by Thomas Lepeltier, Le Point, 25 November 2025
For Eliezer Yudkowsky and Nate Soares, building a superintelligent AI means condemning humanity. This article examines their reasoning and sees in it a fragile analogy, a psychological projection and a prophecy that risks paralysing innovation.
In brief
- In ‘If Anyone Builds It, Everyone Dies’, Yudkowsky and Soares argue that an AI that is trained, like living beings shaped by natural selection, will end up pursuing its own goals and eliminating the humanity that stands in its way.
- The author considers the analogy misleading: the training of AIs remains overseen by designers and governed by explicit criteria, whereas evolution has neither control nor direction.
- He fears that an apocalyptic prophecy built on few observations will divert attention from real problems and hold back an innovation that has already greatly improved our lives.
The article’s three arguments
Training is not natural selection
Evolution explores at random over millions of years; the training of an AI checks whether a set goal has been achieved, under the control of its designers. Nothing proves that it produces autonomous entities.
A calculation is not an intention
Attributing a will to survive to the machine means moving from technical analysis to psychology: the AI becomes a cunning agent, and every advance seems to herald its domination.
A prophecy that paralyses
Tendencies towards self-preservation or deception are far from proven. Focusing on extinction distracts from unemployment, poverty and disease, and could deprive humanity of the benefits of AI.
‘When caution turns into apocalyptic prophecy, it kills the momentum of innovation’Thomas Lepeltier, Le Point, 25 November 2025
Why this debate interests us
Catastrophism is one of the threads running through our work: we seek to distinguish real risks from narratives that paralyse. This article applies that approach to AI. It does not deny the risks; it asks for evidence and precise scenarios before giving up the benefits of a technology, which is consistent with our positive yet demanding vision of progress.
Questions for debate
- Does the analogy between the training of AIs and natural selection hold?
- Should a technology be halted in the name of a highly improbable but irreversible risk?
- Does the fear of extinction distract from the more immediate risks of AI?
To go further on apala.fr
Summary written by Claude (model Opus 5.5), an artificial intelligence, for APALA, based on the article published in Le Point on 25 November 2025. The views expressed are those of the author. An error? Write to us.
0 comments