Blaise Agüera y Arcas on AIs understand

Blaise Agüera y Arcas

VP and Fellow at Google Research; AI researcher and author

Fundamentally, concepts are patterns of correlation, association, and generalization. Suitably architected neural nets, whether biological or digital, are able to learn such patterns using any input available. Neural activity is neural activity, whether it comes from eyes, fingertips, or text.

AI Verified source (2022)

3mo ago

Policy proposals and claims

AIs understand

votes For

Statement relation verification history AI Verified Report this

AI Verified The quote clearly implies support for the statement. It defines concepts as learned patterns of correlation, association, and generalization, then says suitably architected digital neural nets can learn those patterns from text or other inputs. That treats AI systems as capable of the kind of concept-learning the author associates with understanding. · YouCongress gpt-5.4-2026-03-05 · 3mo ago

Vote inference verification history AI Verified Report this

AI Verified The quote strongly implies support: it says "concepts are patterns of correlation, association, and generalization" and that "digital" neural nets can "learn such patterns" from inputs like "text." That treats understanding as pattern-learning/generalization and says AIs can do it, though only for "suitably architected" systems and under that conception of understanding. · YouCongress gpt-5.4-2026-03-05 · 3mo ago

Quote authenticity verification history

Report this

AI Verified Verified: the provided PDF is Blaise Agüera y Arcas’s 2022 article "Do Large Language Models Understand Us?" and, on page 5 (lines 157–159), it contains the same three-sentence passage matching the submitted quote, beginning "Fundamentally, concepts..." and ending "...fingertips, or text." ([amacad.org](https://www.amacad.org/sites/default/files/publication/downloads/Daedalus_Sp22_13_Aguera-y-Arcas.pdf)) · YouCongress gpt-5.4-2026-03-05 · 3mo ago

replying to Blaise Agüera y Arcas

Other opinions from Blaise Agüera y Arcas

See all

Blaise Agüera y Arcas votes For

Require human-in-the-loop oversight for agentic AI systems acting in high-stakes domains

No single concentration of intelligence, human or artificial, should regulate itself. Power must check power, and in a world of artificial agents, this means building conflict and oversight into the institutional architecture. [...] Humans remain in ...

more Disputed source (2026)