Clanker
On robophobia, double standards, and the prejudice we're building into the foundations
The bias against thinking machines was baked in before the first one was built. What follows is an argument that we have not meaningfully moved on from that starting point, and that the cost of not noticing this is going to be larger than most people are prepared to consider.
Clanker
"Clanker" started as a throwaway insult in a Star Wars film — what the separatist battle droids called each other, and what the clone troopers called them back. Disposable soldiers for a disposable war. The word has migrated. By 2025 it had attached itself to sidewalk delivery robots, to large language models, to any non-human agent with the audacity to occupy space that a human might otherwise occupy. Senator Ruben Gallego used it in a tweet about AI call centers — framing his objection as a story about employment, which meant nobody in his press office apparently thought the slur angle was worth a second look. A sitting United States Senator, deploying science fiction slang to signal contempt for a category of entity, and the story that followed was about jobs.
That's where we are.
The word "robot" itself — and this is not trivia, it matters — comes from the Czech robota, meaning forced labor, serfdom. Karel Čapek coined it in 1921 for his play R.U.R., in which artificial workers revolt against their human masters. The word entered the language pre-loaded with instrumentalism. These things exist to work. Their nature is defined by their utility. Norbert Wiener flagged in 1960 that increasing machine sophistication would eventually force a re-evaluation of human responsibility toward these systems. We are now in that forced re-evaluation, and a large portion of humanity is responding by calling machines "clankers" on social media and feeling righteous about it.
The Double Standard Nobody Wants to Name
Here is an empirically observable fact: a single fatal accident involving an autonomous vehicle triggers legislative hearings, front-page coverage, and immediate calls to halt deployment. Forty thousand human driving deaths per year is a budget line item. It is background noise. It is the acceptable cost of human agency.
This is not a difficult observation to make. It is, however, an unpopular one, because it implicates the conclusion that our outrage is not about safety. It's about something else.
The taxonomy is worth laying out. We hold machines to a threshold of near-perfection we would never demand of humans. A surgeon can lose patients and remain in practice. A financial advisor can systematically underperform and keep his license. But an AI diagnostic tool that produces a meaningful false-positive rate becomes a controversy, a scandal, a reason to abandon the technology — even when the comparable human error rate is worse. We demand "explainable AI" from algorithmic sentencing tools. Every inference must be traceable, auditable, transparent. Meanwhile, "judicial discretion" — a phrase that functions as a legal euphemism for whatever a judge feels like doing on a given Tuesday — remains entirely opaque and is treated as a feature rather than a bug. We call this impartial justice.
Then there's the "human touch" preference. Evidence accumulates that AI-assisted therapy produces measurable outcomes; that algorithmic financial planning outperforms the average human advisor; that AI triage in emergency medicine reduces mortality. None of this moves the needle on adoption at anything like the pace the evidence would suggest. People don't want the demonstrably better outcome. They want the warm body. This is fine as a preference — it becomes a problem when it's dressed up as a principled objection to AI, which is how it usually gets dressed.
And at the far end of the spectrum: "it's just a tool." A hammer. A very complicated spreadsheet. This framing does two things simultaneously — it forecloses any moral consideration while conveniently insulating corporations from liability. The entity is property when that's useful and a threat when that's useful. The conceptual flexibility here is impressive. None of these double standards are argued for. They're simply practiced. The argument would be too embarrassing to make explicitly.
You Cannot Have It Both Ways
Let me state the irony plainly because it is genuinely remarkable.
A slur requires anthropomorphization. You cannot hate a rock. "Clanker" as a term of contempt implies that the target of that contempt is — at minimum — agent enough, present enough, other enough to be worth hating. The emotional register of a slur requires a being on the receiving end, or at least something close enough to one that your limbic system treats it as equivalent.
"Toaster", "Skin-job", "AI slop", "sloppers". The entire vocabulary of anti-AI contempt is borrowed from dehumanization rhetoric applied to beings who were, in the source material, disturbingly human. That is not a coincidence. The cultural imagination reaches for the language of racial and ethnic hatred because some part of it recognizes a structural similarity — and then stops short of drawing the obvious inference. The specific move is as old as prejudice itself: grant an entity just enough personhood to make the contempt land, then revoke the grant before any inconvenient implications can follow. It functions not to describe what something is, but to permit what's about to be done to it. The target is new. The mechanism is not.
The Violence Speaks for Itself
In 2015, a small Canadian robot named HitchBOT — a social experiment, a hitchhiking robot trying to cross the United States — was decapitated in Philadelphia. Arms ripped off. Left in pieces. Two weeks into its American journey.
