# How 20% Of Neurons Separate Us From LLM’s | Integrated Information Theory Explained Source: https://www.youtube.com/watch?v=7I0DopbNBM0 Recap page: https://rapidrecap.app/video/7I0DopbNBM0 Generated: 2026-02-13T20:36:02.985+00:00 --- ## Quick Overview Integrated Information Theory (IIT) posits that consciousness arises from a system's degree of intrinsic integration, measured by Phi (Φ), distinguishing it from computational performance like that seen in LLMs, which typically lack the necessary high integration structure, as evidenced by the cerebellum containing 80% of neurons but not being essential for conscious experience. **Key Points:** - The cerebellum holds roughly 80% of all brain neurons, yet its severe damage or absence does not fundamentally affect a person's conscious experience, only motor skills. - IIT, developed by Giulio Tononi, measures consciousness via the degree of integration of nodes in a system, quantified as Phi (Φ), focusing on intrinsic causal structure rather than external information processing. - Consciousness requires two properties: high differentiation (informative) and complete integration (unitary experience), which Jeremiah Hendren notes are accounted for by IIT in physical/neural terms. - A proxy measure for IIT, the Perturbational Complexity Index (PCI), uses TMS combined with EEG to measure the brain's 'echo' after a zap, correlating high PCI (high differentiation and integration) with conscious states. - IIT explicitly dissociates intelligence from consciousness, asserting that LLMs, structured typically in feed-forward processing, have low integration and are therefore not conscious, regardless of their functional intelligence. - IIT starts from the axiom that experience exists immediately ('my experience exists'), contrasting with standard approaches that begin from the external physical world, thus avoiding the hard problem of consciousness. - The substrate of consciousness in IIT is predicted to be a region mostly of posterior cortex, termed the 'posterior hot zone,' which is wired in a fully integrated system, unlike the more modular, feed-forward wiring of the cerebellum. **Context:** The discussion features Jeremiah Hendren, a PhD candidate researching better communication methods for Integrated Information Theory (IIT), and an interviewer, taking place at the Asencia Foundation's workshop 'beyond physicalism' in Angleberg, Switzerland. IIT is presented as a promising theory of consciousness that differs from computational functionalism by prioritizing the intrinsic integration of a system's nodes (measured by Phi) over its external information processing capabilities, aiming to provide a scientifically validated test for consciousness, potentially distinguishing conscious entities like humans from advanced LLMs. ## Detailed Analysis Integrated Information Theory (IIT) offers a formal mathematical approach to consciousness based on introspection, identifying experience as both highly differentiated and completely integrated. Jeremiah Hendren explains that IIT seeks physical properties reflecting these experiences using causal terms, leading to proxy measures like the Perturbational Complexity Index (PCI), which measures brain reverberations post-stimulation; a high PCI, calculated using concepts similar to file compression ratios to assess differentiation and integration, correlates strongly with conscious states versus unconscious states like deep sleep or anesthesia. A crucial distinction IIT makes is between intelligence and consciousness; while LLMs exhibit high functional intelligence due to their feed-forward architecture, they lack the necessary intrinsic integration (low Phi) to be conscious, contrasting sharply with the brain's posterior cortex, the predicted 'posterior hot zone' of consciousness, which is fully integrated. This structural difference is highlighted by the cerebellum, which contains 80% of neurons but is wired modularly like a computer, and whose damage does not impair consciousness. IIT firmly grounds its science by starting from the immediate existence of experience ('experience exists'), arguing that starting from the external physical world inevitably leaves the hard problem of consciousness unsolved, as correlation between neural firing (e.g., seeing red) and subjective qualia remains an explanatory gap for materialist approaches. ### IIT Core Principles - Consciousness requires high differentiation and complete integration (unitary experience) - Integration means modalities feel unified, even if delayed (latency demonstrates integration) - The theory starts scientifically from the direct existence of experience, avoiding the hard problem. ### Measuring Consciousness (Φ and PCI) - Integrated information is quantified as Phi (Φ) based on causal constraints between nodes in toy systems - PCI uses TMS/EEG to measure the 'echo' complexity, which tracks well with conscious reports in healthy subjects - A threshold PCI value separates likely conscious from non-conscious states. ### LLMs vs. Brain Structure - LLMs are functionalist, performing tasks (doing), but IIT focuses on being; their feed-forward wiring results in low intrinsic integration (low Φ) - The cerebellum, with its modular, feed-forward wiring, parallels computer architecture and holds 80% of neurons but is not essential for consciousness. ### The Posterior Hot Zone - The substrate of consciousness in IIT is predicted to be the posterior cortex, wired as a fully integrated system - Damage to this area severely impacts consciousness, unlike damage to the cerebellum. ### Intelligence vs. Consciousness - IIT predicts intelligence and consciousness are totally dissociable; a system can be hyper-intelligent (like an LLM) but lack consciousness, or vice versa - Evolutionary pressure likely favors integrated systems, causing intelligence and consciousness to usually co-occur in nature. ### Addressing Explanatory Gaps - IIT aims to answer why consciousness is present/absent and why specific experiences feel the way they do (qualia) - Simply correlating neural firing patterns with experience (identity theory) does not scratch the explanatory itch for IIT proponents.