http://www.youtube.com/watch?v=LRq_SAuQDec

You are certainly correct that as soon as a machine has any sort of sensory input, it is already beyond the realm of “straightforwardly running a program”.

While these types of robots are still relatively simplistic compared to say, the nervous system of an insect, they’re a long way from a robot running a BASIC program full of procedural instructions like “go straight”, “turn right”, etc. In the end, isn’t a Human just running a complex program resulting evolutionary algorithms? But we make ethical decisions, have moral responsibilities, and arguably, have free will.

I agree it’s still premature to claim any current military robots are making ethical decisions. But these robots are reacting to sensory input in complex ways comparable to simple life forms. My estimation is that the most advanced robots are probably equivalent to the level of complexity found in a one celled animal. Does an amoeba make ethical decisions? What about a reptile, a mammal, a primate?

Out of curiosity, what definition of “agent” are we talking about excludes non-biological entities such as robots? Roboticists use the word all the time; perhaps in a different context than philosophers?

Interesting thoughts about robots learning their own rules. It does seem plausible to me that any entity with human-style agency and moral responsibilty would have to have done that in some sense.

But I have a deeper question. Your approach here, Peter, seems to follow the GOFAI pattern of assuming that robots follow man-made rules. My claim is that a robot following only the rules that it had learned on its own, as we humans must necessarily do, would be a very different beast. Yes, as children, we are sometimes spoon fed with lists of “important” social rules. But the way we go about learning these, and whether or not to give them any importance in our lives, is very different from the way the imputed traditional AI-style robot follows rules by executing its program.

Of course, if the rules, as they were being learned, were stored in the form of program code, one must suppose that such rules would have to be followed in much the same way as a program that was fed in by the programmers. One claim would be that the kinds of rules that could be learned by a functioning, mobile entity, interacting socially, would be like nothing any programmer would ever write. Another claim would be that, even if the set of rules turned out to be essentially the same as a programmer would write, the way those rules get interpreted by the entity would result in very different behavior. This hypothesis rests on the concept that all the time while an organism (or an appropriately programmed machine) is behaving according to some scheduled plan of instructions, whether such a plan was downloaded or learned online, the organism would also be continuously alert to various ongoing world events. This awareness could change the course of action of the organism in ways that no programmer could foresee.

Can a robot really learn to manage in the world given that it must learn the rules on its own? So far, there is little evidence of this. But a project at Michigan State University (http://www.cse.msu.edu/%7Eweng/research/LM.html) seems to point in the right direction. Some of Rodney Brooks efforts at MIT might be said to be following such a trail, but his stated goals are not as broad and general as those of the MSU project. And, of course, learning is not the only thing we as humans do. We also inherit a very large and complex set of autonomic and emotional responses. These built-in factors obviously play large roles in determining our behavior. And even if we one day learn what all of these factors are and how they work, it is by no means given that we will put all or any of them in our robots.

The question is whether such a machine, if such is at all possible, would seem to us humans as being able to decide things for itself. My suspicion is that it would seem to be behaving independently, it would think and say that it did have free will, and it would behave in all the fuzzy and sometimes incomprehensible ways that we humans do. At the very least, we will learn a great deal about ourselves in the process of creating such machines.