Cultural Affordances in AI Perception
- Zachariah A. Neemeh, Department of Philosophy, University of Memphis, Memphis, Tennessee, United States
AbstractAffordances offer AI research an alternative from representations for linking perception to action in autonomous systems. Affordances are based in the informational structure of the environment and the somatic capacities of the agent and arise in the interaction between agent and environment. AI implementations of affordance perception typically utilize relatively basic, natural affordances such as the graspability of a handle. Culturally-scaffolded affordances, such as the letter-mailing capacity of a postbox, pose a more intractable problem for affordance-based robotics. This class of affordances requires acculturation and is highly culture-specific. AI perception has only rarely ventured into incorporating them. I begin by reviewing affordance perception and the difference between natural and cultural affordances. I then critically discuss implementations of cultural affordance perception in autonomous agents, many of which remain in a representational framework. Finally, I argue that AI affordance perception does not require a robust representationalism in order to implement cultural affordances.