Timescales in Natural and Artificial Intelligence
I study how people integrate information over time, as they seek to understand their environment and learn from it. Temporal integration is ubiquitous, because our world unfolds over time: hearing a fragment of sound, we perceive it as part of a mockingbird’s melody; reading one word, we understand it as part of a meaningful sentence. The first part of this talk will review empirical findings and computational models describing how regions of the cerebral cortex integrate new input with context from many seconds earlier; we will also examine how and when cortical regions “forget” prior context. The second part of the talk will explore how learning processes are affected by the processing timescales of our brains, and why certain kinds of information tend to “linger” in our minds.