Modeling Changing Perspectives: Reconceptualizing Sensorimotor Experiences: Papers from the AAAI Fall Symposium
Georgi Stojanov, Bipin Indurkhya, Frank Guerin, Tony Veale, Cochairs
November 15–17, 2014, Arlington, Virginia
Technical Report FS-14-05
Softcover version of the technical report: $30.00 softcover
(For international orders please shipping options before ordering on website.)
Human perception is highly contextual: a perceptual stimulus can be viewed by a human in radically divergent ways depending on the context. In contrast, most AI approaches employ processes that creativity theorists would consider convergent, insofar as they search for a single best or optimal answer. This ability to fluidly change perspectives and dynamically reframe a stimulus is a key aspect of human perception and reasoning that we seek to understand in AI terms.
Divergent choices can be made at the boundaries of different representations and computational levels, so this symposium will focus on the representational basis of divergent thinking that allows humans to change perspectives and reconceptualize stimuli with ease. For cognitive systems need to interpret low-level experiences (for example, neuronal, physiological, sensorimotor) using high-level concepts (for example, belief, intention, identity). Recent advances in bottom-up machine learning allow computational systems to go from low-level sensor data up to useful higher-level features. However, humanlike cognition also requires a top-down process to meaningfully frame our perceptual experiences. Crucially, such top-down processes allow an agent to interpret an object or an experience in divergent ways. This divergence is closely related to imaginative play and may also be integral to modeling social phenomena like empathy, since empathy demands we adopt another's viewpoint and see things via a different lens.
This symposium will explore the following questions from the crossdisciplinary perspective of artificial intelligence, cognitive psychology and cognitive robotics:
1. How might we model divergently unconventional perspectives in a top-down fashion in robotics, AI or machine-vision systems? Though top-down approaches have been used in machine-vision from the get-go in model-driven approaches, the emphasis has been on establishing a convergent ground truth. This symposium will instead focus on how divergent departures from the convergent norm yield creative and playful reinterpretations. In play, an agent deliberately projects a conceptual organization onto an object for which it is not conventionally suited, so that overlooked properties become newly salient and thus suggest novel creative insights.
2. How are social attributes such as empathy, fairness, identity, cooperation and the self/other distinction anchored in the mechanisms of playful reframing and reconceptualization? How might a theory of mind emerge from these mechanisms?