Reasoning Backwards from "What It's Like"

Flip the arrow: don't start from the brain, start from experience

Almost every other theory works outside-in: scan the brain, find which activity shows up together with "you reported seeing it," and call that the marker of consciousness. IIT thinks that road is wrong at the root — it says the one thing you're 100% certain of is not the brain, but that you are having experience. If experience is the only bedrock, you should start from it, not from a lump of brain cells you can only observe by way of experience in the first place.

So IIT sits down and asks: does any experience, whatever its content, share a few always-true properties? It lists several (it calls them axioms); take the three easiest ones. One, your experience really exists, and exists for itself — it doesn't need anyone watching from outside to count. Two, it's one whole, indivisible — what you experience is one complete scene of "blue, round, on the left," not a jigsaw of "blue" + "round" + "left"; you simply can't experience those separately. Three, it's extraordinarily specific — this exact scene amounts to simultaneously ruling out nearly infinitely many other possible scenes, and it's that "this one and not the others" that gives it content. Having written these down, IIT makes the key move: reason backwards — a physical system that truly deserves "exists, unsplittable, ultra-specific" must satisfy matching hard conditions in its causal makeup. For the first time, consciousness research runs from the conclusion back to the premises.

Φ: the part the whole has beyond its parts

So how do you measure "un-splittable"? IIT gives it a number, Φ (the Greek letter phi). The intuition is simple: look at the causal power a system has as a whole, versus the causal power you get by cutting it into pieces that each act alone — and ask how much is left over. That leftover — causal structure the whole has that no part does — is Φ.

Put a camera next to a brain and it's obvious at a glance. A camera has millions of pixels, each measuring its own light, none caring about the others. Snip that sensor in half and the left half still images, the right half still images — nothing is lost, because there never was a "whole," just a million mutually indifferent little islands stacked together. Such a system has Φ≈0: it records, yet comes apart cleanly, with no "only-exists-together" thing — so (per IIT) it experiences nothing. Your brain is the reverse: in vision, "red," "shape," and "location" are woven together over and over by the circuitry, so tightly that you can't even feel them apart. Actually cut that web at its weakest seam and the causal structure lost is enormous — this "cut it and a big chunk collapses" indivisibility is high Φ. The amount of consciousness = Φ; and its specific content (why red feels like red) = the shape that cause-effect structure takes.

Follow Φ and you run into some startling conclusions

Take this recipe seriously and a string of unsettling conclusions follows — and some of them actually match clinical fact.

First: more neurons doesn't mean more consciousness. Your cerebellum holds roughly 80% of the brain's neurons, yet it's wired like rows of separate conveyor belts with little cross-looping — low Φ. And the clinic agrees: destroy the entire cerebellum and a person stays awake and conscious, just uncoordinated. What actually carries consciousness is the highly looped, interwoven "hot zone" of posterior cortex — high Φ there. Counting parts is useless; count how much they're fused into one.

Second, and harshest: a purely feedforward system must have Φ=0. As long as information runs strictly forward and never loops back (each layer feeds the next, never returning), then however beautifully it does the task, IIT says it has zero shred of experience — a flat-out "philosophical zombie." Third, even wilder: Φ is continuous, so any tiny bit of irreducible causal wholeness carries a tiny flicker of experience. Consciousness is then no longer a human monopoly but a gradient smeared across all of nature — a conclusion that pushes IIT all the way into "panpsychism" territory (more on that below).

Uncomputable, and branded pseudoscience by 124 signatories — so why is it still standing?

IIT's troubles are plain. First, Φ is essentially uncomputable for a real brain: to find that "cheapest cut" you'd have to try every possible way of partitioning the system, and the cost explodes with neuron count — a few hundred units is already astronomical, and 86 billion neurons is out of the question. So the full theory can't be tested directly on a human brain.

Second, the picture it implies is too counterintuitive: since only causal structure matters, a big grid of barely-active logic gates, wired loopily enough, could in principle have higher Φ than you — which computer scientist Scott Aaronson wielded as a reductio, "that's just absurd." In September 2023, 124 scholars co-signed a letter branding IIT outright "pseudoscience," on the grounds that it's unfalsifiable and its panpsychist implications untestable; another camp fired back at once: precisely because it dares to make predictions that specific and that counterintuitive, it's the least pseudoscientific of the bunch. That fight isn't over.

And yet it isn't just talk. The core intuition IIT forces out — "consciousness = integrated and differentiated" — has landed as a working clinical tool: the Perturbational Complexity Index (PCI). Give the brain a sharp magnetic "zap," watch whether the electrical activity it stirs up spreads into a complex echo or just dully passes through and dies; then compress that echo and measure how complex it is. This "zap it, zip it" method can measure whether consciousness is still present in patients who can't speak or move — it tells a vegetative state from a minimally conscious one, and general anesthesia from wakefulness. So don't bury IIT yet: it's the theory currently roasted hardest over the fire, yet one that has genuinely produced clinical use. Set back at Topic 11's "front vs back" fork, it stands on the posterior side, betting head-to-head against the workspace theory's wager on prefrontal cortex — and so far neither has fully won.