Linkblog: Building Machines That Learn and Think Like People (arxiv)

More stuff I’ve been reading: Lake, Ullman, Tenenbaum, Gershman. Building Machines That Learn and Think Like People. 2016. arXiv:1604.00289v2.

Review of recent impressive progress in AI, and what challenges remain on the path to the true AI. Pattern recognition still isn’t human-like intelligence.

Longish, ~50 p. Didn’t read the whole text, rather merely scratched the surface (-> todo).

Some notes.

Infants have cognitive capabilities like intuitive physics and intuitive psychology (theory of mind) almost from the ‘get-go’. An unprompted construction of causal models of the world. (Learning as a form of model building.)

Aside. Question. What is the difference between ‘‘model-free’’ and and ‘‘model-based’’ learning.