Descriptions:
Machine Learning Street Talk hosts philosopher and neuroscientist Mazviita Chirimuuta to discuss her 2024 book ‘The Brain Abstracted,’ examining how the scientific practice of simplification shapes — and sometimes distorts — our understanding of cognition and, by extension, artificial intelligence. The conversation surfaces a layer of foundational critique that rarely appears in mainstream AI coverage.
Chirimuuta unpacks two key concepts from philosophy of science: abstraction (deliberately ignoring known details, like frictionless surfaces in Newtonian physics) and idealization (attributing properties known to be false, like infinite populations in genetics modeling). She argues that computational models of the brain carry these same hidden assumptions, and that conflating a model’s tractability with explanatory accuracy is a persistent error running from early cybernetics through modern deep learning. Her ‘protean nature’ view holds that biological and cognitive systems are inexhaustibly complex — they can be ‘pinned down’ to yield true answers, but will keep shifting beyond any single final theory.
For AI practitioners, the implications are concrete: the mechanistic metaphors embedded in how we describe neural networks, attention, and reasoning are not neutral descriptions but theory-laden idealizations that shape which questions get asked and which get ignored. The interview is a rigorous, intellectually demanding conversation best suited to viewers interested in the philosophical foundations of AI, the limits of computational explanation, and what genuine understanding of intelligence would actually require.
📺 Source: Machine Learning Street Talk · Published January 23, 2026
🏷️ Format: Interview
