligence should ever be built, and with my current knowledge and perspective, I would be fine with either outcome.
Some questions:
Has the existence of humanity up to now with no superintelligence been a horrible tragedy?
What would superintelligence allow that would otherwise be forever out of reach?
Are there not also things that humanity could or would lose forever if we did create a safe superintelligence?
Superhuman artificial general intelligence could be created this century and would likely be a significant source of existential risk. Delaying the creation of superintelligent AI (ASI) could decrease total existential risk by increasing the amount of time humanity has to work on the AI alignment problem. However, since ASI could reduce most risks, delaying the creation of ASI could also increase other existential risks, especially from advanced future technologies such as synthetic biology and molecular nanotechnology. If AI existential risk is high relative to the sum of other existential risk, delaying the creation of ASI will tend to decrease total existential risk and vice-versa. Other factors such as war and a hardware overhang could increase AI risk and cognitive enhancement could decrease AI risk. To reduce total existential risk, humanity should take robustly positive actions such as working on existential risk analysis, AI governance and safety, and reducing all sources of existential risk by promoting differential technological development.
Yet before we can pass out of that stage of adolescence, we must, as adolescents, confront an adult problem: the challenge of smarter-than-human intelligence. This is the way out of the high-mortality phase of the life cycle, the way to close the window of vulnerability; it is also probably the single most dangerous risk we face. Artificial Intelligence is one road into that challenge; and I think it is the road we will end up taking. I think that, in the end, it will prove easier to build a 747 from scratch, than to scale up an existing bird or graft on jet engines.
I think it would be really bad for humanity to rush to build superintelligence before we solve the difficult problem of how to make it safe. But also I think it would be a horrible tragedy if humanity never ever built superintelligence. I hope we figure out how to thread this needle with wisdom.
I agree with this fwiw. Currently I think we are in way way more danger of rushing to build it too fast than of never building it at all, but if e.g. all the nations of the world had agreed to ban it, and in fact were banning AI research more generally, and the ban had held stable for decades and basically strangled the field, I’d be advocating for judicious relaxation of the regulations (same thing I advocate for nuclear power basically).
I am not really clear that I should be worried on the scale of decades? If we’re doing a calculation of expected future years of a flourishing technologically mature civilization, slowing down for 1,000 years here in order to increase the chance of success by like 1 percentage point is totally worth it in expectation.
Given this, it seems plausible to me that one should rather spend 200 years trying to improve civilizational wisdom and decision-making rather than instead attempt to specifically just unlock regulation on AI (of course the specifics here are cruxy).
I agree that 200 years would be worth it if we actually thought that it would work. My concern is that it’s not clear civilization would get better/moresane/etc. over the next century vs. worse. And relatedly, every decade that goes by, we eat another percentage point or three of x-risk from miscellaneous other sources (nuclear war, pandemics, etc.) which basically impose a time-discount factor on our calculations large enough to make a 200 year pause seem really dangerous and bad to me.
while I agree for smaller numbers like a few decades, I don’t think I agree with a 1000 year pause.
I think (a) it’s perfectly reasonable for people to be selfish and care about superintelligence happening during their lifetime (forget future people and discount factors thereof—almost every single person alive today cares ooms more about themselves than about some random person on the other side of the planet), (b) it’s easy for “delay forever” people to basically pascal’s mug you this way, as in nuclear power (c) it’s unclear that humanity becomes monotonically more wise over time (as an unrealistic example, consider a world where we successfully create an international treaty to ensure ASI is safe, and then for some reason the entire world modern order collapses and the only actors left are random post-collapse states racing to build ASI. then it would have been better to build ASI in a functional pre-collapse world order than to delay. one could reasonably (though i personally don’t) believe that the current world order is likely to fail in the coming decades and ASI is best built now than in the ensuing chaos)
Yes people are selfish, that is why you should sometimes be ready to fight against them. Point a is not a disagreement with Ben.
This is low probability on time scale of decades but is an argument people can use to justify their self-serving desires for immortality as somehow altruistic.
If I understood Eliezers argument correctly we can shorten those timescales buy improving human intelligences through methods like genetic engineering. Once majority of humans have Von Neumann level IQ I think its fine to let them decide how to proceed on AI research. Question is, how fast can this happen, and it probably would take a century or 2 at least.
>slowing down for 1,000 years here in order to increase the chance of success by like 1 percentage point is totally worth it in expectation.
Is it? What meaning of worth it is used here? If you put it on a vote, as an option, I expect it would lose. People don’t care that much about happiness of distant future people.
Make it a thousand, or two thousand. To Daniel’s point, societal change is not always positive on the scale of centuries. But on the scale of millennia it is. At least the last few ones.
SI can come when we say so. On the human evolutionary timescale, 10K years is short. On a cosmic timescale, it is nothing.
i think it’s plausible humans/humanity should be carefully becoming ever more intelligent ≈forever and not ever create any highly non-[human-descended] top thinker[1]
i also think it’s confused to speak of superintelligence as some definite thing (like, to say “create superintelligence”, as opposed to saying “create a superintelligence”), and probably confused to speak of safe fooming as a problem that could be “solved”, as opposed to one needing to indefinitely continue to be thoughtful about how one should foom
If a superintelligence governs the world, preventing extinction or permanent disempowerment for the future of humanity, without itself posing these dangers, then it could be very useful. It’s unclear how feasible setting up something like this is, before originally-humans can be uplifted to a similar level of competence. But also, uplifting humans to that level of competence doesn’t necessarily guard (the others) against permanent disempowerment or some other wasteful breakdowns of coordination, so a governance-establishing superintelligence could still be useful.
Superintelligence works as a threshold-concept for a phase change compared to the modern world. Non-superintelligent AGIs are still just an alien civilization that remains in principle similar in the kinds of things it can do to humanity (even if they reproduce to immediately fill all available compute, and think 10,000x faster). While superintelligence is something at the next level, even if it only takes non-superintelligent AGIs to transition to superintelligence a very short time (if they decide to do that, rather than to not do that).
Apart from superintelligence being a threshold-concept, there is technological maturity, the kinds of things that can’t be significantly improved upon in another 1e10 years of study, but that maybe only take 1-1000 years to figure out for the first time. And one of those things is plausibly efficient use of compute for figuring things out, which gives superintelligence at a given scale of compute. This is in particular the reason to give some credence to software-only singularity, where first AGIs quickly learn to make a shockingly better use of existing compute, so that their capabilities improve much faster than it would take them to build new computing hardware. I think the most likely reason for software-only singularity to not happen is that it’s intentionally delayed (by AGIs themselves) because of the danger it creates, rather than because it’s technologically impossible.
Every year we don’t build superintelligence is a worse tragedy than any historical tragedy you can name. Tens of millions dead, hundreds of millions suffering, etc.
That doesn’t mean we should rush ahead, because rushing ahead is most likely far worse. But we should be aware of the cost.
I am undecided as to whether superintelligence should ever be built, and with my current knowledge and perspective, I would be fine with either outcome.
Some questions:
Has the existence of humanity up to now with no superintelligence been a horrible tragedy?
What would superintelligence allow that would otherwise be forever out of reach?
Are there not also things that humanity could or would lose forever if we did create a safe superintelligence?
In 2022 I wrote an article that is relevant to this question called How Do AI Timelines Affect Existential Risk? Here is the abstract:
Artificial Intelligence as a Positive and Negative Factor in Global Risk (Yudkowsky, 2008) is also relevant. Excerpt from the conclusion:
How can we know that the problem is solved—and now we can safely proceed?