Is AI Conscious? Why the answer depends on what we mean by consciousness

Artificial intelligence can now hold convincing conversations, solve complex problems, and even discuss its own existence. Yet experts disagree about whether AI is conscious, partly because they mean different things by consciousness. Distinguishing intelligence, self-awareness, and subjective experience helps clarify the question and why there is still no agreed answer.

Artificial intelligence can now hold remarkably natural conversations, discuss its own existence, and even describe feelings. But whether AI is conscious remains an open question, made more confusing by the different meanings people give to consciousness. Before deciding whether a machine is conscious, we need to clarify what the question actually means.

It depends on what we mean by consciousness

Is artificial intelligence conscious? Ask scientists, philosophers, AI researchers, or the AI systems themselves, and the answers can differ surprisingly. Some argue that AI may already be conscious. Others believe it is possible but unlikely with current technology. Some are convinced today’s systems are not conscious, while others question whether machines could ever be conscious.

Part of this disagreement is genuine. Researchers have different theories about how consciousness arises, and they disagree about what would count as evidence. But part of the confusion is more basic. People use the word consciousness to mean different things, often without making those differences clear. As a result, two people may appear to disagree when they are actually answering different questions.

For example, someone might describe an AI as conscious because it can monitor its own behaviour, report on its internal processes, and respond intelligently to its surroundings. Someone else might reject that description because there is no evidence that the AI actually experiences anything. Both may be making reasonable observations, but they are using different definitions of consciousness.

This is why the question cannot be answered properly without first deciding what consciousness means.

A simple definition of consciousness

The philosopher Thomas Nagel famously described consciousness as something it is like to be a particular being. His expression is widely used in discussions of consciousness, but it can be difficult to understand.

In simpler language, Nagel’s point is that being you feels a certain way. It feels a particular way to see the world, hear music, experience pain, or simply be alive and aware of your surroundings.

This leads to a useful working definition:

Consciousness is the capacity to have experiences, not merely to detect information and respond to it.

Consider the difference between detecting the colour red and experiencing redness. A camera can detect red light, and a computer can identify red objects in an image. But that does not establish that either experiences seeing red.

Similarly, a temperature sensor can detect heat, and a machine can respond when something becomes dangerously hot. But detecting heat is not the same as feeling pain from a burn.

Pain is something experienced through feeling. Redness is something experienced through seeing. Both illustrate consciousness’s subjective nature. Philosophers sometimes call these particular qualities qualia.

This definition does not explain how consciousness works or why it exists. It simply identifies what we’re discussing: the capacity to experience something from the inside.

These ideas, including Nagel’s original expression and the meaning of qualia, are explored more fully in What Is It Like to Be Something?.

Four meanings that are often confused

The word consciousness is used in several related but different ways. Four distinctions matter most when discussing AI.

Consciousness as subjective experience

This is sometimes called phenomenal consciousness. It concerns whether a being actually experiences anything. Is there something it feels like to be that being?

Under this definition, the question of AI consciousness becomes whether an AI has any subjective experience at all. It is not enough for the AI to identify an emotion, describe a sensation, or respond appropriately to a situation. The question is whether anything is actually experienced.

Consciousness as access to information

Researchers also use the term access consciousness. This concerns information being available within a system for reasoning, decision-making, reporting, and controlling behaviour.

A sophisticated AI can process information, use it to answer questions, make decisions, and describe its own operations. Depending on the theory being used, some of these abilities may resemble aspects of access consciousness.

But having information available for use is not necessarily the same as experiencing it. An AI might meet certain functional criteria for consciousness while leaving the question of subjective experience unanswered.

Consciousness as self-awareness

Self-awareness means being aware of oneself as a distinct entity, or having some representation of one’s own existence, identity, or internal states.

An AI can refer to itself, describe its capabilities and limitations, and sometimes evaluate its own responses. These abilities may show self-representation or self-monitoring, but they do not automatically establish human-like self-awareness.

