Michael L. Anderson, Ph.D.
	Department of Psychology	Institute for Advanced Computer Studies
	Franklin & Marshall College	Program in Neuroscience and Cognitive Science
	Lancaster, PA 17604	University Of Maryland
	michael -dot- anderson -at- fandm -dot- edu	College Park, MD 20742

What is Action-Grounded Cognition?

The past 50 years has seen two major competing approaches to the understanding of the mind.  The first, most closely associated with cognitivist schools of psychology and classic (“good-old-fashioned”) artificial intelligence, modeled the mind with the use of abstract symbols manipulated by formal, logical rules.   The criticisms leveled against this approach are by now quite well known, and can be summarized by saying that even the successful artificial systems built on these foundations were too abstract. They could “think” all right, in the sense of solve equations and make logical deductions, but they couldn’t do anything autonomously in the world, because their symbols were essentially meaningless to them, ungrounded in any experience. This is all cognition and no embodiment.

 The reaction against this failure was to model the mind as a much simpler transducer, tightly coupled to the environment and to the sensorimotor system, and acting as only one part of a much larger cognitive loop encompassing not just the brain of the agent, but its body and its environment.  This approach has had its successes, too, in that scientists have been able to build systems from the ground up—typically using connectionist architectures—that start and stay grounded in their environments and are able to do actual things with real input.  But there is growing awareness of the limits of this approach.  One might say that these systems are too grounded; although they typically display a certain degree of noise-tolerance, this kind of robustness is not equivalent to the flexibility of higher-order (human and animal) cognition.  They are rarely able to do anything other than the simple things in the particular environments for which they were designed, or in which they were trained. All embodiment and no cognition.

This research group is exploring a new approach to the understanding of the mind, called action-grounded cognition.  Action-grounded cognition combines the strengths of cognitivist and embodied/situated approaches to the mind, by commandeering the overall structure of the cognitivist approach, that is, rule-based transforms of representations, but replacing these abstract elements with their biologically-grounded counterparts. In this model of the mind, abstract symbols are replaced with action-grounded representations defined in terms of the situated perceptual-motor abilities of the agent, and abstract rules are replaced with specialized motor-based transforms (i.e. operations on affordances). This approach allows for both the flexibility of compositional, symbol-based approaches to cognition, and the physical grounding that is the strength of the perception-to-action transducer approaches.

Consider, for instance, that (just as would be predicted by an evolutionary account of cognition) there is strong empirical evidence for the involvement of the motor-control system in supporting higher-order cognition such as language understanding.  How should this be understood?  Action-grounded cognition treats motor control in terms of affordance processing.  Since affordances, the perceived availability of objects for certain kinds of interaction, aren’t just motor programs, but interpretations of the environment, this opens the possibility that the motor control system is also, already, a primitive meaning processor. This would offer one explanation for how it is even possible to leverage motor control to support and constrain higher-order processes like language understanding.  This is not just of great theoretical interest but has significant practical implications; understanding this phenomenon can help us to design a system that could be used both for motor-control, and to ground higher-order representations. Such a breakthrough would solve the symbol-grounding problem (among other things) and could lead to the development of new, more effective robotic control systems, with better integration between reasoning, perceiving, and acting.

What has grown out of Action-Grounded Cognition?

Many things, including a new theory about the organization and evolution of the brain based on neural re-use; new graph-based techniques for mining and analyzing fMRI data; an action-grounded theory of mental content/ intentionality; and more.  Please check out the projects page.