Putting the I back in AI

[2018-05-26 Sat 09:10:00] event: Visions of AI, AI Singapore speaker: Leslie Kaelbling

Ability to improvise

• understand cause and effect

Innate capacities

• people, places, and things

Learning in the factory vs Learnig in the wild

Build in some structures, learn the rest

• make infinite use of finite (learned) means
• e.g. convolution assumes locality, then we compose the local pieces
• what are other structural ideas like this?

Finding a middle ground between a pile of code and a pile of neural goo

• start with hand written control and planning
• learning about the world vs learning how to reason

How can a competent robot acquire a new ability

• pouring liquid from one container to another
• gaussian process to drive sampling during planning and learning
• 2d simulation of making coffee using pouring, stirring, pushing

Advances in planning from STRIPS

• handling discrete and continuous variables
• handling uncertainty in the robot’s model of the environment
• hierachical planning