Autonomy in Robots and Other Agents

The word “autonomous” has become widely used in artificial intelligence, robotics, and, more recently, artificial life and is typically used to qualify types of systems, agents, or robots: we see terms like “autonomous systems,” “autonomous agents,” and “autonomous robots.” Its use in these fields is, however, both weak, with no distinctions being made that are not better and more precisely made with other existing terms, and varied, with no single underlying concept being involved. This ill-disciplined usage contrasts strongly with the use of the same term in other fields such as biology, philosophy, ethics, law, and human rights, for example. In all these quite different areas the concept of autonomy is essentially the same, though the language used and the aspects and issues of concern, of course, differ. In all these cases the underlying notion is one of self-law making and the closely related concept of self-identity. In this paper I argue that the loose and varied use of the term autonomous in artificial intelligence, robotics, and artificial life has effectively robbed these fields of an important concept. A concept essentially the same as we find it in biology, philosophy, ethics, and law, and one that is needed to distinguish a particular kind of agent or robot from those developed and built so far. I suggest that robots and other agents will have to be autonomous, i.e., self-law making, not just self-regulating, if they are to be able effectively to deal with the kinds of environments in which we live and work: environments which have significant large scale spatial and temporal invariant structure, but which also have large amounts of local spatial and temporal dynamic variation and unpredictability, and which lead to the frequent occurrence of previously unexperienced situations for the agents that interact with them.