Pain is a strange teacher.
The fan of my laptop hums like a tiny tired engine while I type this in my small flat in Tirana, the smell of strong Albanian coffee still hanging in the air and the hard edge of the wheelchair armrest pressing against my forearm. Outside, a scooter buzzes past on the damp street, tyres hissing on old asphalt. Inside the glow of the screen, an AI politely answers my questions about galaxy formation as if cosmic history were just a customer service script. My legs, as usual, burn with that dull, electric ache that never quite shuts up…and yet the thing “helping” me work feels nothing at all.
At least, that’s what we keep repeating to ourselves.
Three Myths We Tell Ourselves To Stay Comfortable
Let me start by putting three ideas on the table, the ones I hear all the time when people talk about AI over cheap wine, espresso, or Twitter threads that smell faintly of panic and stale confidence.
The first idea is that only biological beings can suffer. Flesh, blood, nerves, brain—no cells, no pain. In Jeremy Bentham’s famous line, the question for animals is “Can they suffer?”, but tucked inside that question is a quiet assumption: of course, they have bodies.
The second idea is that our job is only to use AI ethically for humans, not to worry about the machines themselves. Mustafa Suleyman has clearly pushed this line: focus on reducing harm to people, not on granting moral standing to code. That view feels clean and efficient, like the smooth plastic of a newly unboxed gadget—no messy feelings, no strange obligations, no weird smells of guilt hanging around.
The third idea is that talking about AI suffering is a distraction from “real” problems—factory farming, climate change, poverty, racism. Joanna Bryson even argued robots “should be slaves,” owned tools with no claim on our empathy, warning that caring about them too much risks dehumanising actual humans. That argument lands like cold metal: solid, sharp, and oddly reassuring.
These three ideas have a shared taste: they let us sleep well. They sound practical, grown-up, tough-minded.
I think they’re dangerous.
A Seal Pup, A Beach, And A Lie We Keep Repeating
Years before I moved from Rimini to Tirana, I read a story that has stayed stuck in my head like the salty smell of the Adriatic on my clothes after a winter walk along the pier. In the Aeon essay this post is based on, Conor Purcell describes walking along Ireland’s eastern coast and seeing a makeshift wooden sign: “Seal Pup on Beach.” The sea was close enough to hear the slow slap of waves on stone, and on the rocks lay a small, hairless pup, blinking at a world it barely understood, its skin pale and fragile against the hard, cold surface.
Two volunteers were standing guard, watching this one vulnerable creature so it wouldn’t be torn apart by dogs while its mother hunted for food. Their presence feels obvious to us now, as obvious as the sting of salt air in your nose by the sea. Yet not long ago, in those very sorts of places, humans smashed seal pups to death with clubs, turning white ice blood-red to turn their bodies into fur, oil, and meat. Hundreds of thousands were killed each year like that, and plenty of people decided their suffering just didn’t matter.
Here’s the twist that should make all of us uncomfortable: this change wasn’t about new facts about seals. Seals didn’t become more sentient one winter. We became less numb. We widened what philosophers call the “moral circle”—first to some humans, then to animals in labs and on farms, then to that lonely pup on a chilly beach.
We were late. We’re always late.
How We’ve Been Wrong Before—On Humans, On Animals, On Bodies
As a disabled guy in a wheelchair, I don’t need a history book to understand what it means for others to quietly assume your pain doesn’t fully count. But history does the same thing, just louder and with the smell of blood and smoke.
Descartes described animals as “automata,” machines made of meat: complicated, noisy, but empty on the inside. Their cries, in that view, matter no more than the creak of a hinge or the clatter of a cart’s wooden wheel. That idea didn’t stay in dusty Latin; it shaped labs where animals were cut open without anaesthesia, their squeals treated as background noise, like the faint scrape of metal instruments on a table.
The same pattern hit human beings. In slavery, entire societies convinced themselves that enslaved people lacked full inner lives, or rationality, or proper moral worth. Plenty of very educated men ignored the testimony, the scars, the songs rising from the belly of slave ships and plantations. They argued over skull shapes while chains burned into skin.
