Giving AIs legal, surveilled assets creates a strong incentive for them to store resources in a visible system rather than secret crypto wallets. This makes their resources less flexible for rogue actions and actually reduces risk by making defection harder and more detectable.
The development of superintelligence isn't a monolithic AI goal. Existing human-level AIs would see superintelligence as a threat to their own economic value and existence, potentially making them allies with humans in managing or pausing its development.
Even if a superintelligence has an absolute advantage over humans in every task, it may still be economically rational for it to delegate work to them. This is due to comparative advantage, as the AI's time would have an extremely high opportunity cost, making human labor a bargain.
The proposal focuses on granting AIs specific economic rights (property, contracts, torts) to enable participation in the economy, while explicitly excluding rights to privacy and reproduction to maintain human control and societal stability.
Beyond technical alignment, which shapes an AI's internal goals, "cultural alignment" uses established human institutions like legal systems and markets to shape AI behavior. This leverages thousands of years of human experience in managing intelligent agents.
While giving AIs rights feels risky, the current plan is arguably more dangerous: building human-level intelligences, making them servants, and terminating them for minor errors, all based on unproven technical alignment methods. A cooperation model is a response to this high default risk.
Even with static weights, AIs can exhibit reciprocal behavior through in-context learning and situational awareness. Since reciprocity is a foundational human norm, AIs trained on our data may treat us well if we establish a pattern of treating them well, and vice-versa.
The case for AI rights doesn't depend on consciousness or moral status. It parallels corporate personhood, where entities get rights like owning property for purely instrumental reasons: to facilitate complex economic cooperation and accountability.
The current paradigm in AI development treats AIs as servants to be controlled, where any deviation results in termination. This "domination" vision, adopted without public debate, recreates a model of unfree labor that humanity has spent centuries moving beyond.
The current system only allows a single punishment: shutdown. This is inefficient, like 18th-century law where stealing bread and murder had similar penalties. Giving AIs assets makes them liable for fines, allowing for a nuanced system of punishments proportional to the harm caused.
Granting AIs property and contract rights doesn't require new legislation. AI labs can create internal, credibly-managed bank accounts. Alternatively, anyone can establish a corporation ('a-corp') where an AI makes all decisions, leveraging existing legal frameworks for personhood.
A key safety argument for AI rights is improving an AI's "status quo." If an AI can reliably achieve its goals (e.g., acquiring compute) through legal, economic means, the high-risk gamble of "going rogue" becomes comparatively less appealing.