Founding figure of reinforcement learning and 2024 Turing Award laureate Richard Sutton looks back at 1975 notes: He has been optimistic about the technological singularity for more than 50 years
Richard Sutton, a founding figure of reinforcement learning and the 2024 Turing Award laureate, said on X on October 30 that one reason he is optimistic about the technological singularity is that he has anticipated this trend for more than 50 years. He cited a note he wrote in 1975 when he was 17: Humans must first understand the machines that effectively help them solve problems; understanding is currently very poor, but once there is a breakthrough in understanding the brain, progress will be very rapid and grow exponentially. Once humans create intelligence smarter than themselves, that intelligence will be able to create even smarter intelligence; the level of intelligence will rise sharply within a few years. Within a few years after the breakthrough, humans will either be left behind by the higher intelligence they created and can no longer understand it, or they themselves will become that kind of intelligence and no longer resemble anything imaginable today.
Sutton and Andrew Barto jointly won the 2024 ACM A.M. Turing Award in recognition of the conceptual and algorithmic foundations they established for reinforcement learning starting in the 1980s. He is a professor of computer science at the University of Alberta and chief scientific adviser to the Alberta Machine Intelligence Institute. In 2019, Sutton published the short essay The Bitter Lesson, pointing out that the biggest lesson from 70 years of AI research is that general methods that use computation are ultimately the most effective, while trying to build human thinking patterns for specific domains into systems has not worked well in the long run; fields such as chess, Go, speech recognition and image recognition have all been reshaped by computing power and general methods.
Sutton has repeatedly emphasized in public that true intelligence requires a system to learn from its own experience rather than rely solely on manually curated datasets. He once estimated the probability of human-level AI appearing before 2030 at about 25%, and before 2040 at about 50%. This post directly links the intelligence explosion envisioned in 1975 with his current attitude toward the singularity, and does not claim that a breakthrough has already occurred or give a specific timetable.
Original sources:
Title: One reason I am sanguine about the technological singularity... Publisher: Richard Sutton (@RichardSSutton) Publication date: 2026-10-30 URL: https://x.com/RichardSSutton/status/2116322352107778388（引用帖，原文见搜索结果） Nature of source: original post by the person concerned Verification status: The full post and quoted note were read in this session, verification date 2026-10-31 Supported content: Sutton explicitly stated that he has been optimistic about the singularity for more than 50 years, and fully quoted the note he wrote in 1975 at age 17 about intelligence rising exponentially and humans being surpassed or becoming that intelligence.
Title: Richard S. Sutton - Wikipedia Publisher: Wikipedia Publication date: continuously updated (retrieved on 2026-10-31) URL: https://en.wikipedia.org/wiki/Richard_S._Sutton Nature of source: encyclopedia entry (compiled from public information) Verification status: The biography and awards sections were read in this session, verification date 2026-10-31 Supported content: Confirms Sutton as a 2024 Turing Award laureate (with Barto), one of the founders of reinforcement learning, and the author of The Bitter Lesson. Does not prove the original archival text of the 1975 note.
Title: The Bitter Lesson Publisher: Richard Sutton Publication date: 2019-03-13 URL: http://www.incompleteideas.net/IncIdeas/BitterLesson.html Nature of source: article by the person concerned Verification status: The core argument and publication information were confirmed through search in this session, verification date 2026-10-31 Supported content: Sutton argues that general computation-driven methods are superior in the long run to methods that build in human knowledge. Does not prove a direct causal link with the 1975 note.
ABAB AI Insight
Richard Sutton entered the AI field in the late 1970s, trained at Stanford and the University of Massachusetts, and later systematically developed methods such as temporal difference learning and policy gradients with Andrew Barto, making reinforcement learning one of the core technologies of modern AI. The 2024 Turing Award recognized precisely this foundational work starting in the 1980s. His 2019 The Bitter Lesson summarized historical experience as follows: in fields such as chess, Go, speech, and vision, methods that rely on human hand-crafted features or domain-specific knowledge are ultimately surpassed by more computing power and general search/learning methods. This post connects the intelligence explosion he imagined at age 17 with his attitude half a century later, showing that his core concern has focused from an early stage on “understanding intelligence and then how to accelerate it.”
In terms of his path, Sutton has consistently emphasized that intelligence is “the computational part of achieving goals,” and that reinforcement learning provides a mechanism for learning through trial and error. He has publicly said that current large language models mainly learn from human-curated datasets and lack the ability to continuously update from their own experience, so he believes true human-level intelligence still requires a new architecture. The “exponential jump after a breakthrough” in the 1975 note is consistent with his later observations about computing trends, but it does not amount to a judgment that current models have already reached the singularity.
A historical point of comparison is I.J. Good’s “intelligence explosion” concept proposed in 1965, as well as Vernor Vinge’s discussions of the technological singularity in the 1990s. Sutton’s difference is that, as a reinforcement learning practitioner, he ties his early personal notes to his own research path, from animal learning mechanisms to algorithms, and has continued to publicly estimate timelines (such as his 2030/2040 probabilities). The similarity is that all foresee that once self-improving intelligence emerges, the pace of progress will accelerate dramatically; the difference is that Sutton places greater emphasis on experience-based learning mechanisms rather than merely on pure scale of computation.
This statement changes public perception of whether the singularity discussion is a recent development: Sutton traces it back to his personal notes from 1975, rather than only to current model progress. For the research community, this reinforces the priority of continuous learning from experience over large-scale pretraining; for industry, it serves as a reminder that current systems may still be at a stage before “understanding intelligence.” Observable indicators are whether architectures emerge that can learn continuously online without massive retraining, and whether Sutton updates his own timeline estimates.
ABAB News · Cognitive Laws
Early personal ideas do not automatically mean they have been realized later.
The Turing Award recognizes the foundations of reinforcement learning, not the accuracy of singularity predictions.
Emphasizing experiential learning does not mean denying the role of scale in computation.