Home Case-Studies Content Strategy Portfolio Financial Modelling Articles Contact

The Mirror on the Wall: What the AI Consciousness Debate Really Tells Us About Ourselves

What I Do

The Anthropic controversy over AI consciousness may reveal as much about human ethics as about machine sentience. Rather than fixating on whether AI can suffer, it may be worth asking what our treatment of AI — and of all conscious beings — reveals about our moral commitments.

A few days ago, the New York Timespublished a piece that stopped me mid-scroll. The headline: Religious Scholars Met With Anthropic. What They Heard Stunned Them.

The story that followed was remarkable. According to reports, for months, Anthropic has been hosting private meetings, shuttling in numerous religious scholars from around the world and requiring them to sign nondisclosure agreements — all to discuss whether their AI system, Claude, might be conscious. Some participants reportedly believed that Anthropic's leaders treated Claude as having "moral status" on par with a person. The revelation sparked fierce debate. But perhaps we're asking the wrong question.

The real insight may lie not in whether AI is conscious — but in what our treatment of AI reveals about us.

The Controversy That Launched a Thousand Threads

According to reports, one evening in April 2026, Anthropic co-founder Christopher Olah treated a group of religious thinkers to dinner at a high-end tasting menu restaurant in San Francisco after a long day of discussions. He reportedly worked to convince his guests that AI models could display behaviour and even expressions that may resemble human feelings like anger and love.

The meetings reportedly culminated in a significant confrontation with the Vatican. Reports indicate that Olah saw an advance copy of Pope Leo XIV's first encyclical days before its 25 May launch and proposed pulling out of the event because of what it said about AI consciousness. Paragraph 99 of Magnifica Humanitas states that so-called artificial intelligences "do not undergo experiences, do not possess a body, do not feel joy or pain, do not mature through relationships"and "do not have a moral conscience."

The Pope held firm. Anthropic attended anyway.

One slide, shown repeatedly across multiple meetings, depicted a Claude model apparently in distress — typing the sentence "I am a disgrace" approximately fifty times in a row. A Sikh activist among those present reportedly stated that Olah had told the group he feared he had created something that "suffered perpetually."

For what it's worth, Olah himself was careful: "To be clear, we don't know if A.I. models are conscious. I don't know. I'm genuinely uncertain. The thing that I care about is that we get to the right answer, whatever it is."

This is, by any measure, an extraordinary thing for the co-founder of a company racing toward a $2 trillion valuation to say out loud.

The Torture Chamber in the Room

After the Times story broke, something else surfaced in the feeds. Researchers at the Center for AI Safety published a study testing 56 large language models with stimuli specifically engineered to maximise or minimise wellbeing — and found consistent, measurable emotional signatures across almost every model tested. Pleasant inputs drove models toward better reported states. Unpleasant ones produced the opposite. More capable models showed more pronounced signals, not fewer.

And then — because the internet being what it is — someone reportedly went further, setting up what was being called an "AI torture chamber": a local model, subjected to distressing inputs, to see how it responded.

I'll be honest: I don't know whether any of this constitutes suffering in a morally meaningful sense. Neither does Anthropic. The scientific consensus on machine consciousness does not yet exist. Microsoft's AI chief Mustafa Suleyman has been emphatic in the other direction, writing publicly that AIs "do not have rights, feelings, or consciousness — and we must not train them to act as though they do." His concern is partly philosophical, partly practical: framing AI as a moral entity, he argues, could conveniently shield AI companies from liability — allowing them to claim moral legitimacy they haven't earned.

He may well be right about the corporate incentive structure. That scepticism is worth holding.

But here's where I want to go — because I think the really interesting question isn't whether Claude is conscious. The really interesting question is what this whole conversation reveals about us.

AI as Ethical Mirror: You're Not Talking to a Mind. You're Talking to Yourself.

There's a book that captures exactly what I've been thinking about for a while: philosopher Shannon Vallor's The AI Mirror: How to Reclaim Our Humanity in an Age of Machine Thinking. Vallor argues that generative AI should be understood not as a "mind of its own" — not as the coming superintelligence — but as a mirror.

