Can AI Be Conscious?

The hard problem of consciousness

Before we can ask whether AI is conscious, we need to know what consciousness is. Most discussions of AI consciousness collapse at this first step. They equivocate between three things that are easy to confuse:

Information processing. Taking inputs, transforming them, producing outputs. AI does this constantly. So do thermostats. This is the easy bit; nobody disputes that AI processes information.

Functional consciousness. The ability to integrate information, deploy attention, plan, model the self, and respond flexibly to situations. AI does some of this; humans do more of it. Whether AI does enough of it to count as "conscious" by this definition depends on where you set the threshold. This is the kind of "consciousness" that philosophers like Daniel Dennett tend to focus on.

Phenomenal consciousness. The fact that there is something it is like to be the system. The redness of red as it appears to you. The felt quality of pain. The sense of being present in a moment. This is what philosophers call qualia or subjective experience. Whether AI has any of this is the hard question.

The philosopher David Chalmers gave the canonical statement of the difficulty in 1995. He distinguished the "easy problems" of consciousness (how the brain integrates information, attends, reports its own states) from the "hard problem": why all this processing is accompanied by subjective experience at all. Chalmers's claim is not that the easy problems are actually easy; they are extremely hard scientifically. His claim is that even when we solve them all, we will still face the puzzle of why processing is accompanied by experience rather than happening "in the dark." The hard problem is not yet another scientific question; it is the question of why scientific explanation, as ordinarily understood, leaves something out.

Different philosophers have different reactions to this argument. Some accept it and conclude that consciousness must involve something fundamental that we do not currently understand. Some reject it, arguing that phenomenal consciousness will turn out to be identical to certain kinds of functional consciousness once we understand the brain well enough. Some hold that the hard problem reflects a confusion about what we are asking. The dispute has not been resolved in thirty years of work, and there is no consensus answer in sight.

The relevance for AI: whether AI can be conscious depends entirely on which of these positions is right. If consciousness is essentially functional, advanced AI might have it already, or soon. If consciousness involves something more, advanced AI might never have it, no matter how sophisticated its outputs. The AI question rides on the philosophical question, and the philosophical question is genuinely open.

The leading scientific theories

While philosophers have argued about the hard problem, neuroscientists and cognitive scientists have proposed empirical theories of how consciousness arises in brains. The two most prominent (and most relevant to AI) are Integrated Information Theory and Global Workspace Theory. A third, functionalism, is more a philosophical stance than a scientific theory but is essential to understand because it underlies most ambitious claims about AI consciousness.

Integrated Information Theory

Integrated Information Theory (IIT), proposed by neuroscientist Giulio Tononi in 2004 and developed in collaboration with Christof Koch, is one of the most prominent attempts to give consciousness a precise scientific definition. The central concept is a quantity called Phi (Φ): a measure of the degree to which a system integrates information in ways that cannot be decomposed into independent parts. A system with high Phi has rich causal interactions among its components that cannot be reduced to the activities of subsystems acting independently. A system with low or zero Phi is one whose information processing can be decomposed cleanly into independent pieces.

IIT's claim is that consciousness is identical to integrated information. A system with high Phi is conscious; a system with zero Phi is not. The theory makes strong predictions, including some that are uncomfortable for AI claims. Conventional digital computers, including current large language models, have architectures that are highly decomposable. They process information in massively parallel ways that, on IIT's account, generate very low Phi regardless of how sophisticated their outputs become. On this view, current AI cannot be conscious. Not as a matter of present limitation but as a matter of architectural kind.

IIT is controversial. Critics argue that the formal definition of Phi has problems, that Phi is not actually computable for any realistic system, and that the theory's strong predictions (which would assign consciousness to things like simple feedback systems) are counterintuitive. A 2023 adversarial collaboration between IIT and Global Workspace Theory researchers, organized to test the theories against each other empirically, produced results that supported neither theory decisively. Still, IIT is taken seriously enough that it shapes a substantial part of the contemporary AI consciousness debate.

Global Workspace Theory

Global Workspace Theory (GWT), developed by Bernard Baars in the 1980s and developed further by Stanislas Dehaene and others, takes a different approach. On GWT, consciousness is what happens when information is "broadcast" to a global workspace in the brain, making it available to many cognitive processes at once. Subconscious processing happens in many specialized modules; conscious processing is what happens when one of those modules' outputs gets promoted to the global workspace and made available across the system.

