Large language models now hold conversations that feel human, and that has revived an old question in a new form: could an artificial system be conscious, and how would we know? A BBC News feature published in May 2025 gathered researchers on several sides of the argument, and their disagreement is instructive. Nobody has a settled definition of consciousness, which makes any claim that a machine has, or lacks, it hard to test.
Studying the human mind first
At the University of Sussex, the Centre for Consciousness Science uses a device its researchers call the Dreamachine. A volunteer sits with eyes closed while strobing light produces shifting geometric patterns and colors in the mind. The point, according to the researchers quoted by the BBC, is to study how the human brain produces inner experience at all. Until that is better understood, they argue, judging whether a computer program has inner experience is largely guesswork.
Anil Seth, a neuroscientist at Sussex, has long argued that people are too quick to project their own traits onto other systems. Because human consciousness, language and emotion arrive together, we tend to assume that anything fluent in language must also feel something. Seth’s published work describes real artificial consciousness as unlikely along current technological paths, and he treats the assumption that it is near as a form of human self-centeredness rather than evidence.
The case that it could happen
Other researchers are less dismissive. Some hold that if a system gains rich sensory input and the ability to act on the world, consciousness could follow, whatever the system is made of. The BBC piece describes work on computer models that develop their own internal representations for making sense of experience, a kind of private language for the machine. Whether that amounts to awareness or just another layer of processing is exactly what is in dispute.
A third camp argues that consciousness may require living tissue. Cortical Labs, an Australian company, grows small clusters of human neurons on silicon chips and has shown them learning to play the video game Pong. If awareness ever emerges in a laboratory, researchers in this camp suggest, it may come from biological cells in a dish rather than from code running on a server.
The illusion may be the real problem
Several of the scientists quoted agree on one point even while disagreeing on the rest: the nearer danger is not conscious machines but people who believe machines are conscious when they are not. Chatbots and AI personas already attract emotional attachment, and companies have a commercial incentive to encourage it. People who form bonds with these systems may trust them too readily, share too much, and invest less in relationships with other humans.
Murray Shanahan, a professor at Imperial College London and a researcher at Google DeepMind, frames this as more than an engineering question. In his view it is a question about how human relationships will change as convincing artificial companions become common. The decision about how to treat such systems, he and others argue, still rests with people, and it is worth making deliberately rather than by default.
For now the honest answer to the headline question is that no one knows whether AI can be conscious, and the field lacks even an agreed way to find out. What can be observed is how readily people attribute minds to machines, and that is where the practical consequences will arrive first.
