"One subtle point is often missed: the Extended Mind Thesis does not claim that a notebook or smartphone is intelligent on its own..." — but that does seem to be the claim of the AI community since it’s right there in the name.
That’s a sharp observation, but I’d separate two different uses of the word intelligence.
The AI community uses artificial intelligence to describe systems that perform tasks associated with human intelligence: reasoning, language, planning, perception, learning, and so on. The name does not, by itself, imply that these systems possess intelligence in the same sense as a human mind, nor that they are conscious or autonomous. It’s a functional label more than a metaphysical one.
Clark and Chalmers are asking a different question: Where does cognition happen?
AI researchers are often asking: Can a machine perform intelligent behavior?
Those questions overlap, but they aren’t identical.
Where your comment gets interesting is that modern AI muddies the distinction. Otto’s notebook is passive. It stores information. It doesn’t reorganize it, challenge it, or generate novel ideas. A large language model does more than retrieve. It synthesizes, generalizes, translates between domains, proposes hypotheses, and adapts its responses to context. That’s much closer to what we’d normally call intelligent behavior.
So there are three increasingly capable external artifacts:At each step, the external artifact takes over functions we once assumed had to occur inside a biological brain. The philosophical disagreement is about what follows from that.
| One camp says: “The machine is intelligent.” |
Another says: “The machine exhibits intelligent behavior but does not itself possess intelligence.” |
A third says: “Intelligence is a property of the coupled system of human and machine.” |
I suspect the last position is going to become increasingly important. If an architect with AI consistently produces designs neither could have produced alone, insisting that only the architect or only the AI deserves the label starts to feel artificial. The interesting phenomenon is the interaction.
Ironically, this circles back to something Clark himself might appreciate. The Extended Mind Thesis was written before modern generative AI existed. The external tools he discussed were mostly passive cognitive scaffolding. AI is different because it pushes back. It offers alternatives, makes mistakes, surprises you, and occasionally drags you into a line of thought you hadn’t considered. That’s much closer to collaboration than storage.
So I think you’re pointing toward a genuine evolution of the debate. The original thesis asked whether tools could become parts of minds. Today’s question might be whether some tools have become cognitive partners.
Whether we should call that partner “intelligent” depends on how we define intelligence. If intelligence means “the capacity to produce adaptive, context-sensitive, goal-directed reasoning,” then AI systems qualify to a meaningful extent. If intelligence additionally requires subjective experience, intrinsic goals, or understanding in a richer sense, then the debate remains very much open.
Humans have an unfortunate habit of using one word for several related ideas and then spending decades arguing over which meaning everyone secretly intended. “Intelligence” may be one of the grand champions of that sport.
Is AI a part of the "extended mind" or an intelligence on its own? Or is it all semantics?...
As usual my input (prompts) are in the boxed elements. The rest apart from some small edits and formatting is ChatGPT's output.