High-level perception, representation, and analogy: A critique of artificial intelligence methodology

Abstract

High-level perception -the process of making sense of complex data at an abstract, conceptual levelCis fundamental to human cognition.Through high-level perception, chaotic environmental stimuli are organized into the mental representations that are used throughout cognitive processing.Much work in traditional artificial intelligence has ignored the process of high-level perception, by starting with hand-coded representations. In this paper, we argue that this dismissal of perceptual processes leads to distorted models of human cognition.We examine some existing artificial-intelligence modelsCnotably BACON, a model of scientific discovery, and the Structure-Mapping Engine, a model of analogical thoughtCand argue that these are flawed precisely because they downplay the role of high-level perception. Further, we argue that perceptual processes cannot be separated from other cognitive processes even in principle, and therefore that traditional artificial-intelligence models cannot be defended by supposing the existence of a "representation module" that supplies representations ready-made.Finally, we describe a model of high-level perception and analogical thought in which perceptual processing is integrated with analogical mapping, leading to the flexible build-up of representations appropriate to a given context.

Generative Art Inspired by Nature, Using NodeBox

“…In a recent research project ("Gravital"), we have attempted to alleviate this shortcoming by providing a node-based interface. 4 Visual building blocks (nodes) can be connected in the interface to create interesting visual effects. Building blocks can be opened to examine and edit their source code.…”

Collaborative and self-generated analogies in science education

“…There are also pragmatic constraints, such as a person's awareness that searching for analogies might be a useful approach to solving a particular problem, and semantic constraints, such as the particularities of the concepts in the involved domains and commonalities of objects and their attributes across the domains. In addition, Chalmers, French and Hofstadter (1992) question the feasibility of exclusively cognitive approaches to analogies and analogical reasoning. Instead of seeing cognition as working in isolation, provided with sensory data from the perceptual system, they propose a more integrated view, where we recruit concepts and memories from our cognition in making sense of situations we encounter through high-level perception.…”

How we learn about things we don't already understand

“…To do so, PHINEAS depends on a previously available language of description well suited to qualitative descriptions of heat flow situations, but not obviously available ahead of time for a real cognitive agent. Furthermore, not only the description terms but also their generic similarity to analogous terms in water flow cases, are defined ahead of time for the system (This is hardly a novel comment: Falkenhainer specifically mentions the use of specialized feature languages as a prominent unexplained aspect of the model, as do (Chalmers et al 1992).) More recently, research on the ambitious Digital Aristotle project (Friedland et al 2004) relies on special-purpose languages designed specifically toward coverage of particular scientific domains.…”

“…Since we can presume that natural problemsolvers are not pre-equipped with special-purpose languages for, say, chemistry labs (though see Fodor 1975Fodor , 1992, then to the extent that these models accurately reflect the concepts that human specialists use to understand their domains, we must wonder where these special-purpose languages come from. While one might try to take cognitive models to task for simply assuming feature languages which are specially geared toward the task being modelled and thus simplifying the real task facing a natural agent (Chalmers et al 1992), we consider the use of specialized feature languages to be a very revealing instance of an intelligent design decision. The reason that models use specialized feature sets is, after all, that they make the task of learning much easier for the artificial agent.…”