Thanks, Jens. I would just scrutinize the technical limitations. The article mentions several areas of potential confusion: the expectation that descriptive systems may extend to interventional systems; assuming vertical solutions generalize to other areas; underestimating the availability of new data; and so on. (For a summary, Gary Marcus provides a good review.)

These limitations are well known and generally acknowledged among technologists, but are often neglected as the discussion moves to practical applications. When the limitations are discounted, the use cases and perceived value of prediction engines tend to inflate.

I don’t find the economic argument at all contentious. It just seems subordinate to the question of whether the technology can deliver on the promise.