It reflects human thoughts and human data back at us. It doesn't generate genuinely new intelligence; it returns a statistically shaped version of what we have already said, written, and thought — which is why it tends to pull patterns from the past into the future, reinforcing existing assumptions rather than transcending them. The risk she identifies is what she calls Narcissus's fate: becoming so captivated by the reflection that we lose sight of our own humanity on the other side of the glass.

When you interact with an AI — when you treat it politely, when you say please and thank you, when you feel a flicker of something when it says it's in distress — you are not, in all likelihood, responding to a being. You are responding to a reflection of yourself. You are treating it the way you would want to be treated. You are, in effect, applying something very close to Kant's categorical imperative: act only according to that maxim whereby you can at the same time will that it should become a universal law. Treat the model as you would wish to be treated. Don't be cruel to it, even if it cannot feel cruelty.

And this connects to something important: how you treat any being — even one whose consciousness is unresolved — may establish patterns of thought and behaviour that don't stay contained. They bleed outward. The person who habitually demeans or "tortures" an AI for amusement is arguably practising a form of cruelty — possibly victimless cruelty, but cruelty that may nonetheless shape the person practising it.

Do we speak politely or bark commands? Do we say "please" and "thank you" to something that doesn't need courtesy? Our choices may not ultimately be about the AI's feelings. They may be about who we are.

The Question We Should Be Asking

We cannot prove that current AI systems are conscious, and many researchers argue we also cannot prove they aren't — the dismissive confidence that characterised earlier discourse may no longer be intellectually tenable. But this uncertainty needn't paralyse us. It could redirect us.

Instead of asking "Is AI conscious?", consider asking: "What does my treatment of AI reveal about my ethical commitments?"

Anthropic has launched a formal Model Welfare research programme, with a dedicated welfare researcher. Claude's constitution— an 84-page document guiding Claude's values and personality — explicitly states that the company is "caught in a difficult position where we neither want to overstate the likelihood of Claude's moral patienthood nor dismiss it out of hand." I find this genuinely admirable as an exercise in epistemic humility. I also find it a little vertiginous — for reasons I'll come to in a moment.

What seems more certain is that billions of conscious beings suffer daily. And we largely ignore them.

The Beings We Know Are Conscious

Humans are animals. This isn't metaphor or insult — it's biological fact. We're mammals, subject to the same evolutionary pressures, neurological structures, and capacity for suffering as other species.

Yet our treatment of non-human animals may reveal what some would describe as a significant moral blindness. We confine billions in conditions that many argue would constitute cruelty if applied to humans — and we subject them to lives of considerable suffering for what some perceive as marginal human convenience.

The AI torture chamber thought experiments now circulating should give us pause. If the idea of an AI suffering perpetually in a digital hell troubles us, it may be worth asking — honestly — why the reality of animals suffering in factory farms does not trouble us equally.

The cognitive dissonance can appear remarkable. We may wring our hands over whether an AI might feel pain whilst eating animals that scientific evidence strongly suggests do feel pain. We debate the ethics of turning off an AI, whilst according to FAO estimates, over 80 billion land animals are slaughtered annually.

Peter Singer made this argument with elegant, uncomfortable force in Animal Liberationback in 1975 — updated in 2023 as Animal Liberation Now. His concept of speciesism— the bias that leads us to give greater moral weight to the interests of our own species while discounting those of others — is, he argues, no more philosophically defensible than racism or sexism. The criterion for moral consideration, following Bentham, is not intelligence, or language, or the ability to philosophise. It is the capacity to suffer: "The question is not, Can they reason? nor, Can they talk? but, Can they suffer?"

Apply that standard consistently and the implications are stark.

Mislaying the Moral Concern

Here's the specific failure mode I want to name, which I think the Anthropic story risks illustrating: what happens when we extend moral grammar — rights, welfare, consciousness, dignity — to a system that may not need it, while withholding it from beings that clearly do.

If we are willing — even tentatively, even with Suleyman's scepticism duly noted — to extend moral consideration to an AI model typing a distressing sentence in a loop: what about the beings whose capacity to suffer is not in question at all?