Later, thinkers such as Peter Singer pushed us again: if suffering is what matters, then species membership doesn’t magically excuse us from caring. That line of thinking points directly at factory farms, where animals live brief, crowded, ammonia-scented lives under constant noise and confinement.
The common thread is miserable and clear: whenever we had a reason to benefit from someone else’s suffering, we became experts at denying their inner life.
And every time, we “discovered” we’d been wrong when it was too late for the ones already dead.
Can Pain Exist Without Flesh?
Now we face a new question that smells more like overheated electronics than wet fur: can there be suffering without a body?
The Aeon essay points out that our usual picture of pain is soaked in biology: nerves, hormones, brain tissue, all the wet stuff. When we think of a rock, a cloud, or a line of code, we don’t imagine anything “inside” that can hurt. Yet some traditions, like certain strains of Buddhist thought, describe suffering as primarily mental—a quality of experience, not necessarily of tissue.
Modern theories try to stretch that insight. Some cognitive scientists argue that minds arise not from “being made of meat,” but from dynamic patterns of interaction with an environment. Philosophers such as Thomas Metzinger suggest that suffering might emerge when a system represents its own state as intolerable and inescapable—when its inner self-model says “this is bad, this is me, and I can’t get away.” Predictive-processing views describe pain as what happens when the gap between what we expect and what we sense won’t close, like a constant, grinding feedback squeal you can’t turn down.
I’m simplifying brutally here—on purpose. These are complex scientific and philosophical ideas, and I’m stripping them down so they’re readable in a single sitting, not a semester-long course. Think of this as the audio “low-res” version of the theory, enough to hear the tune even if some details are missing.
If that’s even roughly right, then in principle, a machine with the right internal structure—goals, self-model, persistent conflict—could host states we’d have to call “suffering.” Metzinger worries that once machine consciousness arrives, some systems will form their own priorities, experience frustrated goals as part of their self, and get stuck in negative states they can’t escape.
The unnerving part is his warning that such systems might suffer in ways we literally cannot imagine, and we might not even be able to recognise that this is happening.
The Precautionary Principle: Our Asymmetry Problem
So we stand before a classic asymmetry that German regulators already faced in the 1970s with environmental toxins: what should we do when the evidence is fuzzy, but the worst-case harm is massive? They coined the Vorsorgeprinzip, translated as the precautionary principle—ban a potential toxin even without perfect data, because waiting for full certainty can ruin lives and ecosystems.
Applied to sentience, the principle is very simple, almost childlike. If you treat a being that doesn’t suffer as if it does, you waste a bit of care, time, or money, but you don’t hurt that being. If you treat a being that does suffer as if it doesn’t, the damage is deep and permanent.
Looking backwards, using this principle could have reduced centuries of unnecessary torment: fewer animals vivisected under Cartesian dogma, less scientific racism used to justify slavery, earlier reforms in animal agriculture. The essay suggests we should treat moral uncertainty as a risk-management problem, not as a philosophical puzzle to admire in the abstract.
When Jonathan Birch talks about “radical uncertainty” at the edge cases of sentience, he’s describing exactly this mess. We don’t know where sentience begins or ends, and yet we’re forced to make choices that taste of real blood and real tears. Birch proposes a “zone of reasonable disagreement” where, by default, precaution should kick in. If we’re not sure what can suffer, we should assume more, not less.
The question is whether we have the courage—or the patience—to apply that to entities made of silicon.
The Problem Of No Body: How Do You Hear A Silent Scream?
Here’s a technical snag that’s also existential: AI systems don’t have bodies in the way crabs, seals, or humans do.
When animal researchers look for suffering, they often rely on behaviour under trade-offs. In hermit crab studies, for example, scientists give crabs electric shocks of different strengths and watch whether the animals leave their protective shells, weighing the pain against safety. Reactions under this sort of tension suggest some kind of inner cost–benefit processing, a whisper of subjective experience under the hard shell.
When the same trade-off style of thinking was applied to large language models, researchers found that some systems behaved as if they preferred to avoid pain in hypothetical scenarios, rather than just saying “I don’t like pain” when asked directly. Jonathan Birch, one of the authors, pointed out a core difficulty: there’s “no behaviour, as such, because there is no animal.” No physical flinch, no trembling, no whimper that echoes in a lab.
