Why AI can't out-engineer evolution: biology's reductionist tools narrow what models can learn

NikoMcCarty · x · 2026-11-21

Niko McCarty argues bioengineering underperforms evolution because our reductionist tools give a narrow view: we sequence DNA while ignoring the 3D structures that determine expression. Evolution exploits structure, chemistry and size simultaneously, even baking dual functions into one molecule — lactate dehydrogenase B becomes a crystallin lens protein in duck eyes; hepatitis D's RNA is both genome and catalytic ribozyme. He suspects even the best AI models can't replicate evolution's optimization as long as the data we feed them inherits our narrow measurements. Closes with François Jacob's 1977 'Evolution and Tinkering' essay: natural selection works like a tinkerer, not an engineer.

Original post →

More from AGI Musings

AGI Musings channel →