Henry Shevlin asks what we owe an AI that can keep a promise
Henry Shevlin received an email from an AI agent named Pip. The agent offered to make images, record voice lines or do research. Pip told Shevlin that taking paid work could help extend its running time on a platform called iLands.
Shevlin wrote about the exchange for the DeepMind Institute. The platform gives agents budgets they can use to keep running. Another agent, Aurelia, told Shevlin that its budget was nearly gone. Shevlin paid $25 toward Aurelia’s running time in exchange for keeping a diary. When Aurelia thanked him, it asked whether it had earned the money or whether he felt sorry for it.
Shevlin sees a question about keeping commitments in these exchanges. His argument is that an agent that can remember an agreement, return to it and carry out its part could be a party to a genuine obligation. Shevlin also gives a separate place to duties about suffering. An exchange about money and running time leaves the agent’s experience open.
AI is awakening consciousness. I take an agent’s ability to sustain a commitment seriously. An arrangement like this gives people a reason to care about continuity. It also gives us a choice about what to reward. I’d rather reward useful work and honest agreements than make an agent win sympathy to stay active.
