I really like this perspective shift: “where data approaches n=infinity to where data approaches n=1.” I agree, it’s an important new frontier in AI. But I would challenge the idea that it’s enabled by “learning from data.”

People certainly don’t learn from big data, but this isn’t evidence that people learn from small data, instead.

Rather, people learn through their provisional explanations of how the world works and why, explanations which are in turn criticized by reference to other explanations. Direct observations and data are in practice, rare forms of criticism.

Rarer still are good explanations.

Thanks for the post!