INTELLIGENCE
What a system can do: solve problems, pursue goals, learn, plan, predict, or reason.
Capability alone does not tell us whether anything is being experienced.FIELD GUIDE · AI, MINDS & CONSCIOUSNESS
CAN A MACHINE FEEL?
Chatbots can sound alive. Nobody has a consciousness meter. Here’s the real fight over computation, biology, and whether experience can exist in silicon.
UPDATED 2026-09-20
FIRST: FOUR THINGS PEOPLE KEEP MIXING UP
CONSCIOUSNESS
Whether there is something it feels like to be the system at all.
This is the part nobody currently knows how to detect with certainty in an AI.SELF-AWARENESS
Representing, monitoring, or talking about oneself.
A machine can produce accurate self-reports without that proving subjective experience.UNDERSTANDING
Whether successful symbol use amounts to grasping meaning rather than merely producing the right outputs.
Searle's Chinese Room made this a central AI argument decades before chatbots.THE CENTRAL ARGUMENT
DOES THE STUFF MATTER?
The cleanest split is not “AI believers versus skeptics.” It is whether the right causal organization is enough, or whether consciousness depends on properties of living physical systems that ordinary computation leaves out.
THE ORGANIZATION MATTERS
COMPUTATIONAL FUNCTIONALISM
Mental states are identified by the causal or functional roles they play. On a strong computational version, the right organization could in principle support consciousness in more than one physical substrate.
Same song, different instrument.
THE STUFF MAY MATTER
BIOLOGICAL NATURALISM
Consciousness may depend on properties of living, self-maintaining organisms that are not preserved by ordinary digital computation alone. Seth argues current AI is unlikely to be conscious on this view, while leaving open more brain-like or life-like artificial systems.
A simulation of a stomach does not digest lunch.
OTHER THEORIES YOU WILL KEEP SEEING
These are not four mutually exclusive political parties. They operate at different levels and often overlap. The point is to know what claim each one is making.
GLOBAL NEURONAL WORKSPACE
A representation becomes consciously accessible when recurrent brain activity amplifies it and makes it globally available to many specialized systems.
WHY AI CARES: If global broadcasting is close to sufficient, researchers can ask whether artificial architectures reproduce the relevant causal organization.
INTEGRATED INFORMATION THEORY
IIT starts from features of experience and asks what kind of physical cause-effect structure could realize them as one integrated whole.
WHY AI CARES: It makes substrate and causal structure matter in a different way from ordinary software functionalism, producing unusual predictions about which machines could be conscious.
PREDICTIVE PROCESSING / ACTIVE INFERENCE
Brains continually generate predictions about their sensory states and update or act to reduce mismatch. Seth uses this framework to connect perception, embodiment, regulation, and consciousness.
WHY AI CARES: The live dispute is whether prediction, homeostasis, and self-maintenance point beyond computation or are themselves computational processes.
THE DENNETT MOVE
Maybe the famous mystery is partly created by how we describe it. Dennett argued that science should explain the brain's access, reports, judgments and convictions about consciousness rather than assume an extra private ingredient that those mechanisms cannot touch.
WHY AI CARES: On this approach, better models of cognition and self-report may shrink the gap that other theories treat as a separate hard problem.
2026-09-17
THE BIOLOGY VS. COMPUTATION FIGHT JUST GOT SHARPER
Anil Seth's 2025 Behavioral and Brain Sciences target article received a large open-commentary round in September 2026. The useful part is not a winner. It is that the disagreement now produces much clearer claims about what a conscious machine would need.
SETH
Current digital AI is unlikely to be conscious because computation may not be sufficient. Conscious artificial systems become more plausible as they become more brain-like or life-like.
RICHARDS + AGÜERA Y ARCAS
Prediction, homeostasis, and autopoiesis do not rescue biology from computation. They argue those life-like processes are computational too, so Seth's ingredients may actually point toward conscious AI.
SOLMS
If biological naturalism says only living systems are conscious, then calling a genuinely conscious nonliving system 'artificial consciousness' creates a contradiction. The fight shifts to what counts as alive.
GODFREY-SMITH
The biological emphasis is valuable, but predictive processing may be doing too much work in Seth's version. Biology and prediction should not be treated as one package.
DUNG
Rejecting biological naturalism does not force you into computational functionalism. There are more positions in the space than the headline binary suggests.
WHAT WOULD ACTUALLY MOVE THE ARGUMENT?
- Build artificial systems that are genuinely self-maintaining and embodied, then ask whether consciousness theories make different predictions about them.
- Compare architectures that implement global broadcasting, recurrent processing, integration, world models, self models, and long-running agency instead of treating fluent language as the whole test.
- Look for interventions that produce different predicted changes in consciousness-relevant behavior or internal dynamics under competing theories.
- Keep substrate as an experimental variable. A theory that predicts different outcomes for silicon, neuromorphic hardware, organoids, or synthetic life should eventually face those differences.
- Separate appearing conscious from being conscious. Humanlike conversation is important evidence about behavior, but competing theories disagree on what that behavior establishes.
TAKE THIS TO DINNER: Fluent behavior tells us what a system can do. It still does not tell us, theory-independently, whether there is anything it feels like to be that system.
THE FILES
FOUR PEOPLE TO KNOW
Anil SethControlled hallucination, predictive processing, biological naturalism — and the argument that silicon may not get consciousness for free.OPEN FILE →David ChalmersThe hard problem of consciousness followed him straight into the LLM era.OPEN FILE →John SearleA room that can answer perfect Chinese without understanding a word. Every fluent chatbot walks straight into the argument.OPEN FILE →Daniel DennettConsciousness, evolution, intentionality: a lifetime spent asking how minds can emerge from machinery.OPEN FILE →
RECEIPTS
Primary papers and reference material behind the map above. The field guide summarizes the dispute; these links are where the qualifications live.
- Conscious artificial intelligence and biological naturalismBehavioral and Brain Sciences
- FunctionalismStanford Encyclopedia of Philosophy
- Facing Up to the Problem of ConsciousnessDavid Chalmers
- Conscious Processing and the Global Neuronal Workspace HypothesisNeuron / PMC
- From the Phenomenology to the Mechanisms of Consciousness: Integrated Information Theory 3.0PLOS Computational Biology
- Current Research HighlightsAnil Seth
- Facing up to the hard question of consciousnessPhilosophical Transactions of the Royal Society B / PMC
- Minds, brains, and programsBehavioral and Brain Sciences
- Life is inherently computational, and consciousness probably is tooBehavioral and Brain Sciences
- Real artificial consciousness is, of course, incompatible with biological naturalismBehavioral and Brain Sciences
- Biological naturalism without the primacy of predictionBehavioral and Brain Sciences
- Why I am not a biological naturalistBehavioral and Brain Sciences