In “Free Will – Even for Robots“, John McCarthy describes what he feels are the minimal criteria for a complex system to be considered as possessing free will, namely the ability to state “I can, but I won’t”.
The decision process of such an entity is divided into two phases, starting with the consideration of all possible actions, i.e. what it can do, and then further searching within that set for what it wants most to do.This manner of breaking down the problem of choice is simply a strategy for building good adaptive goal-seeking systems. It is a strategy that works best when considerations of what can be done and what is wanted are well separated, either in terms of possible perturbations (factors mostly affecting one thing or another) or in terms of time-scale (can possibly changing faster than want). By exploiting this natural factoring, goal-seeking systems are able to adapt faster/more efficiently to changing/different conditions. On top of being structured internally like this, people also try to predict each other’s actions by assessing ability (can) and intent (want) separately. Again, it makes sense to do so because the inference and knowledge used to estimate the two vary greatly.
Incompatabilists, of which there are quite a few, hold that free will and determinism are in conflict. These are the people who drag Quantum Mechanics into the discussion. I find determinism to be irrelevant to the issue of free will – a person deciding to do something by flipping a coin does not seem any freer to me. The issue here is that of subjective uncertainty – when flipping a coin, not even the person knows what’s going to happen next, and hence that uncertainty is shared amongst all the agents, making it objective. A person with free will is one whose actions are unpredictable to others and yet can in retrospect be proven to have been fully decided on by the person internally. The question is, how does one furnish such a proof?