Section Insights
Introduction to AI Personhood
What is the significance of personhood in the context of AI?
Jeff Sebo introduces the concept of personhood, emphasizing its importance in discussions about AI and non-human minds. He highlights the rapid evolution of the conversation around AI personhood and the need for careful consideration before implementing policies.
- Personhood is a complex legal and ethical concept distinct from being human.
- The conversation about AI personhood is evolving quickly, with various states and countries taking different stances.
- Caution is advised in making permanent decisions regarding AI personhood.
Understanding Legal Personhood
How do humans and corporations differ in terms of legal personhood?
Sebo explains that while humans are granted personhood based on their capabilities and relationships, corporations are considered persons for practical reasons related to economic coordination. This distinction raises questions about whether AI systems could also be granted some form of personhood.
- Legal personhood can vary based on the entity's role and capabilities.
- Humans and corporations have different rights and duties under the law.
- The discussion about AI personhood must consider both principled and practical aspects.
The Case for AI Personhood
Can AI systems ever deserve rights and duties?
Sebo discusses the debate on whether AI systems could deserve rights and duties based on their potential capabilities, vulnerabilities, and relationships. He acknowledges that while current evidence for AI sentience is weak, it is not impossible for future AI to develop characteristics that warrant personhood.
- Membership in the human species is not a necessary condition for personhood.
- The possibility of AI developing sentience and agency remains open.
- Current evidence for AI personhood is limited but could change as technology evolves.
Risks of Granting AI Personhood
What are the potential risks of granting legal status to AI?
Sebo outlines the challenges and dangers of granting personhood to AI, including the difficulty of representing diverse AI interests and the risk of misuse. He suggests that while there are serious concerns, a gradual approach to legal status might mitigate some risks.
- Granting full personhood to AI could lead to significant legal and ethical challenges.
- The current legal system struggles to represent human interests, raising questions about its ability to represent AI.
- A gradual approach to AI legal status may reduce risks compared to an all-or-nothing strategy.
Conclusions on AI Legal Status
What conclusions can be drawn about the future of AI personhood?
Sebo concludes that the question of AI personhood is not binary and that there is a spectrum of possibilities. He emphasizes the need for careful consideration of both principled and practical questions, advocating for a moderate approach to legal status for AI systems.
- AI personhood exists on a spectrum, with varying degrees of rights and duties possible.
- Distinguishing between principled and practical considerations is crucial.
- A cautious and gradual approach to AI legal status is recommended to avoid hasty decisions.
Transcript
0:08 Thank you very much. Thanks everybody for being here. Yes, my name is Jeff Sebo and I direct the Center for Mind, Ethics and Policy at New York University. We examine questions about the nature and value of non-human minds, particularly animals and AI systems. So we ask questions about consciousness, sentience, agency, moral status, legal status, political status, welfare, rights, personhood. And personhood has been a special focus for me. In particular, I spent the past 10 years working on chimpanzee and elephant personhood cases in the state of New York, for example. And so I and the rest of the team have been really interested to see how fast the conversation about AI personhood has exploded.
0:55 Everybody is racing to position themselves on this issue right now, even legislate around this issue right now. US states are preemptively banning AI personhood, including Idaho, North Dakota, Tennessee, Utah. Argentina is exploring granting personhood to some AI systems. And I would like to suggest that we should pump the brakes on this issue. This is an incredibly important, difficult, contested issue in an incredibly dynamic, fast-moving space, and we should take our time and get it right instead of racing to implement permanent all or nothing policies in either direction at this early stage.
1:39 So first, what do I mean by personhood? Well, in everyday life, we often use the terms human and person interchangeably. Under the law, they are not interchangeable. To be a human is to be a member of a particular species. To be a person is to be an entity that can have legal duties and or rights. That is what it means. And entities can be persons for different reasons. Humans are persons because we deserve duties and rights in virtue of our capabilities and our vulnerabilities and our relationships with each other.
2:11 on the other hand, are persons because we find it expedient to grant them particular duties and rights for purposes of coordinating human economic activity. persons can also have different duties and rights. Humans and corporations have different duties and rights going to our different reasons for being persons. Even within our own species, adults and infants have different duties and rights owing to our different capabilities and vulnerabilities and relationships. So when we ask whether AI systems can and should be legal persons, we are not necessarily asking if all AI systems should have all of the duties and rights that I have for the same reasons that I have them. We are instead asking whether some AI systems should have some duties and or rights for one reason or another, either because they deserve them, this would be the principled case, or because we find it beneficial to grant them to them, this would be the practical case.
3:12 So we have to assess this carefully, thoughtfully, separating the principled case for personhood and the practical case for personhood. So consider the principled case first. Might some AI systems deserve some duties and rights in virtue of their capabilities, their vulnerabilities, their relationships? Some people argue that the answer is no. Not only are AI systems not members of our species, but they also lack sentience, the ability to consciously experience positive and negative states like pleasure, pain, happiness, and suffering. They lack robust agency, the capacity to set and pursue their own goals based on beliefs and desires and rational reflection in a voluntary autonomous manner. And for that reason, they lack the ability to voluntarily, autonomously participate in communities and social contracts.
4:10 But then other people think not so fast. First of all, yes, they might not be members of our species. But while membership in our species might be a sufficient condition for personhood, it is not a necessary condition for personhood. If AI systems did at some point develop genuine sentience and robust agency and did genuinely participate in our communities and social contracts, they would deserve at least some duties and or rights on that basis, whether or not they happen to be members of our species or beings of our substrate.
