Skill Learning

1
Skill Learning

• What does it mean by skill?
• More complex than remembering information.
• You need to solve a problem.
• Skill learning Problem solving
• E.g., text editing

2
How do we learn?

• Proceeds in predictable pattern.
• Power function
• Interesting studies
• Crossman (1959)
• Observed workers in a cigar factory over a
  10-year period.
• Followed power function.
• At the end, the workers reach cycle time of
  equipment.

3
Similar results...

• Ohlsson (1992)
• Examined Isaac Asimovs writing skill.
• Asimov wrote over 500 books over 40-year period.
• Length of time it takes to write a book and the
  number of completed books - a linear log-log
  function.
• So, skill learning is similar to any other
  learning situation.

4
But...

• Fitts (1964), Anderson (1982)
• Skill acquisition goes through several stages.
• Three stages
• Cognitive stage
• Associative stage
• Autonomous stage

5
Cognitive stage

• First, you need to know what to do.
• You start from declarative knowledge.
• How do we find out what to do?
• Newell and Simon (1972)
• Developed General Problem Solver to simulate
  how humans solve problems.

6
GPS

• A problem solving situation is describe by four
  concepts.
• States - Where you are.
• Goals - The final state you want to be in.
• Operators - Steps that you can take to change the
  current state.
• Search - finding a sequence of operators that
  transform the initial state into the final state.

7
How do we select operators?

• First you have to acquire them.
• This is the domain of conditioning and human
  memory.
• e.g., press a lever in a Skinner box
• But, then you have to select which one to use.
• Two principle mechanisms
• Difference reduction
• Operator subgoaling

8
Difference reduction

• To reach a goal, you must reduce the difference
  between the current state and the goal state.
• To do so, you must select operators that
  accomplish that.
• But, often it is not straight forward like that.
• Often, you must abandon the pursuit of goal.

9
And instead.

• Pursue subgoals - Operator Subgoaling.
• To satisfy hunger.
• State A (Hunger) -----Goal State (Full stomach)
• Operator - eat
• subgoal 1 - find food
• subgoal 2 - build a tool
• subgoal 3 - learn how to build a tool
• subgoal n -

10

• The use of subgoals is characterized as
• push-down pop-up goal stack

11
Example

• Initial state
• 2(3x - 11) 3x 8
• Goal state
• x some number
• How to reach the goal state?
• You need to know operators.
• You need to select operators to reduce difference
  between the two.

12
Do we really do this?

• Hobbits and Orcs problem.
• One side of a river are three hobbits and
  three orcs. There is a row boat on their side,
  but only two creatures can row across at a time.
  All of them want to get to the other side of the
  river. At no point can orcs out number hobbits
  on either side of the river (or the orcs would
  eat the outnumbered hobbits). The problem, then,
  is for the creatures to find a method of rowing
  back and forth in the boat such that they all
  eventually get across and the hobbits are never
  out numbered by orcs.

13
Solution

• 12 states are involved.
• State 6 and 7 are critical and very difficult for
  subjects.
• Reason - Subjects must temporarily abandon the
  difference reduction strategy.
• Conclusion - We do use the difference reduction
  strategy.

14
What about operator subgoaling?

• Tower of Honoi problem.
• Move all the disks from A to C.
• but, you can do it one at a time and you cannot
  put a bigger disk on a smaller disk
• How to do this?
• Optimal - use 15 moves.
• Some moves are in pursuit of subgoals.

15

• Time it takes to make a move is highly correlated
  with the number of subgoals you are pursuing.

16
Associative stage

• Once you know what to do, you can then
  proceduralize.
• Move from declarative to procedural knowledge.
• How?
• Create production system.

17
Production system

• A series of if-then.
• Specify under what condition what action to take.
• Reduce the amount of cognitive involvement.
• Evidence

18
Neves and Anderson (1982)

• Think aloud technique
• Asked subjects to solve problems that used the
  side-angle-side rule.
• At the beginning, subjects had to reason through
  everything.
• Then after a few problems, a drastic reduction in
  reasoning had occurred.
• Subjects simply recognized that the problem
  needed the side-angle-side rule.

19
Characteristics of Production System

• Knowledge becomes asymmetrical.
• Going from If to Then is easy.
• Going from Then to If is very difficult.
• Knowledge becomes less and less conscious.
• E.g., Statistical analysis.
• More difficult when productions become more
  unitized through composition.

20
Expert Rules

• Experts seem to have different production rules.
• Larkin (1981)
• compared novices and experts on a physics
  problem.
• Novices used working-backward rule.
• Expert used working-forward rule.

21

• Novices
• If the goal is to calculate quantity v
• Then try to use the principle v at and set as
  subgoals to find a and t.
• Experts
• If the quantities a and t are known
• Then calculate v according to the formula v at.

22
How to become an expert?

• Acquire hundreds and thousands of production
  rules.
• Hayes (1985)
• Studied geniuses
• found - Most of them took 10 years to become an
  expert.
• Ericsson et al. (1993) - No contribution from
  innate talent.

23
One difference

• Superior ability to remember information about
  the problems related to their expertise.
• E.g., Chess masters
• They can reproduce patterns after single glance.
• No other difference.
• Why? They already have those patterns in their
  memory.

24

• Need a lot of production rules to become an
  expert.
• Artificial intelligence programs
• The same.
• Need to put many production rules before they can
  simulate the experts performance.