Perception and cognition are traditionally considered as sources of knowledge. Consider the opening paragraph of Marr’s book on Vision (Marr 1982), which says “Vision is the process of discovering from images what is present in the world, and where it is.” According to Marr, the role of vision is to acquire knowledge about the geometric layout of the external world. The most important questions (and most productive from a research point of view) have always revolved around the relationship between knowledge and cognition; some of these questions are: what representations underlie our capacities for knowing? What makes a representation veridical? What are the origins of our representations?
Much progress has been made on these questions within — roughly — the computational account of the mind, which assumes that knowledge is a result of computation on representations. In the last two decades, alternate models of the mind such as those considered in embodied cognitive science have challenged the older computational view, and in some cases, have also called for a non-representationist account of perception and cognition. While I sympathize with theories of embodiment and enaction, I think it is premature to go straight from computation to embodiment. Instead, I view cognitions as the primary constituents of perception, beliefs and emotions, i.e., the replacement for “computation on representations.” Instead of knowledge being the outcome of computation on representations, knowledge is a product of cognitions.
The approach I am taking is not that different from the classical Indian philosophers, for whom knowledge is grounded in pramana, i.e., valid cognition. The approach taken here also derives inspiration from recent research on language and conception — for example, in cognitive linguistics — that has been moving away from viewing cognition and perception solely as sources of knowledge. These researchers have emphasized the metaphorical, ‘loose’ structure of concepts and the crossover between representations across semantic fields (Lakoff & Johnson, 1999; Talmy 2000) rather than their formal or modular structure. Perhaps it is now time to invert the relationship between cognition and epistemology and to ground knowledge in cognition rather than the other way around. Further, I also view cognitions are prior to computations. In some cases, computational schemes might model aspects of cognition, this is no more a computational theory of mind than classical physics is a calculus theory of matter.
It is the logic of cognition that one needs for understanding the mind rather than the logic of computation. Coming back to the Indian pramana systems once again, the Indian philosophers understood pramana’s as material entities; it may turn out that the pramana theory will deepen our understanding of the body and embodiment rather than reducing embodiment to the “gross body” that walks and talks. At this stage it is too early to ask for a full theory of embodiment; pramana is a better bet for a model of the mind. My first paper on this model, “Indian Cognitivism and the Phenomenology of Conceptualization,” coauthored with Nirmalya Guha and Chakravarthi Ram-Prasad.