ML is pattern-matching, not learning.

ML models takes a group of problems that are ‘easy for people to do, but hard for people to describe’ and turn them from logic problems into statistics problems. This works tremendously well. However, these models have no structural understanding of the question – they don’t necessarily have any concept of eyes or legs, let alone ‘cats’. These models are matching, recreating, or remixing a pattern.

Footnotes:

• ChatGPT and the Imagenet moment — Benedict Evans (ben-evans.com)