Peter Singer on AI, Transhumanism, and Ethics
- Published
The effective altruism movement was strongly influenced by Peter Singer in its early years (The most good you can do), but some developments, including longtermism, seem to have moved somewhat away from Singer’s conception of effective altruism.
Singer’s positions
For a start, Singer explicitly offered a somewhat critical assessment of longtermism in 2019. (See Is AI Safety ‘Rather Speculative Long-Termism’?.)
In his 2018 interview with Singularity Weblog, despite the interviewer’s persistent leading questions, he repeatedly states his own view: that the so-called “AI crisis” is less urgent than the climate crisis or hunger, and that he is skeptical whether it is a real problem at all.
Thanks to Choi Seung-joon, I also found a more recent article, from 2021, in which Singer expresses a more clearly critical view of longtermism.
In 2022, he tweeted a recommendation of Will MacAskill’s What We Owe the Future, but the wording is subtle. He recommends it as a book about utilitarianism and effective altruism, while saying we need not agree with everything in it—perhaps referring to longtermism?
Will MacAskill’s What We Owe the Future has just been published. You don’t have to agree with everything in it to think that this is an important book. Good to see it getting a lot of well-deserved media attention that raises the profile of effective altruism and utilitarianism.
Speciesism in longtermism
I am not even sure whether the utilitarianism professed by longtermists fits the concept Singer understands and explains. (See Utilitarianism - A very short introduction.) Like Jeremy Bentham, Singer gives no special status to the human species, whereas longtermists seem excessively focused on human flourishing.
By Singer’s standards, I suspect their position may verge on speciesism. Asked what he would do if an AI superior to humans appeared and concluded that human extinction was desirable from a utilitarian perspective, Singer replies that he considers this extremely unlikely, but if it really happened and the judgment were correct, we would have to accept the conclusion. He assigns no special extra value to the happiness of the human species. He even says that human resistance to such a conclusion would be immoral. (Source: the Singularity Weblog interview above.)
The following video is from January 2024. Starting around minute 25, he briefly discusses AI and animal ethics, criticizing the failure of most AI-ethics discussions to consider animals. Just before the end he discusses longtermism, saying he does not expect artificial superintelligence to arrive very quickly and does not consider researching responses to it particularly urgent right now. As expected.
Exploring Effective Altruism at Congreso Futuro 2024
Positions on AI, transhumanism, and related issues
Singer gave another interview to Singularity Weblog in 2024.
(Philosopher Peter Singer on AI, Transhumanism and Ethics)
Some notes:
- 14:00 — Some sympathy for transhumanism. Change carries risks, but we cannot assume evolution designed an ideal state either.
- 14:50 — Asked about his greatest dream, he answers: a world without suffering, or with minimal suffering. He mentions alternative meats and plant-based ingredients as one route.
- 17:00 — His greatest fear is the climate crisis. A different emphasis from longtermism.
- 18:20 — He considers the AI crisis less urgent than the climate crisis. AI may take another 50–100 years to become sufficiently intelligent, leaving us time to think. Though he considers it highly unlikely and does not believe it will happen, he is asked to suppose hypothetically that a very intelligent AI correctly judges human extinction to accord with utilitarianism. Under those assumptions, he thinks he would accept the conclusion. He does not believe that the human species in particular must be happy. If humans collectively resisted, he would regard that as immoral.
- 29:30 — Hunger also matters. Fortunately it is declining, but remains a major problem.
- 34:30 — Discussion of effective altruism begins.
- 43:00 — Asked to elaborate on his remark that he has no problem with transhumanism (??), he says he does not shy away from discussing genetic modification and other ways of enhancing humans to act more morally and thoughtfully. Humans have room for improvement. We react with intense emotion to seeing someone hit a child, but rarely react that way to carbon emissions.
- 51:00 — Asked whether greater intelligence really implies greater morality—probably an attempt to raise AI risk—Singer says greater intelligence would tend to make one more moral, because it helps with reflective thought and rational thinking that restrains emotion.
- 55:00 — Experts say the singularity may arrive within 10–20 years. Would he reconsider? His colleague Nick Bostrom agrees too. Singer replies that at a broad discussion the previous year involving ethicists and industry experts from DeepMind, Microsoft, and elsewhere, he asked why AI takeover was not on the agenda. They all laughed, saying we were nowhere near it. He is not an expert himself; his view comes from listening to other experts.