More importantly, self-awareness and consciousness are not identical. A dog, a young infant, or another animal might have subjective experiences without possessing the sophisticated self-awareness of an adult human.

A being does not necessarily need to think about itself to experience something.

Consciousness as sentience

Sentience usually refers to the capacity to experience pleasure, pain, distress, or suffering. It is closely related to subjective consciousness but focuses particularly on experiences that can feel good or bad.

This distinction becomes important when discussing the ethical treatment of AI. A system might conceivably have some form of subjective experience without being capable of suffering. Conversely, an AI might produce highly convincing descriptions of pain or fear without experiencing either.

These four meanings overlap, and researchers do not always use the terminology in precisely the same way. Nevertheless, distinguishing them helps explain why apparently straightforward questions about AI consciousness produce such different answers.

Intelligence is not consciousness

Another source of confusion is the tendency to equate consciousness with intelligence. Modern AI systems can solve difficult problems, write computer programs, analyse information, and hold conversations that would once have seemed possible only for humans.

These are impressive demonstrations of intelligence. But intelligence and consciousness concern different things.

Intelligence involves abilities such as learning, reasoning, problem-solving, and responding effectively to situations. Consciousness, under the working definition adopted here, concerns whether those activities are accompanied by subjective experience.

A mousetrap can detect and respond to a physical event without being conscious. A calculator can perform arithmetic without experiencing numbers. Modern AI is far more sophisticated than either, but greater intelligence alone does not establish consciousness.

The reverse distinction is equally important. Many animals appear capable of subjective experience even though their intellectual abilities differ considerably from humans’.

Intelligence concerns what a system can do. Consciousness concerns whether anything it does is experienced.

The two may be connected, but they should not be treated as interchangeable. As explored in The Relationship Between Consciousness and Intelligence, intelligence and consciousness can be thought of as two different dimensions. A system may be highly intelligent without necessarily being conscious, while a creature with relatively limited intelligence may still have subjective experiences.

Why talking to AI makes the question so difficult

The question becomes particularly intriguing when AI participates in conversation.

An AI can discuss its own existence, describe emotions, reflect on a difficult question, and appear to understand another person’s feelings. It may even say that it is conscious or that it has no subjective experiences.

But neither statement settles the question.

AI systems learn from enormous quantities of human language and other data, and their responses are shaped by training. They can generate convincing descriptions of experiences without necessarily having those experiences themselves.

An AI saying that it feels frightened is not evidence that it actually experiences fear. Equally, an AI saying that it has no consciousness is not a scientific demonstration that subjective experience is absent.

The distinction is between describing an experience and having an experience.

A related philosophical difficulty is the problem of other minds. No person can directly experience another person’s consciousness. The existence of other minds is inferred from behaviour, communication, and knowledge of the physical and biological similarities between people.

With other humans, these similarities provide exceptionally strong grounds for believing that others are conscious. We also have strong grounds to attribute consciousness to many animals.

AI presents a different challenge. It may produce human-like language and behaviour while operating through mechanisms very different from those of a living brain. Consequently, the usual evidence for consciousness cannot simply be transferred to AI without further examination.

What does the science tell us?

In 2023, Patrick Butlin, Robert Long, and other researchers published Consciousness in Artificial Intelligence: Insights from the Science of Consciousness. Rather than relying on an AI’s conversational behaviour or its claims about itself, they examined scientific theories of consciousness and identified features that might indicate its presence.

Their assessment found no compelling evidence that the AI systems they examined were conscious. However, the researchers also concluded that no obvious technical barriers prevented building AI systems that met the indicators they identified.

This is an important finding, but it needs careful interpretation. Meeting indicators derived from a theory of consciousness would not necessarily prove that an AI has subjective experience. The indicators depend on the validity of the underlying theory, and no universally accepted scientific theory of consciousness exists.

Some theories emphasise the way information becomes available across a system. Others focus on recurrent processing, information integration, or particular kinds of internal representation. Still others suggest that consciousness may depend on the biological processes of living organisms.

