Paul F. Bello, Naval Center for Applied Research in Artificial Intelligence


Paul F. Bello, Naval Center for Applied Research in Artificial Intelligence

Sage 4101

March 9, 2016 12:00 PM - 1:30 PM

Attention is a cognitive capacity that directs resources to a subset of available information. At least in the case of humans, attention allows for selectivity in processing, with focused-on items shaping both what we see and what we subsequently think and do. Attention is somewhat unique in that it can be directed outward at objects, agents, and events in the world, but also inward toward representations in the mind. In this way, it is a natural bridge between perception, cognition, agency and learning. Despite the seemingly central role attention plays in cognition, most modern computational cognitive architectures provide piecemeal treatments for a subset of attention's putative functions.

In this talk, I motivate and describe a new computational cognitive system called ARCA-DIA1 that places attention front and center. I argue for six general principles divorced from any particular implementation that I assume to be definitional of a robust theory of attention. Each principle is motivated by an analysis of empirical results across a variety of experimental paradigms including those that purportedly demonstrate the limits of human attention. I then discuss how each principle is realized either as representational or process/architecture-level assumptions in ARCADIA before providing a series of demonstrations of the system exhibiting human-like performance on a number of different tasks.

Finally, I conclude with a discussion of how ARCADIA is conceptually positioned among other cognitive systems and directions for future research.





Download the paper here. 


Add to calendar