The main idea of this wonderfully-multidisciplinary and intellectually-stimulating book on consciousness can be summarized as:
Consciousness in humans is analogous to how a user interacts with the software on a computing device: a simple, unified user interface hiding an overwhelming amount of details in the underlying implementation (bits and bytes) in order to allow the user to focus on the goal at hand, such as text editing, form filling, and listening to recorded music. Consciousness is “a lie”, albeit a useful one. It enables the individual who possesses it to achieve actions needed for survival and thriving in the complex world: decision making, planning, and interaction with other individuals.
This book is written in clear, engaging English, covering a wide-range of seemingly-remote but in fact deeply-connected topics. This gives the reader a sense of panoramic intellectual beauty centered at the philosophical question of consciousness: one that spans physics, mathematics, computer science, neurophysiology, psychology, and sociology.
The author starts by discussing the question of Maxwell’s Demon and the 2nd law of thermodynamics. He uses this discussion as a launch point for the claim that information by itself is not useful or valuable. Instead, the selective discarding of information, referred to as “exformation“, is the hallmark of intelligence. Exformation is the hard work done to extract the a small amount of most relevant insight from a pile of raw information. This deeply resonates with modern intellectual workers, who are familiar with, e.g., the fact that the most valuable part of a scientific paper is not always what’s explicitly written down, but includes the failed explorations, debates, and discussions along the way of the research that crystallized in the final choice of methods and the results. In fact, the most valuable bit about the paper might be in the selection of the very topic to attack in the first place, which may be the result of years of painstaking deliberation and trial-and-error (i.e., discarded information), little of which is directly visible in the final output.
Exformation is the true intellectual value – a claim that the author connects convincingly to the field of computational complexity (Gödel’s incompleteness theorem and Turing’s halting problem) and the theory of computational depth (Kolmogorov).
This principle of exformation is embodied in the human brain through the fact that the information bandwidth from the senses (1M bps) far exceeds the estimated bandwidth of high-level, conscious cognition (10-20 bps). Evidence for the latter is abundant, including the famous 7±2 capacity of working memory (Miller 1956) and the known facts regarding typical speaking rate seen in humans (200 words/minute, or approximately 3 words/second). Based on this observation of exformation, the author formulates the human mind has consisting of two-levels: the high-bandwidth, low-level “Me” that is non-conscious, and the low-bandwidth, high-level “I” that is conscious. “I” is based on the input information from “Me”, but only a highly-processed version of it. This processing is best seen from a few astonishing and revealing results by Benjamin Libet’s famous perceptual and motoric studies, ones that show that 1) the conscious awareness of self-initiated simple actions such as flexing of a finger lags behind the onset of the EEG correlate (the readiness potential) by ~300 ms; and 2) the conscious awareness of simple sensory stimuli (e.g., tactile stimulation of the skin) also lags the stimulus onset by about the same long latency, despite the fact that the sensation is antedated (moved backward in time) to roughly correspond to the actual stimulus. Such a long latency is the time it takes to perform this exformation process.
The author’s philosophical musing of the relation between “Me” (the grounded self) and “I” (the illusory self) resonates with thoughts such as the “flow state” and those who pursue high levels of artistic and athletic achievement. In those ideal states, one often feels that “I” retreats to the background, if not disappearing altogether (the Buddhist ideal of “no self”), hence giving “Me” an unhindered chance to perform at its impressive high bandwidth. Think of saying such as “dance like no one is watching”, where “no one” includes the “I”. Yet the reader is reminded that in the modern world, “Me” cannot function without an “I”, because it is “I” that navigates social relations, sets the goals, directs “Me” to overcome difficulties and challenges against its animal nature (e.g., initial fear and laziness).
Despite all these intriguing points throughout this chapters, this book is not the final word on the hard problem of consciousness. The formulation of the user illusion itself hints at its biggest weak point. The user-illusion analogy pre-supposes that there is an entity (i.e., the user) as the starting point. In the analogy, the user (human operator of a computing device) exists before the device and the software are made and presented to the user. But in the human brain, there is no such pre-existing entity. So where does this entity that corresponds to the user in the analogy, one that “experiences” the result of the computer’s exformation process, come from? This entity must be bootstrapped into existence by the neurophysiological substrates of consciousness, as if the computer bootstraps a user! Thus following the user-illusion analogy leads to this nonsensical conclusion.
The reader can hardly blame the author for this gaping hole, because it is rooted in the deepest unanswered metaphysical question about the nature of subjective experiences and the self, one that rings throughout the history of philosophy in the form of Cartesian dualism, theological thoughts, and panpsychism. The materialistic and monistic view in this User Illusion book provides experimental data points that a final solution to this problem must fit.
Before ending this note, let me write down a few thoughts about the question of consciousness in modern AI models (e.g., large language models or LLMs). The User Illusion book does not provide much useful insight to the question of whether an LLM is or can be conscious. The exformation process may truly be a foundation for consciousness in humans. But there is no reason to believe that any kind of consciousness or self-awareness must be built on exformation. LLMs are fortuitously similar to the low bandwidth of human language production, but that is just a trivial consequence of their textual input and output modality. This analogy breaks down as soon as we move from LLMs to large multi-modal models such as those with image and video output. There are few key questions surrounding LLM consciousness, such as the nature of symbolic reference (meaning) and self-reference. Questions like those received a deeper treatment in D. Hofstadter’s books (e.g., GEB and I Am a Strange Loop) than in this book. But none of this is to take away the quality and depth of The User Illusion. For example, this book reminded me of two interesting and under-explored questions regarding LLMs: 1) the computational depth of LLMs and how it relates to their mathematical formulation and 2) pre-training followed by auto-regressive sampling as a form of exformation. In this regard this book can serve as a useful source of inspiration and guide.
caisq 
