Solely measuring skill at any given task falls short of measuring intelligence
5 November 2019 (arXiv). François Chollet, a Google AI researcher and creator of the Keras library, posted this paper on arXiv in November 2019. It also introduced the ARC benchmark, the Abstraction and Reasoning Corpus.
On the Measure of Intelligence
More from François Chollet · AI and science · AI and hype · Before ChatGPT
Who else?
What it refers to: The AI community's practice of benchmarking intelligence by comparing the skill of AIs and humans at specific tasks, such as board games and video games.
Other voices
- “Intelligence is largely defined by experience and training” Jeff Hawkins,
- “Surpasses a human expert on three of them” Mnih et al.,
- “Sparks of artificial intelligence” Microsoft Research paper, 2023,
- “Given no domain knowledge except the game rules” Silver et al.,
- “I challenge the claim that next-token prediction cannot surpass human performance” Ilya Sutskever, Builder,
- “Watson is performing at human expert levels” Ferrucci et al.,










