X discussion: Turing Award winner Richard Sutton shares the "intelligence explosion" passage he wrote at 17 in 1975

Background and the problem

In 2026, recursive self-improvement has moved from science fiction to roadmaps that AI company CEOs discuss in public, and developers use ever stronger coding agents every day, so "will AI soon surpass humans" is one of the most asked questions in the community. At 23:42 UTC on October 30, reinforcement learning pioneer Richard Sutton posted on X a passage he wrote in 1975 at age 17, saying it is one reason he is sanguine about the technological singularity: he has seen it coming for over 50 years.

How it works

The 1975 passage argues in four steps. To best aid his problem-solving machinery, man must first understand it, and that understanding was poor. Once a breakthrough in understanding the brain occurs, improvement comes very fast and exponentially. An intelligence greater than man''s should be able to create one greater than itself, just as man created it. So intelligence could shoot up within years, and man will either be left behind by a superior intelligence he can no longer understand or become that intelligence himself. According to ACM, Sutton and Andrew Barto received the 2024 Turing Award for the conceptual and algorithmic foundations of reinforcement learning, and their temporal-difference and policy-gradient methods still underpin techniques such as RLHF.

Benchmarks and results

This is an opinion debate, so there are no benchmarks. What can be verified is engagement: read through the X API around 19:07 SGT on October 31, the post had 2,376 likes, 232 reposts, 45 quotes, 97 replies, 882 bookmarks and about 167,000 views. Among quote posts, Chinese-language users focused on what human intelligence still means, with one Chinese explainer reposted more than 30 times, while an engineer proposed a third path: using AI to learn faster rather than choosing between merging and being left behind.

What developers can do

For developers, the thread is a good entry point into Sutton''s thinking. His 2019 essay The Bitter Lesson argues that general methods that scale with compute, namely search and learning, beat hand-coded human knowledge in the long run. In daily work, that suggests favoring agent designs that improve automatically as models get better and writing fewer prompt tricks that only work for today''s model. Note that the post is a personal look back, not a new prediction of a singularity timeline.