Computational Modeling, Autonomous Robots, & Embodied Cognition
Philosophy 321h, Bates College
This course explores current research models in embodied cognition and artificial intelligence that use simulations, robotics, and genetic algorithms to explore the kids of flexible and adaptive behaviors constitutive of our concept of intelligence. These approaches provide alternative models for intelligent behavior that challenges traditional representational and computational theories of mind. Standard computational theories of mind model intelligent behavior on the human capacity for rational decision making and general problem solving. On this account thoughts are constructed from abstract symbols that represent aspects of the environment, and minds are treated as general purpose symbol manipulation systems that can be realized in any of a potentially infinite number natural organisms or artificial systems. Although the computational model of mind has been a powerful research tool in cognitive science, it has proven difficult to implement in computer simulations and mobile robotics. For instance, general purpose problem solvers fall prey to what is called the frame problem in artificial intelligence: they have difficulty filtering task salient information out of noisy signals and so often follow inefficient procedures or get stuck in blind alleys. This and other difficulties have inspired researchers in embodied cognition and artificial life to look towards insect and animal models of intelligent behavior for alternatives.
Our bodies have evolved in lock step with cognitive systems, and both can be thought of as adaptations fine‐tuned to the kinds of environmental features necessary to help satisfy an organisms’ basic needs and interests. As a result the general structure of an organism's body (i.e., its effectors and the structure and placement of its peripheral sensory organs) is a strong constraint on the ways it acquires, manipulates, and uses information from its environment – a constraint that focuses cognitive systems on task salient information in the environment, simplifying the computational demands of flexible and adaptive behavior. Researchers in embodied cognition and artificial life therefore challenge the assumption that minds are general purpose symbol manipulation systems . They argue instead that intelligent behavior emerges from the interaction between (evolved and well adapted) bodies and the environment. To this end research in autonomous (sometimes called agent‐ or behavior‐based) robotics is used to explore the role of agent‐environment interactions in the production of intelligent behavior.
Embodied theories of cognition have proven to be equally powerful research tools. However, one can question whether the kinds of adaptive behaviors that can be successfully captured by insect and animal models have the structure necessary to explain the full range of human cognitive behaviors that interest us (e.g., counterfactual reasoning processes involved in long term planning). At the very least these types of processes seem to require abstract concepts that enable organisms to consider alternative strategies, model potential environmental change, and represent novel outcomes. This suggests that rather than think of embodied and computational approaches as mutually exclusive alternatives, perhaps it is better to argue that they compliment one another and can be used together to produce a more comprehensive theory of cognition. The course explores this dialectic. Readings are drawn from contemporary sources in philosophy, psychology, neuroscience, and computer science. In addition to course readings and written assignments, students use a range of computer simulations and robotics exercises to explore the ideas introduced in class.
see Rolf Pfeifer's TEDX Talk, "How The Body Shapes the Way We Think"