Evolving Perceptual Categories - a podcast by MCMP Team

Cailin O'Connor (UCI) gives a talk at the MCMP Colloquium (15 July, 2013) titled "Evolving Perceptual Categories". Abstract: Do perceptual categories--green, cool, sweet--accurately track features of the real world? If not, are there systematic ways in which perceptual categories fail to latch onto real world structure? Attempts to answer these questions have persistently led to a further question. Given that human beings can only observe the world through the lens of our perceptual systems, how is it possible to know whether and in what ways perceptual categories are veridical? In this talk, I use tools from evolutionary game theory to attempt to gain traction on this problem. In particular, I employ signaling games to model perceptual signaling and elucidate how and why perceptual categories may or may not track real world structure. Jager (2007) introduced sim-max games, a variation of the standard signaling game where the states of the world are modeled as bearing similarity relationships to one another. This added structure is manifested in payoffs that reward approximate coordination between the sender and receiver as well as perfect coordination. This altered payoff structure appropriately models many situations in which perceptual signals evolve. Jager (2007) showed that in the long run actors in sim-max games evolve strategies that categorize similar states of the world together and dissimilar states of the world separately. When applied to perception, these results would seem to indicate that perceptual categories are natural or veridical in that similar real world objects should be expected to evolve to be part of the same perceptual category. However, this conclusion is not merited when one takes into account how similarity is built into these models. Similarity is manifested in the payoff structure alone. What this means, as I will argue, is that one should expect real world states where the same actions are successful to be categorized together perceptually, and real world states that cannot to be categorized separately. In other words, perceptual categories should be expected to track real world structure inasmuch as payoff to organisms tracks real world structure. As I will argue, this conclusion should not lead to a strong anti-realist stance with regard to perceptual categories. Whenever organism payoff is systematically related to the natural structure of the world, perceptual categories should be systematically related to this structure as well. What this means is that the relationship between perceptual categories and real world structure may be subtle and complex.

Further episodes of MCMP – Philosophy of Science

Further podcasts by MCMP Team

Website of MCMP Team