i3AI

i3AI - the Infinite Improbability Institute for Artificial Intelligence - is the umbrella for my independent research into the substrate of biological intelligence and consciousness. What began toward a PhD thesis in 1999 I revisited during a three-year sabbatical (2017-2020) and have continued since, working toward a biologically plausible neural-network framework for AGI. The neural-network engineering thread is quieter while I focus on agentic AI, but the consciousness research is active again, and the computational-neuroscience learnings continue to guide the agentic work.

The research lives in two main threads, captured as sub-projects below: the AGI experiments sandbox, and the Homeostasis Theory of consciousness. A public site (built on Sitepress/Rails) hosts the long-form manuscripts, the 2018 poster, and supporting essays.

The consciousness thread is where the recent work sits, and the public site is now organized around a single thesis: the capability–sentience gap. Frontier models are climbing the functional axis of mind — reasoning, agency, and increasingly a self they can describe — but not the phenomenal axis of feeling, because feeling requires a homeostatic stake: something genuinely at risk for the system itself. The field is solving intelligence, and scaling alone does not put it on course to solve feeling. The argument leans on a frontier lab drawing the same line — Anthropic's A Global Workspace in Language Models (2026) reports functional-access evidence for a reportable, causally central internal workspace while explicitly declining to claim the model can feel anything.

The central manuscript has been reworked into an illusionist homeostatic theory of valence — an account of why pain is bad — positioned as the "stakes layer" of experience that complements global-workspace and integrated-information theories rather than competing with them. The mechanism is credited as a synthesis of perceptual control, homeostatic reinforcement learning, and active inference; the novel contributions are the illusionist framing, a falsifiable behavioral signature (a state-modulated valence signal), and a two-ingredient criterion for machine sentience — an intrinsic viability stake plus a self-model that is its own implementation — with a small open-data reanalysis as a first empirical probe. Three research threads carry it: why (the Homeostasis Theory of Cognition and Consciousness), what (cognitive homeostatic agents), and how (the Pattern Machine substrate).