Thomas Shultz, Professor @ McGill University
  • Home
  • Research interests
    • Learning & development
    • Neural networks
    • Evolution
    • Cognitive dissonance
    • Problem solving
  • Publications
    • Learning & development
    • Neural networks
    • Memory
    • Evolution
    • Cognitive dissonance
    • Problem solving
    • Decision making
    • Commentaries
    • Blog posts
  • Research highlights
    • Resolving the St. Petersburg paradox
    • Spread of innovation in wild birds
    • Resolving Rogers' paradox
    • Evolution of ethnocentrism
    • Shape of development
    • Connectionist modeling
    • Neural networks
    • Symbolic modeling
    • Causal reasoning
    • Moral reasoning
    • Theory of mind
    • Development of humor
  • Editorships
  • LNSC
    • Current lab members
    • Professors from LNSC
    • Undergraduate awards
    • Funding
    • Join us
  • Contact
  • Links
  • Photo
  • Membership in Research Centres

Connectionist modeling

In collaboration with students and colleagues, I was able to achieve integrative computational models of three major areas of psychological literature: cognitive dissonance, discrimination shift learning, and conservation acquisition. Simulations conducted with Mark Lepper of Stanford covered all of the principal paradigms of cognitive dissonance, including insufficient justification via prohibition, forced compliance, and initiation; free choice; self concept; and emotional arousal. The simulations suggested a novel theoretical interpretation of dissonance reduction based on constraint satisfaction, covered the experimental data better than did classical dissonance theory, and generated predictions, some of which have been confirmed with people.

With former graduate student Sylvain Sirois, I modeled developmental changes in shift learning. Our model was the first comprehensive theoretical integration of the diverse phenomena in this area, and it suggested the novel theoretical interpretation that older children and adults differ in learning from preschoolers primarily because of spontaneous over-training. That is, older individuals learn more from the same exposure time than do young children, perhaps due to rehearsal, more extensive processing, or faster processing. Both continuous discrimination shifts (reversal and non-reversal shifts) and total-change discrimination shifts (intra-dimensional and extra-dimensional shifts) are covered by the model, and several model predictions were subsequently supported in experiments with children and adults.

My model of conservation acquisition was the first to cover several phenomena in that large literature, including natural conservation acquisition, sudden spurts, the problem-size effect, length bias, and the screening effect. This model represents the first coherent, mechanistic account of how perceptual and cognitive skills are integrated as conservation develops.

In other work, we produced successful models of balance scale stages, seriation acquisition, pronoun acquisition, infant habituation to artificial sentences, concept acquisition, early theory of mind, and integration of velocity, time, and distance cues for moving objects. These developmental simulations led to a theoretical contribution, with former graduate student and now Professor Denis Mareschal of Birkbeck, London, showing with constructive neural networks how it is possible to escape from Fodor's paradox (that nothing genuinely new can be learned and thus constructivist accounts of development such as Piaget's are logically impossible). In the case of the balance scale, one of the major benchmarks in the modeling of cognitive development, our model is the only one to naturally capture all of the stage progressions and perceptual effects seen in children. These developmental models are the first genuinely constructivist models of psychological growth, and they correspond well to newly emerging brain evidence on experience-driven neurogenesis and synaptogenesis. These networks grow by recruiting new hidden units and previously acquired networks into the learning network as needed.
Powered by Create your own unique website with customizable templates.