cosc 4126 learning - d goforth automation of learning learning machines
TRANSCRIPT
COSC 4126 learning - D Goforth
Automation of learning
learning machines
COSC 4126 learning - D Goforth
Pavlov – 1904
doing digestion experiments on dogs dogs salivated on seeing/smelling food in lab, a bell rang just before dogs fed dogs ‘conditioned’ to salivate
when bell rang
accidental discovery of learned response
COSC 4126 learning - D Goforth
Automation begins nineteenth century – public education
large numbers of learners cost and number of teachers many-to-one efficient
“scientification” of psychology from ‘mental muscle’ to research of
phenomena: memory and perception
COSC 4126 learning - D Goforth
Edward Thorndike “If, by a miracle of mechanical ingenuity, a
book could be so arranged that only to him who had done what was directed on page one would page two become visible, and so on, much that now requires personal instruction could be managed by print.” Thorndike, E.L. (1912). Education: A first book. New York: The MacMillan Company
is this game levels?
COSC 4126 learning - D Goforth
Edward Thorndike
three conditions that maximized learning Law of effect: stated the likely recurrence
of a response is generally governed by its consequence or effect, generally in the form of reward or punishment.
Law of recency: the most recent response is likely to govern the recurrence.
Law of exercise: stimulus-response associations are strengthened through repetition.
COSC 4126 learning - D Goforth
Claimed advantages of automation
restore one to one attention immediate feedback individualized pace of progress
improve quality have ‘best’ teachers prepare machine
materials
COSC 4126 learning - D Goforth
Machine based learning Sidney Pressey automated testing machine for memory
research multiple choice questions subject presses buttons for choices machine tallies results
revised machine for learning (~Thorndike) machine advances with right answer only learner tries again if wrong
COSC 4126 learning - D Goforth
B.F. Skinner -behaviourism
Psychologist B. F. Skinner believed it
might be possible to
teach arithmetic more efficiently using machines
like those he devised for
teaching animals.
knob cannot be turned until answer is correct
COSC 4126 learning - D Goforth
B.F. Skinner
advanced version with trustWhile not all answers punched // repeat sequence
While more frames // do all framesShow next frame and questionStudent enters answerStudent covers answer with clear windowReveal correct answerIf student answer correct
student punches hole beside answer
COSC 4126 learning - D Goforth
Programmed learning
based on Skinner’s principles (similar to Thorndike)
content is broken into very small steps so new knowledge is ‘obvious’
repetition reinforces knowledge learner is seldom wrong, always
‘rewarded’ (motivation)
COSC 4126 learning - D Goforth
Example – frame 1
MANUFACTURE means to make or build. Chair factories manufacture chairs.
Copy the word here:
_ _ _ _ _ _ _ _ _ _ _
B.F. SKinner. 1958. Teaching Machines. Science 128(3330), 969-977
COSC 4126 learning - D Goforth
Example – frame 2
Part of the word is like part of the word FACTORY. Both parts come from an old word meaning make or build.
M A N U _ _ _ _ U R E
COSC 4126 learning - D Goforth
Example – frame 3
Part of the word is like part of the word MANUAL. Both parts come from an old word for hand.
_ _ _ _ F A C T U R E
COSC 4126 learning - D Goforth
Example – frame 4
The same letter goes in both places.
M _ N U F _ C T U R E
COSC 4126 learning - D Goforth
Example – frame 5
The same letter goes in both places.
M A N _ F A C T _ R E
COSC 4126 learning - D Goforth
Example – frame 6
Chair factories
_ _ _ _ _ _ _ _ _ _ _
chairs.
COSC 4126 learning - D Goforth
Non-linear versions ofprogrammed learning
branch to different content based on answer to a question Norman Crowder. 1960. The Arithmetic
of Computers. Doubleday and Co. TUTORTEXT
COSC 4126 learning - D Goforth
Computer-basedprogrammed learning
use computer to evaluate answers determine question to present next
this form of CAI still exists today WebCT TLM http://webwork.math.rochester.edu/
research goal – intelligent tutoring systems
COSC 4126 learning - D Goforth
Intelligent tutoring systems
learner
content expert
intelligent tutor
repeat
1. tutor gets content from expert
2. tutor presents content to learner
3. tutor evaluates learner response
4. tutor updates model of learner
5. tutor plans next interaction
model of learner
COSC 4126 learning - D Goforth
Programmed learning
psychological basis: behaviourisminternal structure of brain/mind is
unknowable study human behaviour as ‘black box’ study scientifically by controlling
stimulus and studying response show learning as changed response to
same stimulus
COSC 4126 learning - D Goforth
Piaget and Cognitivismopen the black box
make theoretical models of how mind works
strong interaction between computing and psychology mind as information processor input/perception, output/action short-term memory, long-term memory
artificial intelligence
COSC 4126 learning - D Goforth
Cognitivism – learning principles
learner needs to experience, not just be told
learner constructs own ‘model’ of knowledge
learner learns when perception and model in long-term memory conflict
COSC 4126 learning - D Goforth
Cognitivism – learning principles
perception from senses
model in memory
conscious awareness
agreement
add to model
disagreement
revise model
new model
COSC 4126 learning - D Goforth
Cognitivism – implications for teaching and learning learner control of sequence active participation increasingly complex microworlds (LOGO) multiple points of view/representations metacognition simulations tools metaphors reflection – thinking about thinking
COSC 4126 learning - D Goforth
Cognitivism examplenaïve physics
teaching intuition of physics is hard even college students who can do the
math have wrong intuitionsball spinning on a string
cut the string;what’s the path?
(a) (b) (c)
COSC 4126 learning - D Goforth
White and Frederiksen Thinkertools
premise: ‘real world’ intuitions are too strong to overcome main problem is friction
interferes with understanding of force, motion, acceleration, momentum
hypothesis: build intuition with virtual experience ‘air puck’ world
COSC 4126 learning - D Goforth
LOGO programming language
first-person view of motion forward, back, left, rightwith respect to current position and heading
personal motion as metaphor for programming
metacognition – think about thinking