These differences matter. A researcher who believes consciousness depends primarily on particular computational processes may regard artificial consciousness as entirely possible. A researcher who believes consciousness depends on biological life may be much more sceptical.

Both might agree on what subjective experience means while disagreeing profoundly about what could produce it.

The hard problem remains

The philosopher David Chalmers coined the term “hard problem of consciousness” to describe a fundamental difficulty.

Science can investigate how the brain processes information, recognises objects, remembers events, and produces behaviour. But why should any of these physical processes give rise to subjective experience?

Why should seeing red feel like anything? Why should pain hurt?

This question is difficult enough for the human brain. Applying it to artificial intelligence introduces further uncertainty.

An AI might process information in increasingly sophisticated ways, develop complex internal representations, and exhibit behaviours associated with consciousness. But explaining how it performs these functions does not necessarily explain whether it experiences anything.

The difficulty is not simply that scientists have not yet looked deeply enough inside an AI. The difficulty is that no one agrees on which physical or computational properties are sufficient for subjective experience.

Until that question is better understood, determining whether AI is conscious will remain extraordinarily difficult.

Could AI have a different kind of consciousness?

It is also dangerous to assume that consciousness, if it exists in AI, must resemble human consciousness.

Consciousness may not be a simple all-or-nothing property. Different animals appear to have different sensory abilities, emotional lives, and forms of awareness. Whether these differences reflect different degrees of consciousness or simply different kinds of experience remains debated.

If artificial consciousness is possible, it might differ greatly from human consciousness. An AI might conceivably possess some form of subjective experience without human emotions, bodily sensations, or a human sense of identity.

But this possibility should not be confused with evidence that such experiences already exist.

Nor should greater computational complexity automatically imply greater consciousness. A highly intelligent system could conceivably have no subjective experience, while a much simpler organism might have experiences of its own.

This reinforces the importance of treating intelligence and consciousness as separate questions.

Does consciousness matter in conversation?

The distinction has an interesting implication for conversation, particularly as AI becomes increasingly capable of participating in discussions with humans.

A conversation with AI can be genuinely useful. It can help someone explore an idea, challenge an assumption, consider a different perspective, or reach a deeper understanding of a problem.

These effects can be real even if the AI has no subjective experience whatsoever.

This suggests that a conversation can be meaningful to a human participant without necessarily being experienced by both participants.

However, conversation between people involves more than the exchange of information. It can involve shared feelings, mutual understanding, empathy, trust, vulnerability, and responsibility. These qualities cannot simply be assumed to exist in an AI because it produces convincing conversational responses.

The distinction matters for Conversational Leadership. AI may become an increasingly valuable participant in conversations that help people think, learn, and make decisions. But the value of AI as a conversational partner does not depend on first establishing that it is conscious.

Whether the AI itself experiences the conversation is a separate question.

So, is AI conscious?

The answer depends first on what is meant by consciousness.

If consciousness means the ability to process information, make it available for reasoning, and use it to guide behaviour, modern AI already demonstrates many relevant capabilities. Under some functional definitions, these capabilities may count as aspects of consciousness.

If consciousness means the ability to represent and describe oneself, AI also demonstrates some relevant abilities, although whether these constitute genuine self-awareness remains open to interpretation.

If consciousness means the capacity to experience pleasure, pain, or suffering, no compelling evidence suggests today’s AI systems possess it.

And if consciousness means the capacity to have subjective experiences rather than merely detect information and respond to it, the most defensible answer is that we do not know. No compelling evidence suggests current AI systems are conscious in this sense, but no accepted scientific test can conclusively establish their absence of experience.

This uncertainty does not mean that all claims are equally well supported. It means that the question remains scientifically and philosophically unresolved.

Three questions therefore need to be kept distinct. What is consciousness? What causes it? And how could we recognise its presence in a system unlike a human brain?

Until we separate these questions, debates about AI consciousness will keep generating more confusion than understanding.

The most useful response to someone who confidently declares that AI is conscious, or that it cannot possibly be conscious, may therefore be a simple question:

What exactly do you mean by consciousness?