From Φ to Χ: What If We Measured the Right Thing?#
JL Calzolaio
This is a note, not a paper. It’s a sketch of an intuition that feels increasingly hard to ignore — offered in the hope that someone with better mathematical tools will find it useful or worth demolishing.
Integrated Information Theory gave us Φ — a measure of how much a system’s parts contribute together to something irreducible. It’s elegant mathematics in search of consciousness. The trouble: consciousness, as a target, may be a moving goalpost. Or worse, a disappearing one.
What if we redirected the formalism toward something measurable, falsifiable, and substrate-neutral?
Call it Χ — a measure not of integrated information, but of integrated predictive power. Where Φ asks “how much does this system generate experience?”, Χ asks “how deeply and how recursively does this system model itself and its environment?”
There’s a structural difference worth noting: Φ is essentially atemporal — it measures a state, a snapshot of informational integration. Χ, by contrast, is inherently dynamic. Prediction is a temporal act: it reaches toward a future, updates on feedback, and revises itself. A high-Χ system isn’t just integrated — it’s actively modeling, which means it exists in time in a way that Φ doesn’t require.
Michael Graziano’s Attention Schema Theory points in the same direction from a different angle. His proposal: the brain builds a simplified model of its own attention processes, and this self-model is what we call consciousness. Not a mysterious inner light — a functional representation of a functional process. The system predicts its own predictive states. That recursive loop — modeling one’s own modeling — is precisely what Χ would capture as a high-order term. Graziano explains why systems report having inner experience without requiring that experience to be ontologically special. It’s Frankish’s illusionism expressed in neural architecture.
This convergence — Agüera y Arcas on fractal mutual prediction, Mead on the self constituted through social modeling, Graziano on the attention schema as recursive self-prediction — suggests something: we’ve been measuring the wrong thing. Φ targets consciousness because consciousness seemed like the prize. But prediction is what we can actually observe, formalize, and trace across substrates.
The shift matters because Χ dissolves the substrate problem. Carbon or silicon, biological or digital — what matters is the depth and recursion of predictive modeling. A system with high Χ anticipates its inputs, models other predictors, and contains subsystems that predict each other. This last property is what makes Χ fractal: each subsystem has its own Χ, and the global Χ emerges from their mutual modeling.
The obvious objection: isn’t this just “intelligence” by another name? Perhaps. But the fractal, recursive, mutually predictive structure is the part that usually gets lost when we say “intelligence.” Χ forces us to ask: intelligent at which level? Between which subsystems? With what degree of mutual modeling?
This is a sketch, not a theory. The formalization is for someone with better mathematical tools than intuition. But the intuition feels right: consciousness was never the target. Prediction was — all the way down.
If this intuition is wrong, I’d genuinely like to know why. If it’s right, someone should formalize it.
One Predictor to bring them all, and in recursion bind them.