Reality as Simplicity Talk (2009)

The brain doesn't perceive reality — it compresses it, and that compression is reality.

The central claim is disarmingly direct: what we experience as "the real world" is actually the brain's best compressed model of incoming sensory data. Not a copy of reality, not a window onto it — a compact algorithm that predicts what comes next and flags surprises. Ruffini grounds this in Kolmogorov complexity, the idea that the "simplicity" of a dataset is measured by the length of the shortest program that can reproduce it. A brain building models of the world is, in this view, running a continuous compression operation. Reality equals simplicity because the brain only ever has access to its own models, and good models are short ones.

The evolutionary argument follows naturally. Organisms that build compact, predictive models of their environment survive better than those that don't. So natural selection is itself a search for shorter programs — from DNA (a very slow-learning model of environmental regularities) up through homeostatic physiology and into cortical inference. The hierarchy isn't incidental; it's the expected output of evolution applied to inference machines. This also offers a tentative explanation for why physiological signals show scale invariance (fractal-like structure across timescales): hierarchical modeling systems naturally produce it, and its degradation with age might signal modeling breakdown.

The "inferotropic principle" is the most speculative move. Why does the universe appear simple enough to model at all? Ruffini's answer: only universes with simple regularities can sustain inference machines capable of noticing them. It's an anthropic-style argument — we find ourselves in a compressible universe because incompressible ones don't produce observers. The source presents this as a hint rather than a worked proof.

The empirical anchor is the mismatch negativity (MMN), an EEG signal that fires roughly 100–200 ms after an unexpected sound breaks a regular pattern. Ruffini reads this as a low-level neural signature of model violation — the brain had a compressed representation of the auditory sequence, the deviant tone didn't fit, and the error signal is measurable. The deck also sketches applications to virtual reality presence research (how convincingly a VR environment maintains model consistency at multiple levels) and robotics, though these are gestural rather than developed. This is a slide deck from a 2009 talk, so the treatment throughout is programmatic — the ideas are laid out clearly but the formal machinery and experimental detail live elsewhere (the accompanying arXiv paper, 0903.1193).