CogOnt Seminar: U. Feest/A. Stocco

Part of our ongoing online seminar series.  See the full list of talks here.

Register using this link: https://pitt.zoom.us/webinar/register/WN_KMNKu4fmQ9Wh5ZjvXJ3qQA

Please note, registration will be for the entire seminar series.

Uljana Feest (Leibniz Universität Hannover), “Cognitive Kinds and Investigative Practice”

ABSTRACT:

When psychologists investigate their objects of research, such as (kinds of) memory, they operationally define these “objects” in terms of particular experimental tests/tasks, assumed to provide epistemic access to the objects in question. For example, they might treat priming tests as allowing for the experimental investigation of implicit memory. In doing so, they take advantage of conceptual assumptions about their objects. But what is the ontological status of such objects? Are they cognitive kinds? And if so, what kinds of things are cognitive kinds? In my talk I will argue that cognitive kinds are cognitive-behavioral whole-organism capacities, which are comprised of multiple phenomena, including (but not limited to) behavioral phenomena. With this I depart from the assumption that the behavioral criteria by which cognitive kinds are empirically individuated are mere epistemic vehicles that aid in the investigation of cognitive kinds. Rather, they are part of what it is to be such a kind. While I take cognitive kinds to be sustained by the causal structure of the world, I argue that they are not uniquely determined by neural mechanisms. My account of cognitive kinds is relational in that I claim that cognitive kinds are constituted relative to our sensory-conceptual apparatus and maintained by our causal practices surrounding cognitive kinds. While this analysis is conventionalist, I will argue that this does not imply an antirealism about cognitive kinds.

Andrea Stocco (University of Washington), Title TBA

Co-Authors: John Laird (University of Michigan), Christian Lebiere (Carnegie Mellon University), Paul Rosenbloom (University of Southern California)

ABSTRACT:

Historically, the cognitive neurosciences have drawn from three approaches to define their ontologies: (1) constructs from cognitive psychology, which are usually derived from top-down intuitions about the overall organization of cognition; (2) computational models, which produce well-defined formal models of specific processes; and, more recently (3) large-scale neuroimaging data, which can be used to argue for the identification of large brain architectures from patterns of connectivity. These approaches have yet to converge towards a unified and agreed-upon set of constructs.

Here, we describe an integrative approach that leverages the strengths of all three. First, a

consensus model was derived from an analysis of successfully developed cognitive architectures. Focusing on functioning architectures reduces the variability associated with cognitive constructs and provides a preliminary list of “tried and true” mechanisms for human-level cognition. This “Common

Model of Cognition” (CMC) is based on a minimal list of five components (perception, action, working memory, declarative memory, and procedural memory) and their interconnections. To partially validate the CMC, we analyzed fMRI data from 200 participants from the Human Connectome Project, using tasks that cover a representative range of cognitive domains (language, mathematics, working memory, relational reasoning, social cognition, emotional inference, and decision-making). The CMC components were identified with functionally homologous brain regions using an iterative procedure that, starting with large-scale, a priori localization assumptions, (e.g., working memory is mapped to fronto-parietal regions), proceeded through the identification of task-specific and individual-specific regions of interest.

Their communication pathways between components were then translated into predicted patterns of effective connectivity. The resulting dynamic model was implemented and fitted using Dynamic Causal Modeling and compared against alternative architectures using a Bayesian approach. We suggest that a successful ontology should explain brain activity equally well across domains (generality) and better than any other model, even specialized ones (comparative superiority). In fact, our results show that, in all cases, the CMC outperforms all other network architectures, both within each domain and across all tasks.

Thus, we conclude that the CMC provides an existence proof of how cognitive, computational, and neuroscience approaches can be integrated and evidence that a minimal cognitive ontology might exist.