663_AIPsycPresentation

Symbiosis in Teaching AI to Psychology Undergraduate Students: A Case Study : 

Symbiosis in Teaching AI to Psychology Undergraduate Students: A Case Study Dr Jonathan Catling & Dr Colin Price University of Worcester

Introduction : 

Introduction ‘Using AI in teaching other courses’ Specifically focussing on the undergraduate Psychology programme at the University of Worcester Using AI to highlight strengths and weaknesses of the human brain and behaviour. Using specific examples and research papers from our teaching of Cognitive, physiological and developmental Psychology to demonstrate this symbiotic relationship.

Human Language Acquisition : 

Human Language Acquisition Traditional teaching of Language acquisition is often rather dated So…more contemporary ideas are introduced critical periods of development; brain maturation “Age of Acquisition” effects Age of Acquisition: The effect of Age of Acquisition is such that information that is encountered and learned early in life is accessed faster and more accurately than information learned later in life. In essence the earlier in life we are subjected to, and code, this information, the more effective we are in retrieving it. This is true for a variety of lexical processing tasks, including object naming and word naming

Age of Acquisition Effects : 

Age of Acquisition Effects Explanation of Age of Acquisition Effects: Critical Periods for language acquisition. Socially isolated children, Studies of patients who suffer cerebral damage at various ages also provide evidence that prognosis for language recovery is much more positive early in life as opposed to post-puberty (e.g., Duchowny et al., 1996; Bates 1992). Studies of second-language learning Reduction of plasticity: a biological process OR A function of information being entered into a network.

Computer Simulation : 

Computer Simulation Ellis & Lambon Ralph (2000) Standard three-layer backpropagation network Input and output layers of the network each contained 100 units which were fully connected to an intermediate, hidden layer containing 50 units. The input representations were randomly generated, binary patterns distributed across all 100 input units. The 200 input-output patterns were divided into two sets, 100 early and 100 late.

Training Schedule : 

Training Schedule Ellis & Lambon Ralph (2000) 100 early patterns were first trained for 250 epochs (where one epoch involves a single presentation of each pair of input-output patterns). 100 late patterns were then added to the early patterns in the training set and the model was then trained on the combined corpus of 200 patterns. Analogous to human vocabulary acquisition….? Performance on the early and late sets of training patterns was assessed in terms of the mean pattern sum-squared error. This measure reflects the difference between the network's output and the ideal response. The closer the value is to zero, the better the performance of the model.

Computer Simulation Results : 

Computer Simulation Results

Computer Simulation Results : 

Computer Simulation Results Age of acquisition effects are a natural property of connectionist models trained by backpropagation. Age of acquisition effect reflects a gradual reduction in network plasticity.

Physiological and Developmental Psychology : 

Physiological and Developmental Psychology One aspect of developmental and physiological psychology that is regularly taught to psychology undergraduates is that of brain development in early childhood.  Apoptosis (Programmed cell death) Synaptogenesis (Synaptic pruning) In the human visual cortex alone, it is estimated that around puberty, synapses are lost at a rate of around 5000 a second. Basically, over time the brain becomes more hard-wired

Computer Simulations : 

Computer Simulations The fact that programmed cell death occurs, suggests that it gives us a developmental advantage.  However, the average undergraduate student finds it extremely difficult to conceptualise how the death of millions of neurons can in anyway benefit the developing child.  This again is where AI (and again specifically a study of NNs) can be of benefit to learning and teaching.

Computer Simulations : 

Computer Simulations Students train a backpropagation NN on a simple task and then by pruning can interactively observe that having fewer connected nodes need not mean an increase in error rates, but can actually decrease error rates. Pruning of NNs can increase the efficiency and performance of the network.

Physiological and Developmental Psychology (2) : 

Physiological and Developmental Psychology (2) NNs can be used to explain recovery from brain damage and neuron redundancy.  The immature brain is characterised by ‘plasticity’, i.e. it is capable of functional reorganisation in response to damage or injury.    For example, children are often able to recover language skills after sustaining a large injury or even a hemisperectomy on the speech-dominant side, providing the damage happens before the age of 7 or 8.

Computer Simulations : 

Computer Simulations These regenerative processes after damage can be simulated within a NN.  Large proportions of the network can be destroyed without affecting the correct functionality.  Reinforces the concept of distributed networks over serial processing. Helps to illustrate the redundancy found in the human brain.  Network then retrained on the same task The changing of synaptic weights is seen to be analogous to neuroregeneration within the brain, and the relearning of functionality is analogous to the relearning of lost tasks by the brain.

Human and Machine Intelligence : 

Human and Machine Intelligence At a theoretical level one way that we have taught aspects of ‘human intelligence’ to psychology undergraduates has been to turn to ideas from the research into machine intelligence.  Students are asked whether or not certain machines can be defined as intelligent and through this process they are challenged to develop their own working definitions of intelligence.  Students introduced to concepts of weak and strong AI and the Turing test.  Students then develop a ‘better’ test….

Human and Machine Intelligence (2) : 

Human and Machine Intelligence (2) Problem solving and intelligent behaviour.   AI is useful to illustrate approaches to problem solving, concepts which most students seem to take for granted. Means-ends analysis is introduced to our students through the Fox, Chicken & Grain puzzle.  Simple Means-ends analysis will never completely solve the problem. But students can solve the problem… Why? Good way to introduce the concepts of linear algorithms, search space and heuristics into our learning and teaching.

Advantages of using AI in teaching Psychology : 

Advantages of using AI in teaching Psychology Theoretical and practical reasons for adopting this approach to learning and teaching.  At the theoretical level there are a number of educational advantages in using these methods: Learning by the use of analogy or metaphor.  Learning within a Piagetian framework, there is nothing better for learning than playing and experimenting oneself.  At the practical level these methods of teaching are student-centred based on Student feedback.

Conclusions. : 

Conclusions. To conclude, it is the hope that these illustrations of how we incorporate AI to help in the teaching of psychology at undergraduate levels has reinforced the connections and symbiotic relationship between the two disciplines both at an intellectual level and more importantly in this instance within the educational sphere.