Steps Towards More Human-Like Learning in Machines - Josh Tenenbaum

Explore the frontiers of artificial intelligence and human-like learning in machines through this lecture by Josh Tenenbaum at the Institute for Advanced Study. Delve into the myths of machine learning, core knowledge in human intelligence, and the concept of commonsense core knowledge. Examine probabilistic programming, game engines, and simulation techniques used to model intuitive psychology. Discover how babylike learning and learning in game engines contribute to AI development. Investigate examples in perception, plan interactions, and low-level learning using physics engines and amortized inference. Analyze the hard problem of learning, children's learning processes, and one-shot learning in the Omniglot domain. Explore inverse motor programs, Bayesian inference, and probabilistic programs in classification tasks. Examine generative models, drawing styles, and the wake-sleep algorithm in neural components learning.

Introduction
AI is a uniquely exciting time
Point the gap
Myths of machine learning
Human intelligence
Core knowledge
Alan Turing
Commonsense core knowledge
Intuitive psychology
probabilistic programming
game engines
simulation
probabilistic simulation
probabilistic simulation demo
intuitions
building and thinking
learning from scratch
babylike learning
learning in game engines
examples
perception
plan interactions
lowlevel learning
simulation engine
physics engine
amortized inference
a physics engine
simple shape parameters
tackling problems
looking around
trial and error
virtual tools game
trial error learning
SM model
Learning simulation engines
Hard problem of learning
Childrens learning
Oneshot learning
Omniglot domain
Inverse motor program
Bayesian inference
probabilistic programs
classification task
human scale
human version
generative models
drawing styles
more structure
more interesting model
learn neural components
wakesleep algorithm