Hubert L. Dreyfus Publishes “What Computers Can't Do”

Is the entire field of AI just fundamentally flawed, asking the wrong questions and making the assumptions from the very beginning?

Today, as the power of these tools has flipped more to, "Uh oh, maybe we should actually be slowing this down," it seems harder to make this point.

But in 1972, it fell on more receptive and convincible ears. Learning and neural nets were "disproven" and expert systems were showing their obvious limitations. The field was vulnerable, even at its very core.

Enter philosopher Hubert L. Dreyfus and his book What Computers Can't Do: A Critique of Artificial Reason.

Dreyfus is like today's Gary Marcus, someone with a long and consistent track record of AI skepticism, regardless of developments. He couldn't be dissuaded because his opposition came from somewhere deep and unmovable.

Marcus (I mean Dreyfus) had begun his attack seven years prior in 1965's “Alchemy and Artificial Intelligence,” a paper for the RAND Corporation. His argument is familiar and persists to this day: That embodiment and experience are fundamental to intelligence.

Drawing on Martin Heidegger and Maurice Merleau-Ponty, he argued that human intelligence depends on bodies, situations and tacit skills that cannot be written down as rules, so the symbolic programs of the day would stall. To be fair, they were correct in predicting the fall of expert systems. But they were less imaginative about the potential of other technologies to be more successful where expert systems failed (hello deep learning and its roots next decade).

So was this the end of AI? As you probably read an AI summary of this text instead of the real thing, obviously not. AI researchers largely dismissed Dreyfus. The guy didn't quit. He doubled down and revised the book as What Computers Still Can't Do in 1992.

My takeaway from these critiques, which have persisted for decades, is to always be wary of saying what computers or machines can never do. Over time, they just might surprise you.