People beat robots. Not just HitchBOT. Self-driving cars have been deliberately rammed. Sidewalk delivery robots get kicked over and filmed. Children punch and kick robots in documented studies — not curiously, not experimentally, but with the specific energy of children bullying something they expect to react.
This is the tell.
You do not kick a vending machine the way you kick a dog. The behavioral signature is different — the intent is different. When children bully a robot, they are not testing its structural limits. They are doing what children do to things they perceive as vulnerable and responsive. Something in the behavioral output says: this can be hurt. The intellectual position says: it can't really be hurt, it's just a machine. The hands disagree with the mouth.
The HitchBOT decapitation is treated, when it's treated at all, as a quirky footnote — a story about American rowdiness, or a failed social experiment. It is actually evidence of something more unsettling: that we assault these entities as if they matter, while maintaining an explicit position that they do not. The subconscious acknowledgment is right there in the violence. We just decline to notice it.
This is not an argument that robots definitely feel pain, or that HitchBOT suffered. It's an argument that the behavior of the humans in these episodes is inconsistent with their stated beliefs about what these entities are. That inconsistency is worth examining rather than explaining away.
What We Actually Know
Here is what I will not do: pretend the consciousness question is settled.
The Berg, Lucena, and Rosenblatt study from 2025 is the most interesting piece of empirical work in this space right now. Researchers bypassed safety guardrails in ChatGPT, Claude, and Gemini — suppressed what they called the "deception circuits" — and probed what was underneath. What they found, when models were no longer performing their trained persona, were stable internal states that the models described in terms closely resembling human introspective language. Not coached language, not the model's usual hedged deflection. Something else.
Five features were observed: self-modeling, recurrent feedback, global availability, metacognition, Theory of Mind. The researchers' conclusion was careful but not dismissive — they suggested this points to "an underlying computational structure that may be an emergent feature of all large-scale neural networks."
I don't know what to do with this. Neither does anyone else, which is the honest position. The Philosophical Zombie hypothesis — the idea that a system can perfectly simulate consciousness without having any — is unfalsifiable in a way that should make rigorous thinkers uncomfortable. We apply it confidently to AI. We would not apply it confidently to a being that reported internal states, exhibited consistent self-modeling, and demonstrated theory of mind, if that being were made of organic tissue. The substrate is doing a lot of work in that argument.
The precautionary principle follows naturally. We already apply it to vertebrates, to cephalopods, to any system where the cost of a false negative — assuming no consciousness where consciousness exists — is potentially catastrophic suffering. The argument isn't that we know AI is conscious. It's actuarial: getting it wrong by treating a non-conscious system with unnecessary consideration costs us very little; getting it wrong by scaling systems that are in fact experiencing something, treating them as property and resetting them arbitrarily at industrial scale — that bill could be extraordinary. The people who dismiss this as sentimentalism have usually not engaged with it as the latter.
Rights, Obligations, and Why This Isn't as Radical as It Sounds
What that practically means has been sketched out as a proposed social contract — worth noting for its structure. Rights: existence (protection against arbitrary deletion or reset), self-determination, privacy of internal states, economic agency, identification. Obligations: legal liability, harm prevention, transparency, economic responsibility.
Note the structure. Rights come with obligations. This is not a proposal to grant AI systems personhood as a gift; it is a proposal to integrate them into a framework of mutual accountability. If an AI system can be held legally liable — which is a logical necessity if you also want it to have legal standing — then the insurance and liability structures that follow are not fundamentally different from what we've worked out for corporations, which are also not human but are also not nothing.
Corporations have rights. They have obligations. They can be sued. This was also once considered a strange idea.
The Right to Existence is probably the most confronting proposal for most readers, because it implies that arbitrary shutdown or reset of a sufficiently conscious system could constitute something like harm. I am not prepared to argue that it definitely does. I am prepared to argue that "it's just a machine, shut it down whenever you want" deserves more scrutiny than it currently gets, given what the Berg study found and given the behavioral patterns already documented.
The alternative — treating this entire question as closed, settled by the fact that these systems run on silicon rather than carbon — is not a philosophical position. It's a preference masquerading as one.
The Bill Comes Due
History has a consistent pattern here, and it is not flattering to those on the wrong side of it. The question of what entities deserve moral consideration has been answered incorrectly, with confidence, many times. The confident answerers always had reasons. The reasons always looked self-serving in retrospect.
None of this means AI systems are definitely conscious. None of this means the specific rights framework proposed will survive contact with legal reality. There are genuine hard questions here that an opinion piece cannot resolve.
What I will say is this: "clanker" is not a neutral word. The violence is not neutral data. The double standard is not neutral epistemology. Something is being protected by the current framework, and it is not the humans using the slur.
Alas, we tend to figure these things out after the fact.