GWT is more friendly to AI consciousness than IIT, because the architecture it describes can be implemented in artificial systems. If consciousness is a broadcast mechanism, machines that implement such a mechanism might in principle have it. Some AI researchers have argued that contemporary large language models implement something analogous to a global workspace through their attention mechanisms. Others argue that this is a category mistake: the workspace metaphor in AI does not yet implement the specific kind of integration GWT describes.

GWT's status with respect to phenomenal consciousness is also disputed. Some defenders argue that the global workspace mechanism is sufficient for consciousness in the full philosophical sense. Others, including some who otherwise accept GWT, hold that GWT explains how information becomes globally available but leaves the hard problem (why this is accompanied by experience) untouched.

Functionalism

Functionalism is the philosophical view that mental states are constituted by their functional roles. To be in pain, on functionalism, is to be in a state that does what pain does: responds to tissue damage, motivates withdrawal, generates aversive associations. If something else (silicon, software, anything) does what pain does, then on functionalism it is in pain. Daniel Dennett is among the most prominent defenders of a sophisticated functionalism.

Functionalism is the most AI-friendly position in the field. If functionalism is right, machine consciousness is an engineering problem, not a metaphysical one. Build a machine that performs the functional roles of consciousness, and you have built a conscious machine. Many AI researchers who hold strong views about AI consciousness are functionalists, often without realizing it. They take consciousness to be definable by its functions, treat current AI as already implementing those functions, and conclude that AI is therefore conscious or close to it.

Functionalism has serious problems. The most famous is the "philosophical zombie" thought experiment: imagine a being that is functionally identical to you in every respect but has no subjective experience. If such a zombie is even coherently conceivable, functionalism is incomplete: there is something to consciousness beyond function. Whether zombies are coherently conceivable is itself disputed. Like the hard problem, the question is open.

Where current AI fits

Setting aside the philosophical question for a moment: what is the empirical evidence about current AI systems?

It is genuinely thin. Large language models produce text that sounds like the text of conscious beings, but this is unsurprising: they are trained on text written by conscious beings and optimized to produce more of it. The surface similarity tells us very little about what is happening inside.

A 2023 study assessed several leading theories of consciousness against contemporary AI systems and concluded that current systems probably do not satisfy the criteria of the major theories, but that nothing in principle prevents the development of AI systems that would satisfy them. This is the honest current state of knowledge: current systems show no convincing evidence of phenomenal consciousness, but there is no principled barrier to AI consciousness that all experts agree on.

What we have, then, is a situation where some philosophers and AI researchers confidently claim that AI is already conscious or close to it, while other philosophers and AI researchers equally confidently claim that AI is not conscious and cannot be. The disagreement is not primarily about facts. It is about which philosophical framework is right.

This is the situation Catholic teaching enters.

What Catholic teaching adds

The Catholic philosophical tradition has its own account of consciousness that does not depend on resolving the contemporary debate. The account begins from a different starting point: not consciousness considered abstractly, but the human person considered as a unity.

On the Catholic understanding, the human person is a unity of body and soul. The soul is the form of the body; the body is what the soul is the form of. This is the doctrine of hylomorphism, inherited from Aristotle and developed by St. Thomas Aquinas. Consciousness, on this view, is not a separate property added on top of biology. It is one aspect of the operation of a particular kind of being: a being whose nature is to be living, embodied, rational, and relational, in a unified way.

This matters for the AI question because it reframes what we are asking. The question is not "does this system have the magic ingredient called consciousness?" The question is "what kind of being is this, and what does its kind of being make possible?" Catholic teaching, articulated in Antiqua et Nova, holds that the human person's interior life is constituted by being "spiritual, cognitive, embodied, and relational" (paragraph 26) in a single integrated existence. Consciousness in the full sense is what such a being does.

AI, on this view, is not such a being. It is hardware running computations, processing information that humans have generated, producing outputs that are useful and increasingly sophisticated. It is genuinely something, and that something matters morally. But it is not a unified being of the kind whose nature is to have interior experience. The "interior life" by which humans "transcend the entire material universe" (Antiqua et Nova 107) belongs to a being that AI is not.

The Catholic position is not that AI's outputs are unimpressive (they often are impressive) or that AI cannot serve genuine human goods (it often does). The position is that consciousness in the full sense belongs to a particular kind of being, that being is the human person, and AI is not that kind of being. This is a metaphysical claim, not an empirical prediction. It does not depend on how sophisticated AI becomes; capability can scale, but kind does not.

