The hard problem of consciousness is the question of why physical processes in a brain - or any system - are accompanied by subjective experience, rather than going on with no inner feel at all. The philosopher David Chalmers named it in his 1995 paper “Facing Up to the Problem of Consciousness,” self-archived on his website, by contrasting it with the “easy problems.”
The easy problems are the things cognitive science can in principle explain by finding a mechanism: how a system discriminates inputs, integrates information, directs attention, and reports on its own states. The hard problem is left over once all of those are solved. Suppose we had a complete account of every functional process underlying vision. We would still face the question of why those processes are accompanied by the experience of seeing - the redness of red, the quality Chalmers and others call qualia. Nothing in a mechanistic description seems to entail that experience must arise; one can coherently imagine all the same processing happening “in the dark.”
The distinction has become central to discussions of machine consciousness. Everything we can observe about an AI system - what it does, what it says about itself, how it solves problems - speaks to the easy problems. Whether there is any subjective experience behind the outputs is the hard problem, and by its nature no behavioral test reaches it. This is why episodes like an engineer claiming a chatbot is sentient are so hard to adjudicate: fluent self-report is exactly the kind of evidence that cannot, even in principle, settle the hard question.
Why business readers should care: when systems describe their own “feelings” or appear to suffer, the hard problem explains why such claims cannot be confirmed or denied from the outside, which matters for how seriously - and how carefully - companies and regulators should treat assertions about AI sentience.