Basically - the key to AI is in understanding the real world. Traditional AI (well, as it was then...) abstracts the real world details, providing a simplified version that is easier for the AI to understand.
Brooks is arguing that it is better to develop an AI which is capable of acting in a limited way, within a complex environment. i.e. that intelligence comes not from the decision making process, but in the ability to conceptualise and understand real world input.
Part of this argument is that it is incorrect to 'reject' sensory input as being a seperate module to the AI (in that the intelligence is developed, and that sensory input systems can be developed independently). A secondary issue is in that the environment itself determines intelligence - i.e. complex behaviour is caused by simple agent able to act within a complex environment.
