What ChatGPT Still Can't Do (But We Might Do With It): Hubert Dreyfus and Extended-Mind Cyborgs

In the early 1970s, Hubert Dreyfus published a controversial book called What Computers Can’t Do (Dreyfus 1972). Two decades later, he published What Computers Still Can’t Do (Dreyfus 1992). Dreyfus died in 2017, before the widespread availability of generative AI tools using large language models. But it is instructive to consider how his “Critique of Artificial Reason” relates to the latest generation of AI, and what light the new tools cast on the struc- ture and limitations of his critique. We begin with a review of the historical record, clarifying what Dreyfus did, and what he did not say, about the limits of machine intelligence. We adjudicate the controversy over his assessment of the prospects for computer chess and provide an analysis of the master argument upon which his critique of AI relied. In assessing the significance of that critique, we turn our attention to Deep Blue, IBM’s chess-playing super- computer that defeated the then-reigning world champion in 1997. We argue that one element of Deep Blue’s programming architecture points towards a neglected alternative in Dreyfus’ disjunctive argument about the limits of artificial intelligence and an affinity with recent generative AI tools. We intro- duce the concept of extended-mind cyborgs to reflect on the opportunities and limitations of programs such as ChatGPT.