The Cognitive Science and Artificial Intelligence Students’ Association is happy to announce the 8th iteration of the University of Toronto Interdisciplinary Symposium on the Mind (UTism), titled ‘What to Expect When You’re Unexpecting: The Cognitive Science of Disruption’.
Conference Abstract
When a cognitive system goes off the rails — when it steps outside of its typical functional constraints, or breaks out of a previously maintained framework — how does it get itself back on track? What are the mechanisms by which these systems cope with the forces that intervene upon its functionality? Do these coping mechanisms alter the structure and function of the system, and if so in what way? By analyzing potential answers to these questions, UTism 2018 seeks to uncover the importance and role of disruptions (expected and unexpected) in developing a fully integrated account for explaining cognitive systems, and how these accounts should be relevant in the way we approach matters regarding those systems.
Speakers
Laurie Ann Paul (Keynote)
Eugene Falk Distinguished Professor of Philosophy
University of North Carolina, Chapel Hill
Big life decisions are naturally framed using the first personal point of view, where we mentally simulate or imaginatively project different future lived experiences for ourselves. Such decisions are often based on judgments about what these subjective futures will be like. I explore the way that making transformative decisions from this perspective can put us in the position of regarding our future selves as epistemically and psychologically alien to our current selves. I then frame these sorts of radical epistemic shifts as personal epistemic revolutions: they are cases where a person undergoes a Kuhnian revolution writ small. I’ll close by drawing connections to work in cognitive science on intuitive judgments and simulation, and work in developmental psychology on discontinuities in conceptual development.
John Vervaeke
Lecturer, Departments of Psychology, Cognitive Science, and the BPMH Programs
University of Toronto
Is development just a continuous processes of Bayesian updating that is analogous to how science advances knowledge? This talk will argue that the Bayesian model, while powerful, cannot completely account for important aspects of developmental change. If we pay careful attention to the analogy then we will see that development also has discontinuous aspects analogous to Kuhnian revolutions and to the punctuations within biological evolution. This talk will argue that the cognitive analogues of such discontinuous change are plastic and bio-economic self-organization of sub-representational aspects of cognitive competence. These disruptions are key to development.
Jim John
Assistant Professor, Department of Philosophy and University College
University of Toronto
According to the increasingly influential Predictive Processing Theory (PP), the brain is a prediction error minimizer. Some critics of PP allege that it entails epistemological skepticism and, hence, a problematic theory of mind. I will argue that the concern about PP and skepticism is unfounded but that there is a related worry for the view, to do with the ancient “problem of perception,” that is more serious. I will conclude that as long as what is representational about PP is correctly understood, even this problem can be adequately addressed. The moral is that certain criticisms of PP, especially those made by some proponents of “4E” cognitive science, are based on a mistaken conception of the role of representation in the mind/brain.
Second, I will present new behavioral results from a multisession, contextual learning task that tests this prediction. This task requires participants to search for a visual target (the letter “C”) in a cluttered scene made up of the letters “I” and “F”. By systematically varying contextual signals within the scene that are informative of the target location, we can quantify contextual learning. Consistent with predictions of the divisive normalization model of autism, we find that adolescents with autism show diminished contextual learning compared to their typically developing peers.
The data further reveal two qualitatively distinct ASD learning profiles. The first resembles the TD learning profile, but with a delayed ability to disengage a previously learned contextual rule. The second fails to learn the context, from what appears to be a difficulty ‘seeing the forest for the trees’. Together, these results demonstrate how approaching autism as a “disorder of neural computation” can provide insights into the disorder’s underpinnings, and suggest that a synergistic combination of empirical and computational approaches can help guide the development of personalized treatment strategies.
Jennifer Whitson
Assistant Professor of Management and Organizations
UCLA Anderson School of Management
People are motivated to perceive themselves as having control over their lives. Compensatory control theory asserts that people will consequently respond to events and cognitions that reduce control with compensatory strategies for restoring perceived control; one such strategy for protecting perceptions of personal control is imbuing the social, physical, or meta-physical environments with order and structure. A series of experiments establish that people are more likely to engage in illusory pattern perception – i.e., the identification of a coherent and meaningful interrelationship among a set of random or unrelated stimuli – when they lack control. These illusory patterns range from the data-level (seeing patterns in the stock market that do not exist), to the causal (making superstitions connections between events), to the interpersonal (seeing members of one’s organization as conspiring together). Several lines of subsequent research examine other relevant drivers of illusory pattern perception, identify interventions that reduce the effect, and explore potential moderators.
