Early research attempted to instill intelligence into robots focused on automation via designer algorithms. Attempting to create intelligence behavior in this manner is akin to running up a clif. While some progress can be made, the heights remain far out of reach. Modern approaches mix in neural networks, genetic algorithms, and other variety of leraning systems. Learning systems are a major leap forward in being intelligent behaviro to machines, but they are generally hampered by the limted input, feedback and and achievable goals .
I propose a more synergetic aprroach between man and machine, to allow the robots to read our "thoughts", so to speak, and accordingly to do our bidding. Initially a robot "host" would control the movement of a robots limbs through mental and bodily interface. Using pattern recognizion netwoks, such as Bayesian networks, the robots would be able to generalize on this human input, preforming some specific task. These action patterns could then be encapsulated as unified behaviors, such as command to a robotic dog, "fetch paper".
Having the robot under self-control, readily available for general use to preform specialized tasks, is the ultimate achievement of industrial automation. It behooves us to put our own physical skills to work as a means of education of automated learning systems, rather than attempting to instill the behaviors or criteria via our crude means of verbal communictation.