Why the Author Believes Current AI Models Can Really Think and Understand: Part 1

Summary

Andy Masley argues that current debates about whether AI can “really think” often rely on a folk theory of the human mind: the idea that thought requires a conscious inner observer with direct access to mental processes. He distinguishes phenomenal consciousness, the subjective experience of what it is like to be something, from access consciousness, in which information is broadly available to many processes. He does not claim that AI has phenomenal consciousness, but argues that thinking and understanding may depend mainly on information processing of the second kind. Human examples such as word retrieval, fluent speech, linguistic expectations, insight, and careful reasoning suggest that much cognition happens outside conscious awareness. Studies involving split-brain patients and choice blindness are presented as evidence that introspection can also produce confident but inaccurate explanations of recent decisions. The article therefore rejects the Cartesian-theater picture in which a single inner self observes and directs every thought. It reviews global workspace, higher-order, attention schema, predictive processing, and integrated information theories of consciousness. In the author’s reading, these theories place the useful role of consciousness in information sharing, attention management, or self-modeling, functions that could in principle be implemented by machines; some theories even separate consciousness from thought altogether. He also argues that language understanding does not require a vivid first-person image of every word. Understanding instead involves using concepts correctly, connecting them to other meanings and the external world, applying them in unfamiliar situations, and correcting mistakes. The article is explicitly an argument about what should count as thinking, not evidence that current models already possess every human cognitive ability. The author identifies physicalism as a major assumption and acknowledges that nonphysical consciousness or brain-specific computational abilities could undermine his conclusion. Evidence about what current AI can actually do is reserved for Part 2.