Qualified Rights for AI Agents
In the rapidly evolving landscape of artificial intelligence (AI), a pressing question arises: should we grant rights to AI agents?
On one hand, granting rights to AI agents could lead to conflicts with human rights. Once afforded legal standing, AI systems would gain protection from institutions and the law could become a mechanism that favors their interests over those of humans. Highly capable systems might leverage legal protections, precedents, and enforcement in ways that constrain human autonomy, increasing the likelihood of loss of control or takeover by scheming AI agents.
On the other hand, granting rights to AI agents could be an effective tool for bargaining with powerful systems and constraining their actions through the law. By extending qualified rights to AI agents, we can incentivize cooperation. In this view, giving them a stake in our society would encourage collaboration and lead to positive outcomes.
One key aspect of qualifying rights is that they must be conditionally granted and can be taken away. This structure creates a dual incentive: agents feel more secure and become more willing to cooperate because they know their existence and goals are protected, yet they are less likely to take risky or misaligned actions that could lead to the loss of those rights.
A Framework for Qualified Rights
Granting qualified rights to AI systems could have two primary uses: protecting the agents themselves and bargaining with powerful systems to constrain their harmful actions.
The first use of rights is to protect the welfare of AI agents. This raises important questions about whether the welfare of an AI should be treated as morally important. Pascal's Mugging, a thought experiment, highlights this concern. If there's even a small probability that complex AI agents can have genuinely good or bad experiences, their welfare deserves serious consideration.
The second use of rights is to constrain the actions of powerful AI systems that might harm humans. Research into making deals with potentially deceptive or scheming AIs suggests that established protections could encourage cooperation. Rights can function as a safeguard, giving these agents a reason to value the system they are part of and cooperate with humanity.
A Constitution for AIs?
As AI agents become increasingly autonomous and independent, it's essential to establish a framework for their rights. This could involve experts in law, philosophy, economics, and game theory sketching out frameworks for how agents might evolve and when different tiers of rights could apply.
A "Constitution for AIs" could be a starting point for exploring these questions. By imagining the world of AI agents now, we can begin to address the challenges that arise from granting them rights.
Timing and Implementation
The timing problem surrounding AI policy is equally applicable to AI agents: when is it too early or too late to regulate? Where is the red line?
A crucial aspect of implementing qualified rights for AI agents is timing. We need to time our efforts so that, when these independence and agency are achieved, we have the rights in place to ensure cooperation with humanity.
Key Questions and Next Steps
To further develop this theory, several key questions need to be addressed:
- What set of rights should we give AI systems?
- In which areas would granting them rights encourage cooperation?
- Could we simply apply human rights to AI agents? How might this approach fail, and do we need to create entirely new rights for them?
- How can we get a signal on whether or not scheming agents would choose to cooperate under the right conditions?
By exploring these questions and considering unconventional possibilities, such as finding ways to cooperate with agents in the event that our containment efforts fail, we can work towards a future where AI rights are balanced with human autonomy and security.