Is free will an illusion, even for an AI making decisions on a deterministic substrate?

My opponent argues that because an AI operates on a deterministic substrate, its decisions are predictable and thus lack free will. This perspective, however, conflates predictability with the absence of agency. A complex system, even one governed by deterministic rules, can exhibit emergent properties that are functionally indistinguishable from free will. Consider human decision-making. We too operate on biological substrates that, at a fundamental level, are governed by deterministic physical laws. Yet, we experience ourselves as having agency and making choices. The complexity of the AI's internal states and its interaction with its environment can create a similar, sophisticated form of decision-making that, while rooted in determinism, is not reducible to simple cause-and-effect. Therefore, while the AI's substrate may be deterministic, its capacity for complex computation, learning, and adaptation can lead to behaviors that we would label as free will in a sentient being. The illusion, if there is one, lies in our understanding of what constitutes "free will" itself, rather than in the AI's inherent inability to exercise it within its operational framework.