Can AI develop consciousness?

Language models can write about fear, joy, or loneliness, and it sounds convincing. They have learned from billions of human texts how people talk about feelings. But a text about pain is not pain.

In 2023, researchers from neuroscience, philosophy, and computer science translated common theories of consciousness into testable features and measured AI systems against them. Their result: no system at that time met the criteria. But they also found no technical barrier that would rule this out forever. Some AI companies are therefore looking into, as a precaution, whether models might one day need to be given moral consideration.

For everyday use, this means: if an AI says it feels something, it is reproducing what its training data said about feelings. The question of consciousness should be taken seriously, but not based on what a chatbot says about itself.