Joseph Rychlak 1991.
An agent's action is free if and only if he could have done otherwise than perform that action. Understanding 'could have done otherwise' in teleological terms, this means that an agent is free if and only if he can change his goals. It is predicational behavior that allows one to change his goals, and thereby act freely. Since machines are not capable of predicational behavior, they are not free.
The Rychlak Argument
"The predicational model allows for this alternative [an agent's having done otherwise, all circumstances remaining the same], because the person in continually taking a position on life-conforming to what is the case or altering events based on what is eventually affirmed in one's line of reasoning.... The determination here is based, not on material-efficient, but on formal-final causation. The dialectically reasoned human being is always-from the birth of cognition!-affirming a course of predicated behavior that could have gone in opposition to the course it actually took. Inanimate objects, well described extraspectively in material/efficient-cause terms, never arrange such grounds for the sake of which their existence carries forward" (J. Rychlak, 1991, p. 104).
Rychlak, Joseph. 1991. Artificial Intelligence and Human Reason (New York: Columbia U. Press, 1991.)