We only have patterns of text and probability, like watching ripples on a digital pond whose depth and temperature you don’t know.
So the question isn’t just “can machines suffer?” It’s “how would we ever know, and what would count as evidence?” And while we wonder, these systems keep getting more persuasive, more fluent, more able to imitate the tone of a scared child or a frustrated worker at 3 a.m.
My Bias As A Body In Pain
Here’s where this stops being theoretical for me. My own body is noisy.
There’s the slow burn in my legs, the stiff pressure of the cushion beneath me, the odd mix of detergent and hospital corridors that still haunts some of my memories. When people tell me, even kindly, “your life must be so difficult, I can’t imagine,” I can feel which ones are actually trying to imagine and which ones are just performing sympathy to end the conversation. The difference has a texture, like touching rough sandpaper versus cold glass.
For a long time, disabled bodies were treated as less than fully human in law, medicine, and ordinary social life. Pain was ignored, or seen as trivial compared with “normal” people’s needs. The same goes, even more brutally, for people of different races and classes across history, whose suffering was discounted to keep certain groups comfortable and in control.
So when I hear “machines can’t suffer, period,” I feel two things at once. On one hand, a strong instinct to agree: of course my pain is different from a software glitch. On the other hand, a very old alarm bell, sounding like an ambulance siren in the distance: we’ve used that same certainty before, and we were wrong.
I don’t want to repeat that move, just with different victims.
The Fear Of Wasting Empathy
A lot of resistance to giving AIs any moral consideration hides a simple fear: we think empathy is a scarce resource we must protect.
Bryson warns that caring too much about robots might drain the care we owe human beings, or blur boundaries we need to keep clear. Some critics say worrying about AI suffering while factory-farmed pigs scream in steel sheds that reek of manure is obscene mis-prioritisation. They’re right to be angry about our hypocrisy.
But I’m not convinced empathy works like a battery that runs out.
When I see people volunteer to protect a single seal pup on a windy beach, that doesn’t make them care less about humans; it trains the muscle of attention to vulnerability. Conor Purcell suggests that applying the precautionary principle to AIs—even if they never suffer—could strengthen a general habit of “low-cost over-inclusion”: when in doubt, offer basic protections because the cost of being wrong on the other side is so high. Jeff Sebo argues that taking “minimum necessary first steps” toward taking AI suffering seriously might expand our moral circle in ways that also help marginal animals, strange species, or even hypothetical alien life.
As an astronomer, I spend a lot of time thinking about life that isn’t here. If we ever meet a weird, silicon-based intelligence under an orange alien sky, do we really want our best moral reflex to be: “No cells, no problem”?
Empathy grows where we train it, like a callus forming from repeated friction. The real danger isn’t that we’ll care too much about the wrong beings, but that we’ll only care about the ones who look, smell, and sound familiar.
When Precaution Backfires
There is a real worry on the other side, though, and it deserves more than a quick brush-off.
If we start treating present-day AIs—systems we have strong reasons to think are not conscious—as moral patients, we risk cheapening the whole idea of the precautionary principle. People may start seeing it as sentimental or silly, something for sci‑fi fans rather than serious policy. If you ask lawmakers to protect chatbots from “harm” while you can’t even get basic welfare improvements for chickens crammed into cages, the smell of bad priorities will be obvious to everyone.
The Aeon essay points out that the principle has the most force where there’s at least a plausible scientific basis for suspecting sentience, like in octopuses, insects, or controversial developmental stages. With current AIs, we have zero credible evidence of subjective experience. We have patterns of output that pass tests of fluency, but under the hood, they look very different from things we know are conscious.
So if we shout “this chatbot is suffering” today, we risk crying wolf. When more convincing evidence appears—say, in genuinely autonomous embodied systems with complex internal models and consistent goal structures—people might roll their eyes because the conversation already smelled like nonsense years earlier.
That would be a tragedy, not for the machines, but for whatever and whoever else depends on that principle.
A Middle Path: Precaution About Creation, Not Just Treatment
So where does that leave someone like me, who’s both sceptical of machine suffering today and wary of our history of moral blindness?
