Robotics

1
Robotics

• HAL, COG and Embedded Cognition

2
Disembodied AI

• Advantages
• Dont need to worry about navigating through
  space, manipulating objects.
• Treats intelligence as a purely symbolic
  activity. (Monolithic representation and control)
• Problems
• There are no monolithic representations or
  controls in human intelligence.

3
Brooks on classic AI

• Abstraction is dangerous
• Programmers, not computers do the abstraction
• Computers then just search
• Programming still based on introspection
• Newells protocol analysis
• But the Merkwelt may not match up.

4
Intelligence without Representations

• Conclusion and hypothesis
• Conclusion When we examine very simple level
  intelligence we find that explicit representation
  and models of the world simply get in the way. It
  turns out to be better to let the world itself
  serve as its own model
• Hypothesis Representation is the wrong unit of
  abstraction in building the bulkiest parts of
  intelligent systems.

5
Brooks argument from evolution

• Single celled organisms
• 3.5 billion years ago
• Photosynthetic plants
• 2.5 billion
• Fish and vertebrates
• 1 billion
• Insects
• 450 million

6
Evolution, continued

• Reptiles
• 370 million years ago
• Dinosaurs
• 330 million
• Mammals
• 250 million
• Primates
• 18 million
• Man
• 2.5 million

7
Evolution, .

• Human agriculture
• 19,000
• Writing
• 5000
• THEREFORE
• High level knowledge isnt hard, its locomotion
  and action thats hard!

8
Evolution Summary Chart
9
Brooks Approach

• Decomposition by activity
• Not by function
• Engineering problem
• Build creatures completely autonomous mobile
  agents that co-exist in the world with humans,
  and are seen by those humans as intelligent
  beings in their own right.

10
Features of Human Intelligence

• Development
• Social Interaction
• Physical Interaction
• Embodiment
• Physical Coupling
• Integration

11
Development

• Intelligence is developed in humans.
• So we should try to scale intelligence in
  machines by giving them the resources to become
  more intelligent.
• Need to think about structure of the robot and
  the environment, including the caregiver.

12
Social Interaction

• Learning facilitated by relationship between
  caregiver and learner.
• Kismet trains the caregiver.
• Scaffolding
• Reducing distractings
• Marking critical features
• Demonstrating solutions

13
Physical Interaction

• Embodiment
• Environment is the representation
• Physical coupling
• Physical world provides some of the intelligence
  gravity, friction, .

14
Integration

• Sensory and motor coordination
• Head position and visual tracking
• VOR Vestibulo-ocular reflex
• OKN Opto-kinetic nystigmus
• Coordination of different sensory modalities
• The McGurk Effect

15
Key Ideas

• Situatedness
• Embodiment
• Intelligence bottom up
• Emergence intelligence is in the eye of the
  observer