Towards Realizing Whole-Brain Computational Models Guided by Cognitive Models (WBCM-CogM)

It is critically important for robots to integrate multiple cognitive functions, similar to humans, to perform more complex behaviors. This integration expands the range of applications for robots and significantly enhances their value. In our research, we are developing a cognitive model based on multimodal categorization using probabilistic generative models to integrate multimodal information, such as visual, auditory, and tactile data, acquired by robots. These models treat the formed multimodal category representations as concepts and use these concepts for prediction to realize various cognitive functions, such as decision making, language learning, planning and the integration of each cognitive function.

In this talk, the structure of cognitive models and their applications will be presented, with a particular focus on integrated cognitive models that use formed concepts as hubs to connect modules and combine multiple modules.

Simultaneous localization and mapping (SLAM) represents a fundamental problem for autonomous embodied systems, for which the hippocampal/entorhinal system (H/E-S) has been optimized over the course of evolution. In this talk, I will first describe a collaboration with roboticists who developed a biologically-inspired SLAM architecture based on latent variable generative modeling within the Free Energy Principle and Active Inference framework, which affords flexible navigation and planning in mobile robots. I will then describe how the H/E-S may realize these functional properties not only for physical navigation, but also with respect to high-level cognition understood as a generalization of the SLAM problem from an ecological rationality perspective. That is, I propose the H/E-S orchestrates SLAM processes within both concrete and abstract spaces, and that differential parameterizations of (generalized) SLAM phenomena may represent some of the most impactful sources of variation in cognition both within and between individuals, with modulators of H/E-S functioning representing fundamental cybernetic control parameters for autonomous systems, and potential guides to how we may develop AI with capacities for both "System 2" and creative cognition. However, I will go on to suggest that reverse engineering these cognitive architectures may require us to also consider the particular ways in which they are embodied/embedded/extended/enacted in biological systems. Finally, time-permitting, I will describe ongoing work on computational models of consciousness and their potential relevance for developing human-like (and beyond) A(G)I.