It’s All in the Timing: Representing and Reasoning about Time in Interactive Behavior
Papers from the AAAI Spring Symposium
Frank Broz, Marek Michalowski, and Emily Mower, Cochairs
People do not experience the world solely as an ordered sequence of events. The timing of our perceptions and behaviors has as much of an impact on our experiences as the nature of the events themselves. Yet many of the representations currently used to model human behavior do not incorporate explicit models of the temporal expression of these stimuli or actions. Dynamic behavior is often modeled sequentially in such a way that its temporal resolution is reduced and potential non-stationarity is ignored for the sake of computational efficiency (as in Markov state-based models of behavior), or causal mappings between observations and behavior are simplified to mitigate the sparseness of available datasets. Given that any artificial agent designed to interact with people will be dealing with intelligent partners with rich mental representations of time, are we using the appropriate representations?
The issue of timing can be even more critical when designing natural interactive social behaviors for robots or other synthetic characters. Human social behaviors are extremely dependent on a close feedback loop of simultaneous and coordinated activity between multiple interactors. Yet how to best represent these interdependencies and temporal relationships in order to produce interactive behaviors is just beginning to be explored and understood from a computational perspective. Speed, acceleration, tempo, and delay are concepts that AI and robotics researchers recognize as important in everything from motor control to verbal communication, but we do not yet possess a well-motivated framework for how these temporal considerations should be designed into our systems.
This symposium focused on representations of human behavior, including the validity and sufficiency of the current representations of human behavior. It also looked to other communities to determine which representational ideas could be borrowed to create a more complete picture of human behavior. Finally, this symposium addressed a central HRI question: which experimental paradigms can serve as testbeds or eventual benchmarks for the study of timing behaviors?