1. Question
Real me? Self?
How idea create?
Model of brain? computational, logical (lead to un-logical mind??)
What I real want to do? Understand how brain works
2. Thoughts
David Edelman, creation of Conscious Artifact
- Reentrant Architecture
Different than feedback system.
Bi-directionally linked, physical
- Thalamo-Cortical System
Dynamic core
- Value system, reinforcement
For an agent to modify its behavior
- Phenotype
Shape, morphology of the robot. Touch is important
- Motor control
Motor command is fed back to nervous system
Basal Ganglia
Learning and momory. Very complicated model
Problem: how to transfer knowledge to other task
Communication, not language, but showing a degree of self-awareness
Thought
Dynamic core theory
- Language: language is distinctively more sophisticated
2006, UCI, “Build A Brain” meeting
3. Todo
Theory of Neuronal Group Selection
4. Common Sense
Computational model for common sense
Current AI: just pattern recognition, function approximation
Gap: model the world, explain, image, plan, build, share
Self-driving car:
- cost: 100 billion,
- not close to solution
- long-tail problem
Common sense enable flexiable behavior
- understanding the world
- agents have their goal and plan
- abstract of knowledge about
- Physics, motion, force
- Psychology, belief, plan
Smart: learn quickly from example
- Understand others’ goal
- Have a plan
- Trying to help
Starting state: a core cognition, game engine
- learning procedure is very smart
Reverse Engineering
How? Build based on DL platform (TF, PyTorch)
Probabilistic programs
- Symbolic manipulation, enable the auto back prop for super large NN
- Bayesian inference, reasoning in uncertain situation
- Neural network good for pattern recognition and function approximation
We need a fast and good (not perfect) simulator
What to learn?
How to learn?
Intuitive physics
- train a network to inverse guess from the vision to 3d world
Intuitive psychology
What to learn?
- Program to program new
5. Build a Brain
Deep down
- see
- remember
- act
Understand the neuron (single, multiple)