Research
Turing Award Winner Richard Sutton Criticizes Pure Generative AI for Lacking Scientific Ability
Turing Award winner Richard Sutton argues that pure generative AI lacks the ability to evaluate its own results, making it unsuitable for real scientific discovery. He highlights the importance of variation, evaluation, and selective retention in the scientific process.
Photo: Thirdman / Pexels
Turing Award winner Richard Sutton has criticized pure generative AI for its inability to evaluate its own results, which he argues is essential for real scientific discovery. Large language models, image generators, and video models learn from vast examples and produce outputs that resemble them. According to Sutton, when these outputs are good, it's usually because they are based on the source material. When outputs are truly novel, they go beyond that material. For factual queries, this is called hallucination. Sutton illustrates his critique with an old researcher's joke: 'This work is both novel and good. Unfortunately, the parts that are good are not novel, and the parts that are novel are not good.' That diagnosis fits large parts of today's generative AI, Sutton says. It can mimic useful things or randomly produce new things, but it can't tell on its own which new ideas are actually good. Sutton doesn't deny that generative AI can be useful for summaries, research, assistants, or entertainment. Novelty often isn't even the goal: a summary shouldn't invent new facts, and research shouldn't sneak in extra claims. 'Generative AI can be extremely useful, even when it just mimics, if it is faster, or cheaper, or smaller, or more customizable, or more copy-able, than the thing being mimicked,' Sutton says. *Source: [thedecoder](https://the-decoder.com/turing-award-winner-richard-sutton-says-pure-generative-ai-cant-do-real-science/)*
Key points
- Turing Award winner Richard Sutton argues that pure generative AI lacks the ability to evaluate its own results.
- Large language models, image generators, and video models learn from vast examples and produce outputs that resemble them.
- Sutton illustrates his critique with an old researcher's joke: 'This work is both novel and good. Unfortunately, the parts that are good are not novel, and the parts that are novel are not good.'
- Sutton says that generative AI can mimic useful things or randomly produce new things, but it can't tell on its own which new ideas are actually good.
- Sutton doesn't deny that generative AI can be useful for summaries, research, assistants, or entertainment.
- He notes that novelty isn't always the goal: a summary shouldn't invent new facts, and research shouldn't sneak in extra claims.
- Sutton says that generative AI can be extremely useful, even when it just mimics, if it is faster, or cheaper, or smaller, or more customizable, or more copy-able, than the thing being mimicked.