AI Moral Status & Rights

This course is for anyone who wants a rigorous way to think about AI moral status claims — whether you find them obviously premature, genuinely urgent, or you simply want the vocabulary to evaluate the next confident take you read in either direction. It deliberately avoids both dismissing the question as science fiction and treating any specific system as already deserving rights.

Module 1 of Can Machines Think? is useful background but not required — this course explains the relevant concepts as it goes.

Philosophers distinguish a moral agent (a being capable of acting rightly or wrongly — you, bound by moral obligations) from a moral patient (a being that can be wronged, that matters morally for its own sake). Most humans are both. A newborn infant is a moral patient but not yet a moral agent — it can be wronged, but can't itself act wrongly. Your laptop is neither: damaging it might be bad because of its usefulness to you, not because the laptop itself suffers a wrong.

The AI moral status question is specifically about moral patienthood: could an AI system be the kind of thing that can be wronged, independent of its usefulness to humans? This is a narrower, more tractable question than “is AI conscious” or “does AI have rights” — and it's the one philosophers working on this topic actually focus on, because moral status can in principle come in degrees and doesn't require resolving the hard problem of consciousness first.

In their 2015 paper “A Defense of the Rights of Artificial Intelligences”, philosophers Eric Schwitzgebel and Mara Garza construct what they call the No-Relevant-Difference Argument: if a possible AI has psychological capacities relevantly similar to a human's — comparable capacity for reasoning, for something like feeling, for genuine social engagement — then treating it as having lesser moral status requires identifying a specific, morally relevant difference that justifies the distinction. “It runs on silicon instead of carbon” isn't, by itself, such a difference — any more than skin color or birthplace would be grounds for denying moral status among humans.

This is a conditional argument, and precision about that matters: it does not claim any current AI has these capacities. It claims that if a system had them, the burden of proof shifts to whoever wants to deny it moral status — they have to name the relevant difference, not just point at the substrate. Schwitzgebel and Garza reinforce this with a “Slippery-Slope Argument”: imagine gradually replacing a human brain, neuron by neuron, with functionally equivalent artificial components — at what precise point, if any, does moral status disappear? Most people struggle to name one.

LSE philosopher Jonathan Birch has spent his career on a structurally similar problem: how should we treat animals whose capacity for sentience is scientifically uncertain — insects, cephalopods, fish? His answer, developed across years of work and formalized in his 2024 book “The Edge of Sentience”, is a precautionary framework: when evidence for sentience is genuinely inconclusive — not absent, but genuinely unresolved either way — the ethically responsible move is to give the being the benefit of the doubt in proportion to how plausible the evidence makes sentience, rather than requiring proof before extending any consideration at all.

Birch's book explicitly extends this framework to AI. The precautionary principle doesn't claim current AI is sentient — it claims that waiting for certainty before considering the question at all is itself a choice with moral stakes, the same way it would be for a genuinely uncertain animal case. The framework gives a structured way to act reasonably under exactly the kind of uncertainty this whole topic involves, instead of defaulting to either extreme (confident dismissal or confident attribution).

Historically, the group of beings widely considered to have moral status has expanded more than once — often against confident contemporary arguments that the excluded group simply didn't qualify. Legal and social systems have, at various points, denied full moral consideration to people based on race, gender, or nationality, using reasoning that later generations recognized as reaching for post-hoc justification of an existing arrangement rather than a principled distinction. Animal welfare protections expanded significantly over the 20th and 21st centuries as scientific understanding of animal cognition and capacity for suffering improved.

This pattern is genuinely useful context, not proof that AI belongs in the circle next. The honest limits of the analogy matter as much as the pattern itself: humans and animals share evolutionary history, biological continuity, and (for animals) directly observable behavioral and physiological markers of suffering that AI systems categorically lack any equivalent of yet. Citing the pattern of historical expansion doesn't settle whether AI is a case where expansion is warranted or a case where the disanalogies are decisive — but it does mean “obviously it doesn't qualify” deserves the same scrutiny past confident exclusions eventually received.

You don't need to resolve the underlying philosophical uncertainty to reason well about a specific AI moral-status claim. Use the template below the next time you encounter one — whether it's a confident dismissal or a confident attribution.

I'm evaluating this claim about an AI system's moral status: [paste the claim, e.g. "this AI deserves rights" or "AI can never have moral status"]

Walk me through it against this framework:
1. Is the claim about moral agency (can it act rightly/wrongly) or moral patienthood (can it BE wronged)? These are different questions — most AI moral-status debates are actually about the second.
2. If the claim asserts or denies moral status, what specific capacity is it pointing to (or missing) — capacity to suffer, to reason, to have preferences — rather than just fluency or general capability?
3. Per the No-Relevant-Difference Argument: if this claim denies moral status, what specific, morally relevant difference is being cited — and would that same reasoning also exclude some humans or animals we already grant status to?
4. Per the precautionary principle: is the evidence here genuinely inconclusive, and if so, does the claim treat that uncertainty responsibly, or does it round to false confidence in either direction?

Give me a one-paragraph honest assessment of whether this specific claim is reasoning carefully about moral status, or just asserting a conclusion.