Topic 7 Flashcards
A not B Task
- Experimenter hides a toy from a baby in one of two holes. They repeat this a few times, then “hide” it in the second hole. The baby will still look for the toy in the first hole even though they saw it go in the second.
- If the child’s position is changed before the trials with the second hole, they will reach for the second hole
- Suggests that this is a motor mistake (moving how we’re used to moving) rather than a knowledge mistake
- Children make the mistake more often if attention is drawn to their arms
- Combine “act” and “think” in sense-think-act
Topology of the race
-A race’s last mile overlaps its first, but the pamphlet depicts the hills in the last bit much steeper, even though they’re the same hills
Perception of the steepness of hill - Top/Bottom
- People at the top of a hill overestimate how steep it is, people at the bottom are closer to the actual slope
- If we think of sense-think-act, this doesn’t make sense, because the actions should come at the end, rather than influence the “think” step
Steepness - Before vs. After a run
-People estimate slopes to be steeper after a run than before
McGurk Effect
- Face says “ga,” audio says “ba,” most people hear “da,” which is ostensibly in the middle
- We use visual perceptions to influence audio ones
McGurk’s Face
- Same as McGurk effect, but instead of watching someone’s face a robot manipulates your face
- Same result, which indicates that motors are part of decision making and perception
Reading Activates Motor Representations
- Mirror neurons, but for reading
- “Active” language causes more brain activity than abstract language
- Rather than purely symbolic representations, information can be stored as your physical relationship with that thing
Robot moving around a cluttered office
- Robot’s navigation based on studies of a rat brain
- Landmarks are denoted by the movements performed near them
Subsumption architecture - Behaviour-based AI
- Several robots share the same approach
- Use the evolutionary approach - solve one problem, then the next, then the next, etc.
- Don’t store info about the world that can be easily observed
- Use simplest possible cognitive systems
Cog Video
- Robot “Cog” built to learn from the ground-up how to do things
- Most robots can be completely thrown off if their environment is changed; clearly this is not how it happens in nature
- Cog should eventually learn to be human-like
Decomposing - simultaneous layering vs sequential stages
- Traditional model is sense-think-act
- With behaviour-based decomposition, we have multiple sense-think-act systems running simultaneously
- No “central” process
Allen - example of layering, minimal interaction
- Simple robot with multiple iterations
- Level 0: Avoids approaching people and running into objects
- Level 1: Adds a wandering system
- Level 2: Adds following hallways
- Each level adds onto the previous one without changing anything about the previous one
Herbert - soda can robot
- Several systems in a robot designed to take soda cans
- None of the systems “know” about each other, they only work on information available to them
Dorsal/Ventral Streams
-“Where” vs “What” streams for object detection
-
Letter Posting Example
- Patient with damage to ventral stream cannot align different lines in a circle
- If given a different task (put a letter in a slot) which is the same problem redone as a motor problem, they succeed
- People with damage to dorsal stream have the opposite problem
- This suggests our cognitive processes are not centralized, ie. one task can be done in multiple ways by different parts
