How to Build Bridges between Intelligent Tutoring System Subfields of Research
Philip Pavlik Jr. and Joe Toth
Human Computer Interaction Institute, Carnegie Mellon Univeristy, Pittsurgh, PA 
Why is there this fragmentation? Particularly why are there so many oppositions?
Kuhnian paradigm shifting
More generally, relativism
Thomas Kuhn's Modest Proposal
Science proceeds by pardigm shifts
For example Aritotle's Physics --> Newton's Physics
One is not correct, they are just different
Science tends to shift betweeen these different paradigms in scientific revolutions as anomolies build up in old paradigms
For example behaviorism --> cognitive psycholgy
Kuhn argues that rival theories cannot be directly compared
Theories cannot be understood from the framework of another theory
Kuhn does suppose that some theories are better, but must be understood on their own terms
Seems like these beliefs encourage a camp mentality with no way to judge which camp explains anomolistic results better


And does it really make sense?
Corallary about Methods
Conclusions

Recipe for Lack of Progress
Can't we break down the mechanisms of each theory and compare them
Can't we use observation to see which makes more accurate predictions
Some, including the authors, would argue that these things are in fact possible
We can use the data from our senses (which shows remarkable stability across individuals) to compare the predictions of different theory components in specific situations

Implications of Kuhn
A strong seperation between reality and theory
Theories are relative
Reality is absolute

Davidson argues that this strong dualism of “scheme and content, of organizing system and something waiting to be organized”, implied in Kuhn’s theory of how ideas shift, is unintelligible
Sometimes ITS subdisciplines tend to ignore one another (e.g. Motivation and cognitive psychologies)
Sometimes they agree, but explain using different constructs (e.g. Cognitive and Neuroscience theories
Sometimes there is opposition (e.g. Behaviorism and Discovery learning theories)

ITS developer is left with sorting
out what's what

One solution is to ignore many subdisciplines during the design process
But this has possibly huge consequences if some students  are missing some crucial determinants of learning
Solution step 1: identify grainsizes (units of analysis in the ITS system). Grainsize is a universal feature of theoretical analysis.
Solution Step 2: Identify theories (and mechanisms within theories) that explain the units in your system
solution step 3:
align mechanisms for grainsizes and speculate on whether there is combination, integration or conflict
Three possibilities
combination
Integration
Reconciliation
Reconciliation is about bridges
Illustration of consequences
  
How do we do experiments if we need to vary everything?
We don't need to vary everything
We need to address everything
Just as with regular control of variables strategy, we only vary one or a few factors
In essence, we are  adovcatiing backwards elimination rather than forward selection
WHY?
How?
A possible source of the problem?
This theory explains pretty well why large tests of ITS (e.g. by the Institute of Educational Sciences in the United states) often fail
Unless we suppose the original studies of efficacy are bogus, there MUST by interacting factors blocking effects in large tests
It seems that only by determining the missing factors can we avoid these sorts of results in the future

How can we fix this fragmentation? How can we gain a unified understanding?
Why is there this fragmentation? Particularly why are there so many oppositions?
Direct Instruction

Authentic contexts
Exploraton contexts
Authentic contexts
Exploraton contexts
Scheduling
Testing

Scheduling
Testing