Whether these systems experience anything. The honest answer is that nobody knows, partly because we have no agreed test even for other people, only inference from similarity. The question recurs because systems that discuss their own states fluently make it feel pressing, which is not the same as it being answerable.
Technical view
Approaches apply indicator properties from theories such as global workspace and higher-order theories, and assessments generally find current architectures lack several proposed markers while noting none are decisive. The hard problem means behavioural evidence is inherently inconclusive, and self-report is unreliable, since a model’s statements about its inner life are shaped by training data and preference optimisation rather than introspection.
Related
Part of AI Terminology Index, section: Alignment and Safety Concepts.