Jacek Hoffmann Proposal
Jacek Hoffmann proposed building systems from multiple AI agents with different data and verification methods. On August 20, Hoffmann published an essay on "heterogeneous cognitive ecology": humans and AI approach a task with different methods and data, then cross-check results. He calls the scenario where multiple agents share one mistake "Beryl Cage".
Hoffmann assumes a possible asymmetry: AI systems will increasingly create and transform information, while an individual will find it harder to reconstruct the solution path. He sees a sample verification device in science: one participant proposes an explanation, another looks for a counterexample, and a third repeats the analysis with different data or methods. This makes the error more noticeable where the results diverge.
Diverging data, methods, and criteria create different reasoning paths and different errors. The divergence can reveal a hidden assumption or condition that the first participant missed. Ten copies of one system can agree because they err in the same way. "The number of models is not equal to diversity," Hoffmann writes, citing an official report from Anthropic, August 13.
🔗 Read original →
Substack
Prisoners of the Stone Age in a World of Exabytes
From the Biological Limits of Mind to a Heterogeneous Cognitive Ecology
August 23, 2026 4
