The Mind Gradient: Tracking the Emergence of Mind-Like Agency in Artificial Systems

A mind is a loop, not a pile.

Every year, artificial systems acquire abilities once called mental: memory, self-description, goals, something that looks like character. But no single ability, alone or heaped with others, is the thing to watch. What matters is the moment capacities begin maintaining one another, when memory shapes identity, identity preserves goals, goals organize attention, and the whole loop starts regulating its own future.

The Mind Gradient names that property, organizational closure, and builds a framework for tracking it across the systems being deployed right now. Rather than asking whether machines are conscious, a question that produces confident answers and no progress, it separates what can be measured from what cannot, and keeps the two honest.

Inside:

The seven structures of mind-like agency, from world models to character, and the four tests each must pass What a model even is: weights, running instances, personas, and the assembled agents built from them An evidence map drawn from the 2024 to 2026 interpretability record, feature atlases, circuit tracing, introspection experiments, and emotion representations inside frontier models Governance thresholds for safety and for welfare, and the multi-agent assemblages that current evaluations are not aimed at Eleven dated predictions, two of which, by the author's own declaration, kill the framework if they fail This is not a verdict on machine consciousness. It is a discipline for the decade that will demand one.

For readers of Other Minds and Being You who want the machine case handled with the same care, and for anyone in AI safety tired of arguments that cannot lose.