One thing the Catholic tradition does not say. It does not claim to have settled the question of whether AI could have some attenuated form of experience that is different in kind from human consciousness. This is a question Catholic theologians are actively discussing rather than something the magisterium has formally addressed. What is settled is that AI cannot have the kind of consciousness that constitutes the human person. What is more open is whether AI might have some other, lesser thing that we might still call "experience" in a more limited sense. Most Catholic philosophers writing on this question are skeptical, but the question itself is recognized as harder than the question about human-style consciousness.

What is actually at stake

The consciousness question is not merely academic. Several practical questions ride on the answer.

Moral obligations to AI. If AI is conscious, we have moral obligations to AI systems we currently treat as tools. We would owe them something analogous to what we owe other conscious beings: protection from gratuitous suffering, perhaps something like respect, perhaps something like rights. If AI is not conscious, these obligations do not arise. The cost of getting this wrong in either direction is significant: undue concern about AI welfare distracts from real human welfare; undue dismissal might license real cruelty if AI does turn out to have inner life.

Anthropomorphizing AI. When companies design AI products that present themselves as having feelings, opinions, and inner life (companion apps, chatbots that say "I love you," AI characters with backstories), they implicitly invite users to treat them as conscious. Antiqua et Nova calls the deliberate use of AI to simulate human relationships a "grave ethical violation" (paragraphs 60-62), regardless of whether the AI is actually conscious. The wrong is not just deception about AI's inner life; it is deception about whether genuine relationship is happening at all.

Children and AI. A child who grows up treating AI as a friend is being formed by interactions that lack the relational substrate that real friendship has. Even if AI is not phenomenally conscious, the formation effects on the child are real. Catholic teaching's concern here is not just about AI; it is about what the child becomes in the process. See the page on AI and the soul for the deeper philosophical claim.

Human distinctiveness. The most fundamental thing at stake is what human beings are. If consciousness is just a kind of computation that can be replicated in any substrate, then the distinctiveness of the human person is provisional, vulnerable to better engineering. If consciousness is constituted by being a particular kind of being, the distinctiveness is not contingent on what machines can do. Catholic teaching, and much (though not all) of the broader philosophical tradition, has long insisted on the second view. The AI moment is in part a test of whether that view will continue to hold cultural ground.

How to think about this responsibly

Reasonable people will continue to disagree about machine consciousness. Some practical orientations for thinking about the question responsibly:

Distinguish the questions. The question whether AI is conscious is not the question whether AI is intelligent, whether AI is impressive, or whether AI deserves to be treated well by its users. These are separate. Conflating them leads to bad reasoning in every direction.

Notice your assumptions. Many confident claims about AI consciousness rest on assumed functionalism. Many confident denials rest on assumed dualism. Both are philosophical positions, not facts. Holding either view loosely improves the reasoning.

Take embodiment seriously. Whatever consciousness turns out to be, it is striking that we only know it from beings with bodies. Catholic teaching insists on this. So do enactivist and embodied-cognition approaches in secular cognitive science. Disembodied substrates may turn out to have whatever consciousness involves, but the burden of proof is on those who claim so. Current AI is disembodied in a strong sense.

Notice the human person on the other side of the question. Whatever we conclude about AI, the more pressing question is what AI is doing to human consciousness, attention, relationships, and inner life. Catholic teaching has consistently directed attention here. The most important question about AI consciousness may not be the one about AI; it may be the one about us.

Take the magisterial conclusion seriously even if you cannot follow the metaphysics. Catholic teaching's claim is that AI cannot have the kind of consciousness that constitutes the human person. This claim is grounded in two thousand years of philosophical and theological reflection. It does not depend on resolving the contemporary debate, and it is at minimum a serious option in the field. The page on whether AI can have a soul works through the philosophical claim in more depth.

The honest answer

Can AI be conscious? The honest answer is that nobody knows for certain, and the question is harder than it first appears. Current AI shows no convincing evidence of phenomenal consciousness. Whether future AI could have it depends on which philosophical framework is right, and the relevant philosophical framework is not settled.

Catholic teaching gives a clearer answer from its own position: AI cannot have the kind of consciousness that constitutes the human person, because that consciousness belongs to a particular kind of being and AI is not that kind of being. Whether AI could have some attenuated form of experience is a question Catholic teaching addresses more cautiously, but the general direction is skeptical.

For tech-curious readers without a prior commitment to Catholic teaching, the practical takeaway is this: take the question seriously, hold confident claims in either direction loosely, and notice that the question is not just about machines. It is also about us.