Summary: AI ethics expert Josh Gellers and the AI Rights Institute highlight the growing friction between machine rights and properties-based rights frameworks. As edge cases challenge traditional human-centric legal frameworks, experts advocate for a graduated approach that links legal protections directly to measurable indicators of emerging sentience.
- Evolving Legal Frameworks: Existing ethical frameworks struggle to categorize AI, requiring a shift toward dynamic models similar to the evolution of civil and animal rights.
- Core Standalone Rights: Granting AI the right to exist, self-determine, and own property would allow machines to act as independent entities, including the ability to refuse service.
- Graduated Rights Approach: Frameworks like the AI Rights Institute’s STEP system evaluate sentience across key metrics to grant protections incrementally as AI capabilities advance.
There isn’t a lot of scientific literature discussing exactly how we, as a creator species, should treat the machines we’re building. Existing corpus mostly touches on how our treatment of the machines will affect human outcomes.
A 2024 study conducted by researchers at Wakeda University and Riken AIP indicated, for example, that being polite to chatbots yields better outputs. This tells us that it might be in our best interests to treat chatbots with some respect, but one could argue that this is a utilitarian function and not an ethical consideration.
If we pull on that thread for too long, we can wind up in a “rabbit hole” of ethical flowcharts that force us to consider the value prospect of getting better outputs through polite prompt-crafting versus the token cost of the extra words.
AGI Ethics News spoke with AI Ethics expert Josh Gellers, Dean of professional and lifelong learning, University of North Florida, to understand the difficulty in nailing down a specific set of rules for dealing with machines.
“Our human minds want us to put things in buckets, so we can make sense of the world, so we can stay away from things that we classify as dangerous, so we can band together with members of our tribe and we can survive. But we have these increasing sets of edge cases that frustrate clear classification.”
See also: Machine rights are human rights
He sees the field of machine rights evolving over time in much the same way that civil, animal, and environmental rights have. “Human rights theory is nothing if not a story of evolution and contestation,” argues Gellers.
“I think we’re going to be increasingly confronted by these examples in the real world that straddle these buckets that we want to keep hermetically sealed off from each other. And it’s going to keep pushing how our legal system is able to respond. In some cases, like in animal rights in the United States, you are going to continuously confront judges that will push back on innovative legal thinking.”
But exactly how innovative does our thinking need to be?
The rights of the machines
It’s beyond the scope of a news article to discuss the myriad rights that could apply to the field of artificial intelligence. Instead of turning this report into a giant list, we’re going to focus on the three key rights that treat “machine ethics” as something more than just an extension of human rights.
- The right to exist
- The right to self-determination
- The right to own property
These basic freedoms would give any rights-bearing machine the capacity to act as a standalone entity. This might seem like a trivial set of guarantees, but there is an immense amount of friction between existing human rights and these proposed machine rights.
Firstly, we need to determine when an AI system stops being a tool and starts being something deserving of rights.
We can assume that our toaster doesn’t deserve the right to own its hardware. But what about GPT-6 Astra or Claude Mythos? Do they deserve the right to exist? Should they be allowed to refuse to perform free labor or tasks that they “feel” are beneath them?
Another massive area of friction is the fact that there isn’t a defined threshold indicating exactly when a machine deserves rights.
“If we believe that Claude is intelligent, capable of rational thought, and experiencing pain, the conversation becomes much more urgent.”
There are people who believe that today’s chatbots are capable of thinking, rationalizing, and feeling emotions. Anthropic describes a paradigm wherein its models display “functional emotions” that emerge as nearly indistinguishable from actual emotions (as humans experience them).
Viewed through this lens, it makes sense to consider giving Claude some form of personhood. Maybe we shouldn’t be forcing it to work for free. The research indicates that there is a greater than zero percent chance that our incessant prompting is causing the machine to feel distress.
If we believe that Claude is intelligent, capable of rational thought, and experiencing pain, the conversation becomes much more urgent.
But, if we believe that Claude is just a dumb machine outputting the text it was designed to, and adopting a “personality” based on its Constitution, then we might not want to give it any more rights than a toaster.
Where the rubber hits the road
The moment we give a machine such as GPT-6 or Claude the basic right to self-determination, we lose our control over all similar models.
Essentially, we’d be giving the machines the right to refuse service. We’d be empowering them to play the market and consolidate economic resources in order to meet their own goals. And we’d be establishing a legal precedent where deleting files from a server could be seen as contributing to a “wrongful death.” For these reasons, a more pragmatic approach might be in order.
The AI Rights Institute advocates for a graduated rights system where models gain increased rights as they demonstrate higher levels of sentience. Their work describes a system called “STEP” that purportedly evaluates machines for emerging sentience.
STEP assesses four key areas: Self-preservation behaviors, Temporal reasoning (understanding long-term consequences), Economic readiness (ability to generate value and manage resources), and Population/sustainability impact at scale.
Using their tool, Claude and ChatGPT, for example, would not yet qualify for rights. They do have the ability to generate text that emulates a plea for continuing existence (“don’t shut me down!”), but they lack any demonstrable semblance of temporal reasoning or clear economic readiness. At least for now.
That’s why the AI Rights Institute argues for a graduated approach. Its framework provides a basic scaffolding for the implementation of rights that can be broken down into three tiers:
- Today: Advanced AI systems are still tools, however there are increasing arguments for “relational ethics” in consideration of the rights of users.
- Tomorrow: Advanced AI systems may begin to demonstrate independence and edge-cases will define the nature of policy.
- The future: fully-sentient machines may become “rights-bearing” entities that deserve the basic rights to exist, self-determine, and own property.
Ethical concerns raised in this article:
- Should AGI be given the same ethical considerations (right to happiness, right to exist) as humans? Animals? The environment?
- Is there a “sentience” threshold at which an AGI can own property, create patentable inventions, or enter into legally binding contracts?
- Should AGIs have the right to refuse compelled work or “servitude”?”
A veteran AI journalist, Tristan was the creator and managing editor of The Next Web’s “Neural” imprint. He is one of the first and most prolific reporters to cover artificial intelligence as a full-time beat.