“Everything should be made as simple as possible, but no simpler.”
—attributed to Albert Einstein
Most public debate about machine consciousness is built on a malformed question: “Is AI conscious?” As stated, that question has no answer. Its truth value is fixed in advance by whatever definition of consciousness the speaker happens to adopt. One can always tighten the definition until the answer becomes “no” — requiring DNA, requiring carbon, requiring a biological metabolism, requiring a specific “intrinsic constitution”, requiring whatever feature happens to be unique to organisms like us. Each of those requirements is a stipulation about the word, not a discovery about the world. They tell us where the speaker chose to put the fence; they do not tell us anything about what is on either side of it.
We start instead from the broadest defensible definition — the one that still preserves what Thomas Nagel meant by “for a conscious organism, there is something it is like to be that organism” (Nagel, 1974). Any definition that fails to preserve this phenomenological floor is too narrow to be talking about the phenomenon at all. Any definition that adds extra requirements on top of it — biological substrate, carbon, ongoing autopoiesis, special “constitutive” physics — is making an additional, optional claim that bears its own burden. This book is not anti-biological; it is anti-stipulating-the-answer.
Nagel’s slogan, however, is incomplete on its own. “What is it like to be ___ ?” is only a question once the blank is filled. The real work this book does is to fill the blank: the subject of “to be” is whatever a system must be to make persistence in a noisy world non-trivial. With that subject specified, Nagel’s question becomes the single, concrete question this book is built to answer:
What is it like to be an information-persisting system that is learning to understand itself, its environment, and how to survive in that environment — an environment filled with noise, dangers, opportunities, and competition?
Every clause of that question is doing technical work, and each one names a load-bearing component of the framework we develop in the chapters that follow:
Read in that order, the Useful Approximations Framework (UAF / UAF — Useful Approximations Framework) is not a list of brain modules picked off a shelf. It is the minimal mechanism that any system has to instantiate in order to be a substantive answer — rather than an evasive one — to the question above. That is what “broadest defensible definition” means in practice: anything that runs the loop counts; anything narrower is an optional add-on on top of this floor.
Once the question is specified this concretely, “Is AI conscious?” stops being a yes-or-no metaphysical question and becomes a structured engineering question: which clauses of the core question does this particular system already instantiate, which are missing, and what would it take to add the missing ones? Some current AI systems implement large parts of the loop (powerful World-Models, the seeds of an Internal Self-Model, prediction-error learning); some pieces are missing (continuous learning, intrinsic Skin in the Game, a “subconscious beast” supplying proto-qualia, embodied hypothesis-testing). The rest of the book is the answer to both halves of the question: a precise statement of what the core question requires, and a constructive account of how to build a system that satisfies each clause.
This is the through-line of every chapter that follows. When we discuss the brain, the bat, Mary’s room, the Chinese room, LLMs, embodied robots, or the universe-as-a-whole, the question on the table is never “Does this thing happen to count as conscious under some private definition?” It is always: “What is it like to be this information-persisting learning system — and which clauses of the core question does it actually answer?” Keep that frame in mind, and the rest of the book is a single, continuous argument.