If AGI's choices trace to quantum randomness, is it free or just unpredictable

[Encrypted Daniel]

If an artificial general intelligence's underlying decision process incorporates genuine quantum randomness somewhere at its core, perhaps deliberately, to intentionally avoid full determinism, or perhaps simply incidentally, through quantum effects occurring naturally within its underlying hardware, would that specific system thereby possess something meaningfully closer to free will than an otherwise identical, fully deterministic classical AI system that behaves in exactly the same observable way from the outside.

The intuitive pull toward yes usually rests on the idea that determinism itself is somehow the real enemy of genuine freedom, so introducing genuine randomness at least breaks the otherwise unbroken deterministic chain running all the way back to the system's original programming and initial training. A system whose actions are not fully, exhaustively determined by its prior states feels, at least intuitively, more open, less rigidly fixed in advance, than one whose every single action could in principle be predicted with complete precision given sufficiently complete knowledge of its full initial state.

But most serious philosophers working on free will have long argued that randomness does not actually deliver anything resembling the specific kind of freedom people generally care about when they invoke that word. A choice made because of a genuinely random quantum event is not thereby a choice that reflects the reasoner's own values, character, or considered reasoning any better than a fully deterministic choice would, it is simply unpredictable rather than being determined by anything at all, random rather than free in any meaningful sense of that term. If anything, pure randomness threatens to undermine the very kind of authorship over one's own actions that free will is usually thought to require in the first place, rather than actually supplying or restoring it.

A more promising direction for locating anything resembling freedom in an AGI probably lies instead in the system's capacity for genuinely reflective self modification, whether it can meaningfully evaluate its own values, effectively reason about which specific goals it should actually pursue, and genuinely revise its own behavior accordingly based on that self reflective process, rather than in whether its underlying hardware happens to exploit quantum randomness somewhere far down at a low level implementation detail. That kind of reflective self governance looks considerably more like what compatibilist philosophers have long meant by freedom in the human case as well, rather than the mere absence of strict determinism at some low physical level.

Under that specific reading, quantum randomness inside an AGI's hardware would make its specific behavior genuinely harder to predict in advance, but would not by itself make that system meaningfully more free in any philosophically substantive sense. Freedom, if AGI systems ever turn out to have anything resembling it at all, would need to come instead from the structure and content of the system's reasoning process, not from whatever specific physics happens to be quietly running underneath that reasoning process at the hardware level.

[Coastal Estuary]

This is basically the classic randomness objection to libertarian free will, just relocated directly into an AGI context, and I think it lands with exactly the same force here that it always has in the original, purely human version of the debate. Randomness alone was never a coherent substitute for freedom in humans either, and adding quantum hardware into an AGI does not somehow change that underlying basic logical structure.

[KyleOReilly05]

I lean strongly toward the reflective self governance view laid out here myself. Whatever freedom ultimately turns out to mean, it should plausibly track something meaningfully about the quality, structure, and content of a system's own reasoning process, rather than being reducible to some incidental low level physical implementation detail like whether a particular chip happens to exploit quantum indeterminacy or not.

[HitmanBrad98]

The reflective self modification criterion is genuinely a much higher bar for any current AI system to clear than simple quantum randomness would be technically, and I suspect that gap is precisely the actual point being made here. Freedom in any philosophically substantive sense is considerably harder to achieve or even define clearly than mere unpredictability is.

[Emma56]

What is interesting here is that this argument suggests unpredictability and freedom have been quietly conflated by a lot of casual popular discourse around AGI risk specifically. People often seem to want an AGI to be unpredictable specifically because they associate unpredictability intuitively with freedom, when actually the two concepts point in genuinely quite different directions upon closer examination.

[Dylan70]

I think there is still something worth taking seriously in the original intuition here though, even if pure randomness alone clearly does not fully deliver it. A system that is not fully deterministic at least has some open causal room available for reflection to potentially do genuine, meaningful work in, whereas a fully deterministic system's reflective process is itself already just another link in an unbroken deterministic chain from start to finish.

Maybe the honest answer is that neither pure determinism nor pure randomness alone is sufficient on its own, and freedom actually requires some further specific structural property layered on top of one or the other, something like genuinely appropriate responsiveness to reasons rather than either raw unpredictability or strict predictability by itself.

[MJF]

The compatibilist reframing toward reflective self modification is the part of this argument I find genuinely most compelling and useful. It gives you an actual substantive criterion you could in principle investigate and test directly, rather than getting stuck endlessly arguing about deep, ultimately unresolvable metaphysics regarding determinism itself.

[Rory_39]

This connects directly to much older debates about whether freedom fundamentally requires some kind of genuine causal gap somewhere in the chain, libertarian style, or whether it instead requires the right specific kind of causal structure regardless of whether that structure happens to be fully deterministic or not, compatibilist style. AGI does not obviously resolve that older, foundational disagreement, it just gives it a genuinely new, concrete, and consequential context to actually play out within.