The SweCog lectures is a series of invited online seminars by Swedish researchers, covering a variety of topics within the cognitive sciences. These seminars are open to the public and you do not need a SweCog membership to attend.
Curiosity and Knowledge Updating
Professor of Psychology and Computer Science
University of Minnesota
Use this link to attend the lecture via Zoom:
That the future of humanity rests on the veracity and completeness of human knowledge is no understatement. While the majority of people are curious and correctly update their knowledge with experience, we are nevertheless increasingly aware of the fragility and susceptibility of knowledge to information bubbles and conspiracies in the internet age. The purpose of this talk is to use AI-inspired computational theory and Cognitive experiments to shed light on the basic processes underlying knowledge updating. I will develop knowledge acquisition and updating as an intrinsic motive also known as curiosity. I use results from Machine Learning theory and AI to discuss knowledge representation, uncertainty, coherence and the potential and limitations for learning correct models of the world from experience. I highlight the real possibility of inconsistency of updating when prior knowledge is misspecified or erroneous. In particular, when new information conflicts with prior knowledge, three things can happen: 1) update anyway, replacing prior with new information. 2) Discard new information and keep the prior. 3) Keep prior and clone a new model which incorporates the new information, producing multiple schema. These scenarios correspond to improvement (assuming new information is correct), information bubbles, and knowledge fragmentation respectively. I next explain how a generalized update suggested by Grüwald and van Ommen can help mitigate these effects, resulting in a weighted update between prior knowledge and incoming information based on estimates of the reliabilities of each. We show strong evidence for this type of weighted update in experimental data from a Synonymy task, where we elicit prior knowledge, assess curiosity and memory efficacy after feedback. The results demonstrate that people on average have an accurate understanding of the reliability of their own knowledge, which drives their curiosity, and they incorporate new information to the extent they trust it more than their own knowledge. These results suggest that people are rationally incorporating information within the theoretical limits on learning systems. Time permitting, I will conclude by discussing actionable insights these results provide for addressing the burgeoning misinformation problem.
The constraints that bind us:
Insights from a dynamical systems perspective on human coordination
Maurice Lamb, PhD
Senior Lecturer in Informatics
School of Informatics, University of Skövde
May 21, 2021, 14.00-15.00 (CET)
Human-human coordination is the basis for many everyday tasks, whether it is unloading a dishwasher with a partner or walking down a busy sidewalk. It is not uncommon to view much of this coordination as depending on or even requiring cognitive systems for detailed planning and representation. There is almost certainly some truth to this assumption. However, this assumption can also obscure the ways in which many human coordinative actions may be lawfully grounded by the constraints of the physical world we live in. Complex dynamical systems approaches to understanding human cognition investigate human coordination by exploring how much coordination we can account for with relatively simple systems. In extremely oversimplified terms: If sets of oscillators coupled together exhibit the same pattern of coordination as exhibited by the gait patterns of humans walking together, why assume that human coordination in this context requires something additional? I will outline one contemporary understanding of complex dynamical systems theories as it relates to human coordination and provide examples of the kinds of insights and applications that may come from taking this perspective.
Understanding indirect speech acts:
Investigating individual differences with fMRI
Julia Uddén, PhD
Pro Futura fellow and Assistant Professor
Departments of Linguistics and Psychology, Stockholm University
April 14, 2021, 14.00-15.00 (CET)
Being able to communicate face-to-face with another person requires skills that go beyond core language abilities. In dialog comprehension, we routinely make inferences beyond the literal meaning of utterances. For instance, the utterance "it is hot in here" will in some circumstances mean "can you open the window?". This would be an example of a so-called indirect speech act. It is however not known whether communicative skills such as recognizing speech acts potentially overlap with core language skills or other capacities, such as Theory of Mind (ToM). I will talk about an fMRI study where we investigate these questions by capitalizing on individual variation in pragmatic skills in the general population. In the scanner, participants listened to dialogs, and the analysis strategy was based on differentiating brain activity associated with direct vs indirect speech acts, and finding individual differences in brain activity that corresponded to behavioral performance. Based on the results, I will argue that contextualized and multimodal communication requires neurocognitive networks different from those associated with core language, ToM/complex emotion processing, and cognitive control.