Brooks - behaviour languages

Possibly the most vocal critic of the symbolic AI notion of agency has been Rodney Brooks, a researcher at MIT who apparently became frustrated by AI approaches to building control mechanisms for autonomous mobile robots. In a 1985 paper, he outlined an alternative architecture for building agents, the so called subsumption architecture [Brooks, 1986]. The review of alternative approaches begins with Brooks' work.

In recent papers, [Brooks, 1991a][Brooks, 1991b][Brooks, 1990], Brooks has propounded three key theses:

1. Intelligent behaviour can be generated without explicit representations of the kind that symbolic AI proposes.
2. Intelligent behaviour can be generated without explicit abstract reasoning of the kind that symbolic AI proposes.
3. Intelligence is an emergent property of certain complex systems.

1. Situatedness and embodiment: `Real' intelligence is situated in the world, not in disembodied systems such as theorem provers or expert systems.
2. Intelligence and emergence: `Intelligent' behaviour arises as a result of an agent's interaction with its environment. Also, intelligence is `in the eye of the beholder'; it is not an innate, isolated property.

If Brooks was just a Dreyfus-style critic of AI, his ideas might not have gained much currency. However, to demonstrate his claims, he has built a number of robots, based on the subsumption architecture. A subsumption architecture is a hierarchy of task-accomplishing behaviours. Each behaviour `competes' with others to exercise control over the robot. Lower layers represent more primitive kinds of behaviour, (such as avoiding obstacles), and have precedence over layers further up the hierarchy. It should be stressed that the resulting systems are, in terms of the amount of computation they need to do, extremely simple, with no explicit reasoning, or even pattern matching, of the kind found in symbolic AI systems. But despite this simplicity, Brooks has demonstrated the robots doing tasks that would be impressive if they were accomplished by symbolic AI systems. Similar work has been reported by Steels, who described simulations of `Mars explorer' systems, containing a large number of subsumption-architecture agents, that can achieve near-optimal performance in certain tasks [Steels, 1990].