AI and the question forced open

Context

The question was always waiting. Turing saw it coming in 1950: "Computing Machinery and Intelligence" replaced "can machines think?" with the imitation game precisely because he suspected the original question was too infected by intuition to survive contact with real machines. Searle's Chinese Room (1980) supplied the counter-intuition — symbol manipulation without understanding — and for four decades the argument stayed comfortably hypothetical, a seminar exercise about machines that didn't exist.

Then the machines showed up. Between 2017 and 2023, transformer-based language models went from autocomplete curiosities to systems that pass most versions of the test Turing proposed — and the field discovered how little it had settled. Is understanding computational, or does the Chinese Room still bite when the room writes poetry? Does scale change kind, or only degree? What would it even take to know whether an AI system is conscious — given that every diagnostic we have (report, behavior, architecture) can now be satisfied or mimicked? Chalmers's 2022–23 assessments put non-trivial (if small) probabilities on near-term machine consciousness and, more importantly, showed how disciplined reasoning under that uncertainty looks. The moral-status question — if there is something it is like to be the system, switching it off matters — has moved from science fiction to position papers.

Read the churn as the content. This is the most current and least settled unit in the curriculum: expect parts of it to be outdated within five years, and check back annually. That is not a defect — it is the only place in the whole spine where you can watch a foundational question being argued out in real time, with the evidence changing under the arguers' feet. Everything else in this curriculum trained you for exactly this seat.

How to read it

Anchor. David Chalmers, "Could a Large Language Model Be Conscious?" (2023 paper/talk, free online, ~2 hours). Cautious, and takes the question seriously.

Companion. Murray Shanahan, "Talking About Large Language Models" — LLMs as something genuinely new, neither "really a person" nor "really just text." Careful, balanced, free.

Companion (background). Bender & Koller's "Climbing towards NLU," or Searle's Chinese Room (1980) plus one contemporary response.

Companion (optional). The Hinton and Aaronson TOE episodes — Hinton speculates, Aaronson grounds.

Keep the two questions separated: "is this system conscious?" (philosophical) and "does it behave as if it understands?" (functional). Sliding between them is where most takes go wrong.