What does "conscious" mean here?
In consciousness science the word usually means phenomenal consciousness: having experiences at all, such as seeing red, feeling pain or hearing a melody. Philosopher Thomas Nagel put it as there being "something it is like" to be a creature. This is different from intelligence. A system can be very capable without experiencing anything, and some simple creatures may experience a great deal without being very capable.
So "can AI be conscious?" is not the same question as "can AI be smart?" Today's AI systems are increasingly smart. Whether any of them experiences anything is an open question.
Why an AI saying "I feel" proves very little
Large language models learn from enormous amounts of human writing, and humans write constantly about their feelings. A model can therefore produce fluent, moving descriptions of inner experience whether or not anything is going on inside. Self-reports are useful evidence for people because we know human brains generate them. For AI, the words alone cannot separate experience from imitation.
That is why serious work on AI consciousness looks at mechanisms: what the system is actually doing internally, not what it says.
How researchers try to assess it
A widely discussed approach is to take the leading scientific theories of consciousness and turn each into checkable "indicator properties". In 2023, a group of 19 researchers led by Patrick Butlin and Robert Long published a report doing exactly this. Examples of indicators include:
- Recurrent processing: information loops back through the system rather than flowing in one pass.
- A global workspace: a limited-capacity stage where information is broadcast to many specialised modules.
- Higher-order representations: the system represents some of its own states, such as how reliable a perception is.
- Agency and embodiment: learning from feedback to pursue goals, and modelling how its actions change its inputs.
The report concluded that no current AI system was a strong candidate for consciousness, but that there were no obvious technical barriers to building systems that satisfy many of the indicators. Satisfying indicators is evidence, not proof, because each theory might be wrong.
The main positions
- Computational functionalism: what matters is the computation being performed. If a machine runs the right kind of processes, it could be conscious regardless of being made of silicon.
- Biological views: consciousness may depend on features of living brains, such as metabolism or specific biochemistry, that digital computers lack.
- Integrated information theory: consciousness depends on a system's physical causal structure, so two machines running the same software could differ in consciousness depending on their hardware.
Why it matters
If future AI systems could suffer or flourish, how we build, deploy and switch them off becomes an ethical question. If they cannot, treating them as if they can could mislead people. Either way, better tools for telling the difference are needed, and that is a scientific problem as much as a philosophical one.
Our first paper builds directly on the indicator approach. We build two comparable systems, one with a recurrent workspace and a persistent self-model and one without, then disable components and measure what changes. The goal is mathematics that distinguishes mechanisms, rather than relying on what a system says about itself.
Quick answers
Is any AI conscious today?
There is no scientific consensus that any current AI system is conscious. A 2023 expert report found no current system was a strong candidate, while noting no obvious technical barrier to building systems with more of the relevant properties.
Does an AI saying it has feelings mean it does?
No. Language models learn to describe feelings from human text, so their self-reports cannot by themselves distinguish real experience from imitation. Researchers focus on internal mechanisms instead.
What is the difference between AI intelligence and AI consciousness?
Intelligence is about what a system can do, such as solving problems. Consciousness is about whether it experiences anything. The two can come apart.
Further reading
- Butlin, P., Long, R. et al. (2023). Consciousness in Artificial Intelligence: Insights from the Science of Consciousness. arXiv:2308.08708.
- Nagel, T. (1974). What is it like to be a bat? The Philosophical Review, 83(4).
- Chalmers, D. (1995). Facing up to the problem of consciousness. Journal of Consciousness Studies, 2(3).
- Seth, A. (2021). Being You: A New Science of Consciousness. Faber & Faber.