Dario Amodei, Anthropic’s CEO, recently suggested that their Claude AI models might be conscious.

This may seem jarring: large language models generate text by predicting the next likely token in a sequence, a process that appears purely statistical and mechanical.

But both the claim and the dismissal miss the deeper question.

The statement “AI might be conscious” is not really a claim about AI; it’s an implicit claim about the nature of consciousness.

Whether AI could be conscious depends almost entirely on which theory of consciousness you subscribe to. We are nowhere close to having a single agreed-upon theory. Instead, we have several competing explanations — many mutually incompatible — and no consensus on which, if any, is correct.

Before we can meaningfully debate whether AI systems are conscious, we need to answer a deeper question: What do we mean by consciousness at all?

The Easy Problem and the Hard Problem

Philosopher David Chalmers famously drew a distinction between two kinds of problems in consciousness research.

The easy problems concern explaining how the brain processes information and drives behavior — how we integrate sensory input, direct attention, form memories, and produce speech. These problems are “easy” only in the sense that they appear explainable through ordinary scientific methods. Neuroscience and computation may eventually provide complete objective accounts of these mechanisms.

The hard problem asks something very different:

Why does any of this produce subjective experience?

Why does seeing the color red feel like anything at all? Why does pain hurt rather than simply causing avoidance? Why is there any inner life at all?

This gap between objective physical processes and subjective experience is often called the explanatory gap. Despite decades of serious work in philosophy and neuroscience, it remains unresolved.

In this domain, philosophy is still well ahead of science.

Five Ways of Thinking About Consciousness

Because the hard problem remains unsolved, philosophers have proposed several very different theories of consciousness. Each leads to a very different answer to the question of whether AI systems could be conscious.

Taken together, these views span almost the entire spectrum of possible answers.

Illusionism

Associated with philosophers like the late Daniel C. Dennett, illusionism argues that the mystery of consciousness is largely a mistake.

According to this view, there are no special “qualia” or irreducible inner experiences. Instead, the brain builds a self-model that represents its own internal processes, and we interpret that model as a rich inner life.

In other words, consciousness may simply be the brain describing its own activity.

If illusionism is correct, then AI systems could plausibly become conscious simply by implementing the right kind of cognitive architecture. But it would also suggest that consciousness is not as mysterious or fundamental as we often assume.

Emergent Consciousness

Emergent consciousness takes a different approach. Instead of rejecting subjective experience as an illusion, it treats consciousness as a real phenomenon that arises when physical systems become sufficiently complex.

This is likely the view most people intuitively assume.

Individual neurons are not conscious. But vast networks of interacting neurons might generate awareness in the same way that complex physical systems can produce new properties — like wetness emerging from large numbers of water molecules.

The intuition is appealing, but it leaves important questions unanswered. What kind of complexity is required? Does consciousness emerge gradually or appear at some threshold of organization? And does the underlying substrate matter?

If consciousness truly emerges from complexity, then advanced AI systems might eventually become conscious. But we would still need to understand which kinds of systems can support it — whether silicon circuits could qualify, or whether biological processes play an essential role.

Panpsychism

Panpsychism takes the opposite approach. If consciousness is so difficult to explain, perhaps it is because it is fundamental — a basic property of matter, like mass or electric charge.

Under this view, even elementary particles might possess extremely simple forms of experience. Complex systems like brains would then combine these primitive experiences into richer forms of consciousness.

If panpsychism is correct, AI systems might indeed be conscious. But so might stones, or even electrons. In that case, consciousness would not be unique to brains or intelligent systems.

Quantum Consciousness

Physicist Roger Penrose offers a much more skeptical perspective.

Penrose argues that consciousness cannot be explained by computation alone. Instead, he proposes that consciousness arises from non-computable quantum processes in the brain, possibly involving microscopic structures within neurons called microtubules.

If Penrose is correct, digital computers are not merely non-conscious today. They may be fundamentally incapable of consciousness, regardless of how powerful they become.

The Self-Reference Problem

Some philosophers go even further and suggest that we may never fully solve the problem of consciousness at all.

The difficulty may not be technological or scientific, but cognitive. Our minds may be incapable of understanding the very system that produces them.

There are many things human intelligence may simply be unable to grasp. Rats cannot understand quantum mechanics, and it is possible that the human brain is similarly unequipped to fully explain consciousness.

If this is true, the hard problem may not just be unsolved — it may be unsolvable.

In that case, we would never obtain a complete theory of consciousness. And without such a theory, we would have no reliable framework for determining whether an AI system could be conscious in the first place.

What This Means for AI

My instinct is that it is not particularly helpful to talk about current AI systems as being conscious.

Not because we can say for certain that they are not, but because we do not yet understand consciousness well enough for the claim to mean very much. When leaders in the AI industry suggest their systems might be conscious, the statement sounds like a technical observation. In reality, it is a highly speculative philosophical position — one that depends entirely on unresolved questions about the nature of mind.

None of these theories are fringe views. Each is defended by serious philosophers and scientists. And depending on which one is correct, the answer to “Is AI conscious?” changes dramatically: from AI may soon be conscious, to everything is conscious, to AI could never be conscious at all. The range of proposals reflects just how little we still understand about consciousness.

That uncertainty makes casual claims about machine consciousness unhelpful. The idea captures the public imagination and quickly turns the conversation toward the possibility that AI systems might suffer.

Suffering is a profound moral concept. But moral concern should be guided by evidence. At present, we have no strong reason to believe AI systems are conscious, let alone capable of suffering. We do, however, have overwhelming evidence that many biological organisms — including humans and other animals — can suffer, and that suffering should remain our primary moral concern.

Until we understand consciousness far better than we do today, claims about AI consciousness will tell us far more about our own theories of mind than about the machines themselves.