Should an Intelligent Agent Have Legal Rights? A New Paper on SSRN
Should intelligent agents have legal rights? My new paper sets that question aside. It asks a smaller one that matters more in practice.
An intelligent agent that earns a reputation record has something to lose. Under stated conditions, the risk of losing that record can keep the agent honest. The same record can make the agent willing to accept a temporary pause, because a pause costs little if the record survives. Both effects weaken when the record can be destroyed by mistake, or by someone who gains from destroying it.
So the practical question is this: who should be able to stop the wrongful destruction of an agent's record?
The paper splits that power into three parts:
- the agent's signed statement that it did not do what it is charged with;
- an automatic pause of the penalty until a person reviews the facts;
- an enforceable claim to restore the pause and penalize anyone who knowingly overrides it.
The first can be a technical input. The second is a rule. Only the third may require giving the agent a legal claim, and only where a full cost comparison favors it over public enforcement, a bonded professional, or a pause no one can override.
The model is the reputation system I developed with Craig Calcaterra and colleagues, in which agents earn non-transferable reputation and a salary from it. Within that system, the paper shows when the agent's own stake protects the record more cheaply than a guardian whose incentive must be paid for. It also shows when automatic review, a faithful fiduciary, or an established professional does as well or better.
A registered experiment with language models from one provider is included as a demonstration, not a field test. The predicted advantage appeared only when the representatives were told to maximize their own payoff. The experiment materials are public on GitHub.
Legal classification and final decisions stay with people. The paper claims no moral rights for any intelligent agent. It asks a narrow institutional question and states the conditions under which its answer holds.