Reading through this I have some pretty significant concerns.
First the model behind the “$0.24 for each life saved” figure seems very suspect:
The assumption of radical life extension technology being developed by 2100 is totally unsupported, with the one citation being to a survey of machine learning researchers which gave a 50% chance of AI reaching human level in all activities by 2062. It is unclear how this relates to the development of radical life extension technology however, something significantly out of reach of (current) human level ability
It assumes that Metformin would definitely extend human life expectancy by 1 year. However since much of the current evidence is from animal or cohort studies, it cannot be assumed that it definitely has an effect.
Even given all of the above it is not clear to me that the model actually provides a good estimate of the number of people likely to be saved. It is based on an extreme simplification (assuming 5 billion people are all born in 2020 and take Metformin for their entire lives) and as far as I can tell there is no attempt to justify its accuracy.
I am also unconvinced by the quality of argument elsewhere in the paper. For instance in the section “False arguments against badness of death” they list common arguments against the badness of death and then claim to refute them. However the responses are often extremely shallow and do not engage at all with the core of the argument. Here are some examples:
1) Stopping death will result in overpopulation. Only the number of births counts for overpopulation (Gavrilov & Gavrilova, 2010), and short-lived organisms like lemmings are the type of species that suffers from overpopulation.
2) Stopping death could result in stagnation, infinite totalitarianism, or other bad social outcomes. Our world changes so quickly that there is no time for such “stability” to take root.
3) Stopping death takes opportunity from non-born people, who would be born if resources were freed up by death of aging humans. The idea of an infinite universe where everything is possible kills the objection.
The paper contains many other arguments of a similar level of quality, and so although I largely agree with many of its conclusions, I find it generally very uncompelling.
Finally, as the most minor point there are quite a high number of grammatical issues.
Taking median date of the AI arrival like 2062 is not informative as in half cases it will not be here at 2062. The date of 2100 is taken as the date when it (or other powerful life-extending technology) almost sure will appear as a very conservative estimate. Maybe it should be justified more in the text.
Yes, it is assumed by Barzilai and gwern that metformin will extend human life 1 year, based on many human cohorts studies, but to actually prove it we need TAME study, and until this study is finished, metformin can’t be used as a life-extending drug. So any year of delay of the experiment means a year in the delay in global implementation. For now, it is already delayed for 2 years by luck of funds.
Given all uncertainty, the simplified model provides only an order of magnitude of the effect, but a more detailed model which take into account actual age distribution is coming.
As the paper is already too long, we tried to outline the main arguments or provide links to the articles where detailed refutation is presented, as in case of Gavrilov, 2010, where the problem of overpopulation is analysed in detail. But it is obvious now that this points should be clarified.
The next round of professional grammar editing is scheduled.
Thanks for the reply. Despite my very negative tone I do think this is an important work, and doing good cost benefit analysis like these is very difficult.
Taking median date of the AI arrival like 2062 is not informative as in half cases it will not be here at 2062. The date of 2100 is taken as the date when it (or other powerful life-extending technology) almost sure will appear as a very conservative estimate.
I don’t share the intuition that human level AI will rapidly cause the creation of powerful life-extending technology. This seems to be relying on a rapid takeoff scenario, which while plausible I don’t think can be taken as anything like certain. I think if this is the argument it should be spelled out clearly.
With regards to the effectiveness of metformin, my argument is that you should include a discount factor of a half or so to include the probably that it does not pass the human level trial.
Given all uncertainty, the simplified model provides only an order of magnitude of the effect
My issue is that I don’t see any arguments that the model is even likely to be accurate to within an order of magnitude.
I’m glad to here a more detailed model is in the works, as I said I think this is important work, but that makes getting it right all the more pivotal.
As the paper is already too long, we tried to outline the main arguments or provide links to the articles where detailed refutation is presented, as in case of Gavrilov, 2010, where the problem of overpopulation is analysed in detail. But it is obvious now that this points should be clarified.
I think if the intention is just to link to other articles with detailed refutations you should just do that and not attempt to summerise (or make it clear this is at most a very rough outline). However for two of the examples I listed no other article is linked.
Reading through this I have some pretty significant concerns.
First the model behind the “$0.24 for each life saved” figure seems very suspect:
The assumption of radical life extension technology being developed by 2100 is totally unsupported, with the one citation being to a survey of machine learning researchers which gave a 50% chance of AI reaching human level in all activities by 2062. It is unclear how this relates to the development of radical life extension technology however, something significantly out of reach of (current) human level ability
It assumes that Metformin would definitely extend human life expectancy by 1 year. However since much of the current evidence is from animal or cohort studies, it cannot be assumed that it definitely has an effect.
Even given all of the above it is not clear to me that the model actually provides a good estimate of the number of people likely to be saved. It is based on an extreme simplification (assuming 5 billion people are all born in 2020 and take Metformin for their entire lives) and as far as I can tell there is no attempt to justify its accuracy.
I am also unconvinced by the quality of argument elsewhere in the paper. For instance in the section “False arguments against badness of death” they list common arguments against the badness of death and then claim to refute them. However the responses are often extremely shallow and do not engage at all with the core of the argument. Here are some examples:
The paper contains many other arguments of a similar level of quality, and so although I largely agree with many of its conclusions, I find it generally very uncompelling.
Finally, as the most minor point there are quite a high number of grammatical issues.
Thank you for review.
Taking median date of the AI arrival like 2062 is not informative as in half cases it will not be here at 2062. The date of 2100 is taken as the date when it (or other powerful life-extending technology) almost sure will appear as a very conservative estimate. Maybe it should be justified more in the text.
Yes, it is assumed by Barzilai and gwern that metformin will extend human life 1 year, based on many human cohorts studies, but to actually prove it we need TAME study, and until this study is finished, metformin can’t be used as a life-extending drug. So any year of delay of the experiment means a year in the delay in global implementation. For now, it is already delayed for 2 years by luck of funds.
Given all uncertainty, the simplified model provides only an order of magnitude of the effect, but a more detailed model which take into account actual age distribution is coming.
As the paper is already too long, we tried to outline the main arguments or provide links to the articles where detailed refutation is presented, as in case of Gavrilov, 2010, where the problem of overpopulation is analysed in detail. But it is obvious now that this points should be clarified.
The next round of professional grammar editing is scheduled.
Thanks for the reply. Despite my very negative tone I do think this is an important work, and doing good cost benefit analysis like these is very difficult.
I don’t share the intuition that human level AI will rapidly cause the creation of powerful life-extending technology. This seems to be relying on a rapid takeoff scenario, which while plausible I don’t think can be taken as anything like certain. I think if this is the argument it should be spelled out clearly.
With regards to the effectiveness of metformin, my argument is that you should include a discount factor of a half or so to include the probably that it does not pass the human level trial.
My issue is that I don’t see any arguments that the model is even likely to be accurate to within an order of magnitude.
I’m glad to here a more detailed model is in the works, as I said I think this is important work, but that makes getting it right all the more pivotal.
I think if the intention is just to link to other articles with detailed refutations you should just do that and not attempt to summerise (or make it clear this is at most a very rough outline). However for two of the examples I listed no other article is linked.
May I share with you the next version when all that changes will be done? I expect that the next revision will appear in 2 months.
Sure, although I’m not sure how much time I will have to look it over. My email is alexbarry40@gmail.com.