Yang Xu
Assistant Professor, Departments of Computer Science and Cognitive Science
University of Toronto
Human language relies on a finite lexicon to express an infinite set of emerging ideas. One result of this challenge is that words tend to acquire novel meanings over time, e.g., gay (‘happy’->’homosexual’). The other way in which this challenge is met is by creating new words, e.g., skitch. Previous research has suggested that these time-varying processes of the lexicon may be non-arbitrary, but little work has explored their cognitive underpinnings in formal terms and tested those at scale. We present computational models that predict the emerging patterns of word meanings and forms, dating back hundreds of years in the English lexicon. Our results show that these processes are not only predictable, but they also tend to occur in ways that minimize cognitive effort.
Philip Monahan
Assistant Professor of Linguistics, Centre for French And Linguistics
University of Toronto
Successful interpretation of the dynamic spoken language signal appears to rely quite heavily on the propagation of top-down information flow and subsequent integration with bottom-up sensory cues. In this talk, recent magnetoencephalographic (MEG), electrophysiological (EEG) and behavioural findings are brought to bear on the role of this top-down information flow and the extent to which this knowledge is used in a predictive manner during speech perception and spoken word recognition. This work demonstrates that we make relatively specific predictions about the content of incoming linguistic information and that evidence for these predictive knowledge sources can be observed in the induced neurophysiological response prior to encountering the relevant exogenous stimulus.
In particular, listeners appear to use relatively abstract phonological and morpho-syntactic knowledge as the bases for these predictions and evidence for these predictions is evident in early brain responses. Specifically, I advocate for a model of linguistic comprehension wherein hypotheses about the upcoming signal are internally generated and tested against sensory information, e.g., Analysis-by-Synthesis models, and the source of these hypotheses is our rich linguistic knowledge. Disruptions in the incoming signal (either due to environmental sources or mismatches in our expectations) and language parsing are better handled via the exploitation of rich knowledge sources. Very recent work also points toward the consequences on the nature of the neurophysiological when these bottom-up sensory cues contradict with our expectations, potentially causing disruptions in speech comprehension.
Ari Rosenberg
Assistant Professor, Department of Neuroscience
University of Wisconsin-Madison
Neural circuits and the computations they perform bridge physiology and behavior. In this talk, I will discuss how understanding neural circuit function can provide insights into how changes in physiology produce behavioral consequences observed in mental health disorders. First, I will introduce computational work showing that physiological and perceptual consequences of autism can be parsimoniously accounted for by alterations in a nonlinear, canonical neural computation called divisive normalization. Divisive normalization balances the excitatory input to neurons with an inhibitory signal composed of neural population activity. Consistent with the widely held hypothesis of an increased ratio of excitation to inhibition in autism, I will present neural network simulations showing that a reduction in the divisive normalization signal accounts for both physiological and perceptual autism data reported in the literature.
This result implicates the context-dependent, neuronal milieu as a key factor in autism, with autism reflecting a less “social” neural population. An important behavioral prediction of reduced divisive normalization is that individuals with autism will show diminished use of contextual information in interpreting current sensory evidence. Second, I will present new behavioral results from a multi-session, contextual learning task that tests this prediction. This task requires participants to search for a visual target (the letter “C”) in a cluttered scene made up of the letters “I” and “F”. By systematically varying contextual signals within the scene that are informative of the target location, we can quantify contextual learning.
Consistent with predictions of the divisive normalization model of autism, we find that adolescents with autism show diminished contextual learning compared to their typically developing peers. The data further reveal two qualitatively distinct ASD learning profiles. The first resembles the TD learning profile, but with a delayed ability to disengage a previously learned contextual rule. The second fails to learn the context, from what appears to be a difficulty ‘seeing the forest for the trees’. Together, these results demonstrate how approaching autism as a “disorder of neural computation” can provide insights into the disorder’s underpinnings, and suggest that a synergistic combination of empirical and computational approaches can help guide the development of personalized treatment strategies.