ACTR: Moving beyond production systems

1
ACT-R Moving beyond production systems

• CS/ISYE/PSYC 7790
• Instructor Ron Ferguson
• Fall 2003

2
Administrivia

• Lab 1 due today
• Lab 2 today at 400 in Media Lab
• Today ACT-R system
• Next week ACT-R performance and learning (ch.
  3-4)

3
Assignments for next week

• Monday Weekly ticket (1 page)
• Wednesday in lab One page description of
  possible project

4
Requirements for ticket

• Two tendencies that you should try to avoid
• Summary of readings
• Attempting to make modeling technique pass the
  Turing Test

5
Potential questions for next weeks ticket

• ACT-R has a specific model of goals and subgoals
  which depends on a goal stack. However, do humans
  handle goals and subgoals in the same way (or as
  well) as the stack mechanism in ACT-R?
• Contrast two of the learning methods in ACT-R.
  Are there other learning methods that might be
  useful?
• ACT-R is very concerned about things like fan
  effects in memory. When might this be an
  effective prat of modeling, and when might it not
  be?

6
One-pager on possible project

• Strictly informal
• Designed to get you thinking about a project
• Guidelines
• Focus on input/output characteristics of the
  model
• Determine a general knowledge representation
  scheme
• Try to give at least one example task
• Dont worry too much about implementation yet.

7
Example for one-pager

• Domain when do we use a metaphor or a simile?
• Example
• Atom is like a solar system
• Atom is a solar system
• Idea Use Ortonys salience imbalance model with
  some tweaks
• Input/output, KR
• Set of features for each item
• Salience ratings for each of the features (toxic
  in cigarettes, for example)
• Might use a connectionist network, might not

8
Last time

• What constitutes a production system?
• How can production systems be mapped onto
  psychological phenomena?
• Productions as mediating states

9
Outline for today

• How ACT-R attempts to move beyond some of the
  limitations of production systems as cognitive
  models
• Memory
• Knowledge representation

10
Four claims

• Claim 1 ACT-R is not just a production system
  it is a production system that attempts to make
  specific claims about the interaction of
  problem-solving and memory. These additions to
  the architecture give it more explanatory power
  than a classical problem solver
• Claim 2 The representation used by ACT (and
  then ACT-R) is frames, not predicate calculus or
  symbols, and there is a reason for that (even
  though they are technically equivalent)

11
More claims

• Claim 3 Anderson claims that we can also
  generalize from the condition/action asymmetry of
  rules. In some places this makes a lot of sense,
  but in other places it does not, because not all
  asymmetries have the same character as that of
  rules.
• Claim 4 Anderson succeeds in showing how memory
  and problem solving may interact. His claims for
  handling other cognitive phenomena are less
  convincing

12
Claim 1 ACT-R is not just a production system
13
(No Transcript)