Topic 5 - Symbolic Systems

Question 1

Four Properties

Answer 1

• Representations
• Transformations
• Serial Processing
• Sense-think-act

Question 2

Eliza

Answer 2

• Rogerian Therapy dictates that people can’t be told how to solve their problems, they need to discover it themselves
• Very simple chatbot called Eliza basically repeats things back to the patient
• Representations: User input strings
• Transformations: Predefined responses to specific inputs
• Serial Processing: Running the program
• Sense-think-act: Input problem, process, output therapy

Question 3

Shrdlu

Answer 3

• Chatbot that can manipulate objects in a 3D virtual environment
• Representations: User input, blocks in world
• Transformations: Responses, rules of how blocks can be manipulated
• Serial Processing: Running the program:
• Sense-think-act: Input problem, figure out how to solve it, move blocks accordingly

Question 4

Soar

Answer 4

• Problem being solved: Move train engines and flatcars from one configuration to another along a split track
• Representations: Problem spaces with different problem states and operators (actions)
• Rules: When there’s an impasse, a new “selection” problem space is created with the goal of breaking the tie, which uses more switch spaces recursively
• How Soar learns: When an impasse is resolved, Soar stores the configuration of states and moves and the state they lead to in chunks
• How many times does Dr. Rosenbloom say Chunk: 10

Question 5

Mycin

Answer 5

• Blood disorder diagnosis bot
• Eliminates possible diagnoses until it is reasonable certain of a particular result
• Outperformed Stanford med students and faculty, but isn’t used for legal/ethical reasons
• Can’t diagnose anything that isn’t a blood disease

Question 6

WATSON

Answer 6

• IBM bot that won jeopardy

- Weighs evidence to answer questions based on a lot of different factors

Question 7

Pedro

Answer 7

• Lego MINDStorms bot that would move in response to sounds or proximities
• Representations: Sensory data
• Transformations: Specific instructions for when sensory data is acquired (move forward when sound is loud, etc)

Question 8

Marr’s Three Levels

Answer 8

• Explanation has three levels: Task description (eg math), algorithm description (eg computer program), physical description (eg transistors and bits)
• We won’t be truly “done” cognitive science until we understand all three for our brains

Question 9

Edge Detection

Answer 9

• Detecting boundaries between objects is hard
• Difference of Gaussian’s Filter (algorithm level)
• There are cells in the LGN in our brains that seem to more-or-less do this (physical level)

Question 10

Neural Network Edge Detection

Answer 10

• Neurons get inputs from other neurons, some of which repress other signals
• If there’s a strong reaction in the middle and a weaker connection to the sides, then we find an edge
• Try using simulated neurons to figure out how the brain works