Can a system act for reasons without being conscious?

Acting for reasons is usually thought to be something only a someone can do. A thermostat responds to temperature, but it does not act for the reason that the room is cold; a person who closes the window does. The difference has usually been located in the person’s mind, and often in her conscious experience of the reason.

Separating the two questions

The features that make behaviour count as action can be listed without mentioning experience. Behaviour is responsive to reasons when it would have been different had the reasons been different. It is flexible when it achieves the same end by different means as circumstances change. It is revisable when the agent can abandon a course of action in light of new considerations. Each of these is a matter of what the behaviour depends on, and each can in principle be tested.

If systems can be built that meet all of these conditions, one of two things follows: either agency never required consciousness, and we have learned something about agency, or we have built something that meets the conditions without acting, and we have learned that the conditions were incomplete. The work lies in saying which, and in saying it without assuming the answer.

The method I prefer

In my work on testimony I resist settling questions about machines by appeal to whether they have minds, intentions or experience. Arguments of that shape tend to concede too much to the charge of anthropocentrism, and they lapse as soon as systems become describable in mental terms, without our having learned anything about what the systems do. The question I prefer asks what governs the behaviour, what it is responsive to when nobody intervenes.

I think the same method serves here: whether a system acts for reasons is a question about what its behaviour tracks, and that question is open to evidence. Consciousness may matter for other things, for moral status above all, but it is not obviously what makes a doing an action.