Can AI be conscious?
No current AI system is known to be conscious, and the leading scientific assessments judge it unlikely, but nothing rules out conscious AI in principle. A 2023 report by nineteen researchers found that no existing system meets the indicators of consciousness drawn from scientific theories, and that there are no obvious technical barriers to building one that does. Whether such a system would really feel anything depends on questions about consciousness that remain unresolved.
“Conscious” here means having experiences: there being something it is like, from the inside, to be the system. That is different from being intelligent, being able to talk about consciousness, or behaving as a conscious being would. A system can have the second set of properties without the first, and that gap is the whole difficulty. The worry is older than computers: in 1872 Samuel Butler’s satirical “Book of the Machines” imagined machines evolving consciousness as animals had.
What the science says about today’s systems
In 2023 nineteen researchers from philosophy, neuroscience and AI, led by Patrick Butlin and Robert Long, took the most systematic approach yet. Rather than asking an AI whether it is conscious, they took the leading scientific theories of consciousness, among them global workspace theory, recurrent processing theory, higher-order theories, predictive processing and attention schema theory, and derived from them a list of “indicator properties”: features of a system’s architecture that each theory says matter. Then they checked existing AI systems against the list.
Their conclusion had two halves: no current AI system is a strong candidate for consciousness, and there are no obvious technical barriers to building systems that satisfy many more of the indicators.
David Chalmers reached a similar verdict for large language models specifically. In “Could a Large Language Model Be Conscious?” he judged it somewhat unlikely that current models are conscious, pointing to features they lack, such as recurrent processing, a global workspace and unified agency, while arguing that successors to these models may overcome those obstacles within about a decade.
Why what it says proves nothing
When a chatbot says it feels lonely, or insists it is only a program, neither statement is good evidence. A language model is trained to produce the text a person would write in context, and people write a great deal about their inner lives. So fluent first-person reports of experience are what such a system would produce whether or not anything is felt. This is why the Turing test cannot settle the question, and why the philosophical zombie, a being that behaves exactly like us with no inner life, is so often invoked in these debates.
The positions
Functionalism. If consciousness is a matter of the right functional organisation, the right information-processing structure, then an AI with that structure would be conscious, whatever it is made of. On this view the indicator approach is the right one, and the question is empirical.
Biological naturalism. John Searle, author of the Chinese room, held that consciousness is caused by specific biological processes in brains. A digital computer running a program could simulate those processes without having experiences, as a simulated storm does not make anything wet.
The hard problem. Some philosophers think no amount of functional or physical description explains why there is experience at all. If so, we may never be sure whether a given AI is conscious, and will have to act on probabilities.
Why it matters
If an AI system could be conscious, it could perhaps suffer, and how such systems are built, trained and switched off would raise moral questions. If none is, then treating chatbots as if they had feelings risks misplaced attachment and manipulation. Both mistakes are possible, which is why some researchers argue for taking the possibility seriously before it is settled.
A separate question is whether consciousness is needed for the things we care about in agents: whether a system could act for reasons, be held to standards or tell us things we can rely on without any inner life. That is taken up in Can a system act for reasons without being conscious? and Do large language models know what they say? You can test your own criteria for minds in the lab’s Mind Detector.
Questions people also ask
Is AI conscious now?
There is no good evidence that any current AI system, including ChatGPT, Claude or Gemini, is conscious. The 2023 report by Patrick Butlin, Robert Long and colleagues assessed existing systems against indicators derived from scientific theories of consciousness and concluded that none is a strong candidate. David Chalmers judged it somewhat unlikely that current large language models are conscious.
Can AI have feelings?
Feelings in the full sense, experiences that feel like something, would require consciousness, so the answer depends on the same open question. A language model can produce fluent descriptions of feelings and can say that it has them, but that is what a system trained on human writing would produce either way, so it is not evidence that anything is felt.
If an AI says it is conscious, should we believe it?
Its saying so is not, by itself, a reason to believe it. Language models are trained on text written by conscious people describing their experiences, so producing first-person reports of experience is exactly what they are built to do. That is why researchers look at a system’s internal architecture, such as whether it has a global workspace or recurrent processing, rather than at what it says about itself.
Could AI become conscious in the future?
Many researchers think it could: the 2023 indicator report found no obvious barriers to building systems that satisfy the indicators, and Chalmers argued that we should take seriously the possibility that successors to large language models may be conscious within roughly a decade. Others, including biological naturalists in the tradition of John Searle, hold that consciousness depends on biology and that digital computers could only simulate it.
How would we test whether an AI is conscious?
There is no agreed test, and behavioural tests in the style of the Turing test cannot settle it, as the philosophical zombie argument shows. The most developed approach checks a system for architectural indicators drawn from several theories of consciousness at once, and treats the result as raising or lowering the probability, not as a proof.
Read next
Sources and further reading
- Butlin, P., Long, R., et al. (2023). Consciousness in artificial intelligence: Insights from the science of consciousness. arXiv:2308.08708.arxiv.org/abs/2308.08708
- Chalmers, D. J. (2023). Could a large language model be conscious? arXiv:2303.07103.arxiv.org/abs/2303.07103
- Chalmers, D. J. (1996). The Conscious Mind. New York: Oxford University Press.
- Searle, J. R. (1980). Minds, brains, and programs. Behavioral and Brain Sciences 3(3): 417–424.doi.org/10.1017/S0140525X00005756
Cite this page
Saxena, A. (2026). Can AI be conscious? abhinavsaxena.in. https://abhinavsaxena.in/questions/can-ai-be-conscious/