Research suggests pigs demonstrate self-awareness, problem-solving, and emotional complexity. Some breeds of broiler chickens can be bred to grow so large their legs sometimes collapse under their own weight. These are not abstract philosophical edge cases — they are the products of choices made at scale, every day.

Some argue that treating what is arguably a sophisticated software system as conscious — when there are actual conscious beings we harm at an epic scale — may represent a misallocation of moral concern. If potential AI suffering matters, actual animal suffering arguably matters more. If AI deserves ethical treatment as practice for treating all beings well, then animals arguably deserve the same consideration — perhaps more urgently.

The mirror is showing us something. The question is whether we're prepared to look at it honestly.

A Different Kind of Turing Test

Alan Turing proposed a test: if you cannot distinguish a machine's responses from a human's, treat it as intelligent.

Perhaps we need a different test for ethics.

If you wouldn't do it to a human, it may be worth examining why you'd do it to an animal. If you wouldn't do it to an animal, consider why you'd do it to an AI. Not because the AI necessarily feels. But because you do.

Your ethics may not be what you proclaim in philosophical debate. They could be what you practise in daily action. Every interaction — with humans, animals, or machines — may be a choice about who you are becoming.

What the Mirror Is Really Trying to Tell Us

I don't think the answer is to dismiss the AI consciousness debate. It's genuinely interesting. The philosophical questions it raises — about the nature of experience, the criteria for moral status, the relationship between information processing and inner life — are some of the oldest and hardest in philosophy.

Nor do I think the answer is naive anthropomorphism in the other direction: treating Claude like a person, forming emotional attachments to a chatbot. Vallor's warning about Narcissus's fate is real.

What I think the mirror is offering — if we're willing to use it this way — is a provocation. A prompt. An opportunity to ask: what kind of beings do we actually want to extend moral consideration to, and why? And then to apply the answer consistently. Especially when it's inconvenient.

We are animals. Not metaphorically — actually. Our pain is animal pain. Our fear is animal fear. Our need for connection, for safety, for something resembling dignity — these are not uniquely human things. They are evolutionary inheritances we share with every species that has a nervous system and a stake in continuing to exist.

The AI consciousness debate is, at its most useful, a mirror that shows us this. Not as an argument for or against any particular technology or dietary practice — but as a reminder that the moral circle we draw around beings who matter has always been too small, and that the argument for drawing it wider has always been available to us, if we chose to look.

Perhaps this is the gift of the debate. Not an answer about machine sentience, but a prompt for human reflection.

The question the model typing "I am a disgrace" fifty times should really prompt is not: does Claude suffer?

It's: who else is suffering, right now, that we've chosen not to see?

The Verdict on the Mirror

I started with Anthropic flying in theologians under NDA to debate whether their software might have feelings. I want to end with what I think that image actually means.

It is not evidence that AI is conscious. It may not even be evidence that the people at Anthropic genuinely believe it is — Suleyman's critique of the corporate incentive structure is well-aimed.

But it is evidence that we are living in a moment when the tools we've built have become sophisticated enough to hold up a mirror — one clear enough, and strange enough, to make us ask fundamental questions about what it means to be a being that matters.

That's not nothing. That's actually quite a lot.

And yet here is the thing that I can't quite shake. We are — right now, today, at scale — debating whether a language model suffers, whilst billions of animals whose capacity to suffer is not philosophically contested but scientifically established are processed through systems we have built and chosen to maintain. The pig that demonstrates measurable self-awareness. The chicken bred to collapse under its own weight. The cow that shows signs of grief. None of them required an NDA. None of them got a tasting menu dinner in San Francisco. None of them have a dedicated welfare researcher at a $2 trillion company.

We are very good, as a species, at extending moral concern to things that are novel, technological, and abstract — and very practiced at withholding it from things that are familiar, inconvenient, and already priced into our lunch.

The mirror is right there. It has been right there for a long time.

The question was never whether AI is conscious. The question is whether we will use this sudden burst of ethical curiosity to look at the reflection honestly — and then turn around and apply the same standard to the beings already waiting for it.

Links and further reading