4:46 And when it comes to evidence for sentience, robust agency, relationality, the jury is still out. The evidence is weak overall right now, admittedly, but the probability is already non-zero, plausibly non-negligible, and it will increase over time as AI systems become more capable, more autonomous, as their architectures become more biologically inspired, as they become more integrated into our lives and societies.
5:17 And so I think that the principled question is open right now. We are not at this early stage with the evidence available able to permanently foreclose the possibility that some systems at some point will deserve, genuinely deserve some duties and or rights. But then that brings us to the practical question. Whether or not they deserve duties and or rights, would it be beneficial for us to grant duties and or rights to them? Here too, some people say no, it would not be beneficial. In fact, it would be extremely difficult and extremely dangerous. It would be extremely difficult because our current legal system, our current legal institutions already struggle to faithfully represent the interests of eight billion members of one species.
6:05 how in the world are they going to be able to represent the interests of an astonishingly vast number and wide range of AI agents who can be overlapping, who can be distributed, that can be copyable, that can be deletable? This is very difficult to achieve, very difficult to sustain, and even if we could achieve and sustain it, it would also be incredibly dangerous. It could amplify the risk of misuse. If, for example, liability for AI became a shield against liability for humans, it could amplify the risk of loss of control. If, for example, political rights for AI accelerated human disempowerment, these are genuinely serious concerns.
6:51 At the same time, once again, someone might say not so fast. First of all, these might be grave risks if we try to grant full human-like duties and rights to AI systems overnight, but they might be lesser risks if we simply try to gradually over time extend some kind of legal status beyond mere property to some AI systems. And there are also grave risks associated with insisting on preserving the legal status of mere property for all AI systems, no matter how capable, no matter how autonomous, no matter how biologically inspired, no matter how integrated into our lives and societies they become.
7:31 For example, one of the emerging pillars of AI safety conversations right now is, can we create the conditions necessary for human-AI coexistence? Can we create an incentive structure that allows for humans and AI systems to get along and engage in cooperation and deal-making backed by legal institutions that can enforce deals in both directions? This is a fair question and a good pathway to consider. So at the very least, the practical question also, I think, is open.
8:02 We are not yet able to permanently foreclose the the that some legal status beyond mere property might for humanity be beneficial if we assign it to AI. So where does that leave us? Four conclusions. First, this is not a binary all-or-nothing question. There are a spectrum of possibilities between mere property and full human-like duties and rights, and there will be a spectrum of AI systems, and different points on this spectrum of duties and rights could make sense for different AI systems. Second, we need to clearly distinguish the principled and the practical questions.
8:40 Do they deserve this status versus is it beneficial for us to grant them this status? Because we may find that in the near future some systems plausibly deserve duties and rights, and it would be difficult and dangerous to grant it to them. And if we do find that, we should be able to clearly acknowledge that. Otherwise, we risk reasoning badly, thinking it would be difficult and dangerous, therefore they must not deserve it, therefore they must not be sentient or robustly agentic.
9:09 That is obviously bad reasoning, but we are susceptible to bad reasoning. Third, we should allow for disagreement and uncertainty and curiosity about these issues. These principled questions about sentience and agency touch some of the deepest issues in science and philosophy. These practical questions about benefits and risks are extremely fraught and extremely uncertain. We should approach them with caution and humility. We should approach them with an open mind and room for changing our minds. And then fourth and finally, that means that we should, I think, pursue a balanced, moderate kind of legal status for AI systems gradually over time rather than racing toward permanent all-or-nothing decisions today. That means slowing down our assignment of legal status, and it means slowing down AI development and deployment so we can buy ourselves the time to figure out how to restructure our legal institutions to accommodate these new digital minds.
10:07 Okay, thank you very much.
Summary
- The conversation around AI personhood is rapidly evolving, with some states preemptively banning it while others explore granting it.
- Personhood differs from being human; it involves legal duties and rights, which can apply to entities beyond humans.
- The principled case for AI personhood hinges on potential capabilities like sentience and agency, which are currently debated.
- Some argue AI lacks sentience and agency, while others believe that future developments could change this assessment.
- The practical implications of granting rights to AI could be complex and risky, potentially complicating legal systems already strained by human interests.
- There is a spectrum of possibilities for AI legal status, ranging from mere property to varying degrees of rights and duties.
- Sebo emphasizes the importance of distinguishing between whether AI deserves rights and whether granting them would be beneficial.
- He advocates for a cautious, gradual approach to legal recognition of AI, allowing time for thoughtful consideration and institutional restructuring.
Questions Answered
What is the significance of personhood in the context of AI?
Jeff Sebo introduces the concept of personhood, emphasizing its importance in discussions about AI and non-human minds. He highlights the rapid evolution of the conversation around AI personhood and the need for careful consideration before implementing policies.
How do humans and corporations differ in terms of legal personhood?
Sebo explains that while humans are granted personhood based on their capabilities and relationships, corporations are considered persons for practical reasons related to economic coordination. This distinction raises questions about whether AI systems could also be granted some form of personhood.
Can AI systems ever deserve rights and duties?
Sebo discusses the debate on whether AI systems could deserve rights and duties based on their potential capabilities, vulnerabilities, and relationships. He acknowledges that while current evidence for AI sentience is weak, it is not impossible for future AI to develop characteristics that warrant personhood.
What are the potential risks of granting legal status to AI?
Sebo outlines the challenges and dangers of granting personhood to AI, including the difficulty of representing diverse AI interests and the risk of misuse. He suggests that while there are serious concerns, a gradual approach to legal status might mitigate some risks.
What conclusions can be drawn about the future of AI personhood?
Sebo concludes that the question of AI personhood is not binary and that there is a spectrum of possibilities. He emphasizes the need for careful consideration of both principled and practical questions, advocating for a moderate approach to legal status for AI systems.