A fascinating read from Vol 360 of Science (May 2018, Issue 6391) on efforts of learning from humans how AI could acquire skills in the future. Ongoing attempts are thrilling, such as of listing human instincts to be baked later into AI (10 so far), however there’s much more to learn from the article:
Interaction networks (incorporating assumptions about physical world) and methodological bias of charming learning people against putting background knowledge in large amounts
Intelligence Quest - MIT initiative to understand human intelligence in engineering terms,
Schema networks (idea from Vicarious, a robotics software company) to assume the existence of objects and interactions, and trying to infer the causality that connects them.
Probabilistic programming language - and effort of MIT in creating Pyro.
Highlights and call outs from the article (made using a whiteboard on an iPad) below:
Full text – see Science Archive here.