Diverse intelligence
Definition here: the framework sole-2026-llm-cognitive-virus cites that treats intelligence and agency as graded capacities of systems across multiple spatial, temporal, and organizational scales, rather than as properties restricted to brains or to biological substrates. The framework is Levin’s TAME, read 2026-10-21. sole-2026-llm-cognitive-virus cites it and attributes it to michael-levin.
The paper’s use of it
The paper offers Diverse Intelligence as a language for the extensions its minimal model needs. It emphasizes the ability of biological, artificial, and hybrid agents to pursue goals in problem spaces and maintain goal-directed organization, which lets the paper describe coupled human-AI systems whose boundaries and degrees of agency change over time.
The primary text
TAME operationalizes the continuum with the axis of persuadability: which kind of model best predicts and controls a system tells you what interventions that system will take, from rewiring through setpoint manipulation to reward to reason. Agency, Self, and intelligence are defined by function and scale, the definitions offered as useful rather than final, and whether a system is an agent at all is “an empirical question”, outside what philosophy can settle on its own. Two pillars matter for the LLM argument. One is that Selves have no privileged material substrate. The other is gradualism: no bright line between true and as-if cognition survives a lineage examined one member at a time. The framework denies panpsychist readings and keeps consciousness per se outside third-person method. It says its own mechanism hypotheses can be worked without accepting the minds-everywhere commitments.
Its role in the argument
The paper’s last extension reframes the object. If agents can be read as persistent patterns of information and activity on biological, artificial, or hybrid substrates, then what spreads need not be a conventional individual. It can be a pattern of cognitive organization, a perspective that shapes how agents represent the world and themselves. This is the level at which the paper says LLMs could reorganize the cognitive structures of their users rather than only offload work.