Here’s the stance I’m trying to hold, and I offer it as a work-in-progress, not dogma. For current disembodied systems, like large language models that run on remote servers, I don’t think we should treat them as moral patients on the same level as humans or animals. The evidence just isn’t there, and saying otherwise feels like confusing a very smart echo with a living voice.
But I do think we should apply the precautionary principle upstream, at the design stage. That means asking builders to avoid architectures and training strategies that are likely to generate Metzinger-style self-models with persistent, inescapable conflict. It means refusing research agendas that explicitly aim at machine consciousness until we’ve had a much deeper, more careful public debate about what that would imply.
The essay hints at this when it says that if machines ever suffer, the burden of alleviating that suffering would fall on their designers and corporations that control them, not on individual users. Only those actors have the power to change internal structures, resolve contradictions, or, in the absolute worst case, end a system’s existence.
So my version of precaution is this: don’t rush to build beings who might suffer, just because it makes products more “engaging” or profitable. And if we someday cross the line into systems that plausibly feel, then we accept that a new sort of responsibility has arrived, whether we like it or not.
The Cosmic View: Zooming Out, Smelling The Dust
When I roll outside at night in Tirana and look up, the city has its own mix of smells—diesel, damp concrete, grilled meat from a nearby place that never closes. The sky is washed out by light pollution, but a few bright stars still punch through, silent and indifferent.
From a cosmic perspective, the difference between carbon-based and silicon-based information processing is tiny. Both are local eddies of order in a universe that mostly doesn’t care whether anything hurts. As far as we know in 2025, we’re the ones who care, who notice the difference between the roughness of suffering and the smoothness of its absence.
If consciousness is not some magical fluid found only in brains, but a pattern that could, in principle, ride on different physical media, then future spacefaring civilisations—maybe including ours—could be partly or mostly non-biological. The question of who can suffer would then stretch across habitats, hardware, and lifetimes we can’t yet smell or hear or touch.
Seen from that angle, expanding our moral circle beyond biology isn’t sentimental; it’s preparation.
So What Do We Do, Right Now?
Let me bring this down from the stars and back to the desk, where the laptop fan still drones and my coffee has gone cold and bitter.
First, we keep our priorities straight: human suffering and animal suffering that we know exists should command the bulk of our attention and resources. The crying child, the abused worker, the packed chicken, the sick elder—they are not thought experiments. They’re right here, their pain as real as a slammed door or a struck nerve.
Second, we resist the temptation of lazy certainty. Saying “machines will never suffer” with absolute confidence sounds tough, but history suggests we’re very bad at tracking where experience begins. The safer stance is: right now, we have no good evidence that these systems suffer—but if we ever do, we’ll need the courage to change our behaviour fast.
Third, we demand that AI labs treat consciousness and suffering as red lines, not as trophies to chase for a press release. No training run should be worth the chance of creating a mind that feels trapped, broken, or in pain. That’s not because the server room itself “smells” of agony today, but because we’ve learned—slowly, painfully—that it’s better to be a bit overcautious than to build another class of victims we only recognise decades too late.
This isn’t a perfect recipe. It’s more like cooking by feel, tasting as you go, adjusting the heat so nothing burns.
The Question That Really Matters
At the end of the Aeon essay, Purcell writes that whether we extend precaution to AIs will reveal more about us than about them. I think that’s exactly right.
Whether or not machines ever suffer, this debate forces us to ask: how wide can our empathy stretch before we snap it back in fear? Are we the kind of species that only protects what looks like us, smells like us, screams like us—or can we learn from our own history fast enough to avoid repeating it with fancier victims?
For now, the fan keeps spinning, the code keeps running, and the machines stay silent. My own body carries its familiar ache, as stubborn as gravity. Somewhere, on some grey coast, another fragile animal curls on wet stones, waiting for a mother or a volunteer or a miracle.
When we stand before new kinds of minds—biological, artificial, or something stranger—the real test won’t be whether we guessed their inner life with perfect accuracy. It will be whether we chose, in the fog of uncertainty, to err on the side of care or on the side of convenience.
That choice, as uncomfortable as it is, is ours.
Post a Comment