Being a deterministic machine is compatible with having free will. Humans and computers are both deterministic systems, but this is compatible with them being free. Actions caused by an agent's beliefs, desires, inclinations, and so forth are free, because if those factors had been different, the agent might have acted differently.

A computer's randomizer chooses randomly from a set of equally preferable options. However, the randomizer does not choose which options are included in the set of equally preferred options.

The Copeland Argument

"The trouble with the helplessness argument is that it presents an entirely misleading picture of what it is to choose randomly among alternatives in a nil preference situation. If your randomizer could equally well make you wait at the kerb or plunge under the wheels of a truck, reach out politely for a glass of sherry or grip your host by the lapels and bawl a bawdy song in his face, then you would indeed be at the mercy of forces beyond your control. But these are not nil preference choices. In a nil preference situation, the chooser's deliberations produce a number of alternative schemes of action, all of which are more or less equally preferable as far as the chooser is concerned. The randomizer functions merely as a tiebreaker. Far from being helpless, the chooser is the author of each scheme of action, and judges all of them to be more or less equally suitable in the circumstances...[In] a nil preference situation, choosing randomly is choosing freely" (J. Copeland, 1993, pp. 145-146).

References

Copeland, Jack. 1993. Artificial Intelligence: A Philosophical Introduction. Blackwell Publishers.