Reverse-engineering human intelligence, and engineering more - - PowerPoint PPT Presentation

Reverse-engineering human intelligence, and engineering more human-like AI

Josh Tenenbaum MIT Computational Cognitive Science Group CSAIL Department of Brain and Cognitive Sciences

March 2013

A success story: “Intelligence as statistics on a grand scale”

Long term potentiation

t n t t t n t t t

x x w y w E w x w y E

∑ ∑

= =

⋅ − = ∂ ∂ − ∝ Δ ⋅ − =

1 1 2

) ( ) ( 2 1

t n t t n t n t t t

x y w E w x w y E ⋅ = ∂ ∂ − ∝ Δ ⋅ = =

∑ ∑ ∑

= = = 1 1 1 2 2

) ( 2 1 2 1

How do we get so much from so little?

(or, where sparse data meets rich knowledge)

“tufa” ¡

How do we get so much from so little?

(or, where sparse data meets rich knowledge)

How do we get so much from so little?

(or, where sparse data meets rich knowledge)

The approach: probabilistic programs

World state (t-1) World state (t) World state (t+1) Image (t-1) Image (t) Image (t+1)

physics graphics … …