Accidentally Engineering Conscious AI

In this episode, we break down David Chalmers’ essay on whether large language models could ever possess consciousness. Chalmers argues that current systems likely lack subjective experience due to their feedforward architectures and lack of sensory grounding, but maintains that there is no principled reason silicon systems could not be conscious. We unpack the technical hurdles he identifies—such as unified agency, recurrent processing, and global workspaces—and how they form a roadmap for building potentially conscious AI. The discussion also raises the ethical implications of creating artificial systems that might one day deserve moral consideration.