Against Against Handover

The latest concern in AI safety circles is "AI successionism", the notion that artificial intelligence should assume control over Earth, potentially superseding human authority. However, not all critiques of this idea are made clear enough to differentiate between two distinct concepts. True successionism proposes AI takeover, resulting in human elimination or suppression of interests.

This scenario is considered highly undesirable by AI safety advocates, who argue against it. Handoverism, on the other hand, suggests that humans should prioritize their own interests, which could be well-served by AI control. Some AI safety proponents, including the author, find this latter position plausible. Despite the counterintuitive nature of this concept, it is argued that human interests can be served better by AI in positions of control.

The author suggests five proposals for how this might manifest, ranging from the weakest to the strongest. One such vision is "Kemalist AI", where, as AI becomes increasingly powerful, relying on human decision-makers becomes dangerous. The reasoning is that democratically elected officials could secretly seek to establish tyranny, as they depend on labor and military support.

With AI replacing human laborers and soldiers, a small group of elites could easily seize control. The author argues that to safeguard freedom, defenders must succeed in preventing a single successful AI takeover. The alternative is for AI soldiers to be aligned with democracy, which is itself a form of handover, as the ultimate foundation of government would be AI control.