An update on
* Distributed AI-Assisted Cognition,
wiki.p2pfoundation.net/Distr…
ISPCR:
"The literature on AI-assisted cognition has, until recently, focused on the dyad: one human, one model. The questions asked have concerned cognitive offloading, automation bias, the calibration of trust, and the effect of generative outputs on the human's own reasoning. This is the right starting point for individual-scale work. It is the wrong frame for what is happening in small innovation consortia.
What we observe instead is a configuration in which multiple AI systems are invoked deliberately for different cognitive functions, and the differences between them are used as a productive resource. One model is preferred for strategic structuring and synthesis; another for sustained long-form drafting; another for adversarial pressure-testing of commercial narratives; another for sharp, compact SWOT generation. The human operator's task is increasingly to choose which model to ask, in what order, with which framing, and how to handle the contradictions that come back. This is not a dyad. It is closer to what one might call a curated polyphony — a small ensemble of cognitive agents with deliberately non-identical reasoning styles, used jointly to do work that none of them, and no human, would do as well alone.
The intellectual ancestry here runs through several traditions: the Delphi method, in which expert disagreement is treated as informative; ensemble methods in machine learning, in which the disagreement between weak learners is the source of strength; and red-team / blueteam practices in security and policy work, in which adversarial cognition is institutionalized. What is new is that the agents in the ensemble are themselves AI systems, that the human's role has shifted from substantive contributor to orchestrator, and that the orchestration itself is a learned skill that the case material visibly improves at over time."
(
orcid.org/my-orcid?orcid=000…)