Emily M. Bender has spent her career trying to get people to ask better questions about artificial intelligence, so when an Anthropic engineer resigned and warned that AI companies are reckless and gambling with our lives, her reaction was more complicated than a simple amen. She agrees with the accusation: companies are being reckless. But she does not agree that the threat is best described as an existential, civilization-ending catastrophe in the way so many doomsayers imagine it. For her, the talk of superintelligence and runaway machines is not just a distraction; it is a failure of responsibility, a way of treating a man-made technology as if it were a supernatural force. She says she has spent her whole career trying to say the same thing from different angles. In 2021, she and her coauthors wrote the famous “Stochastic Parrots” paper for exactly this reason. OpenAI had claimed that GPT-2 was too dangerous to release. The hype around it was enormous. People were asking whether these models were becoming conscious, whether they would replace us, whether they would eventually decide we were irrelevant. Bender and her colleagues were saying: you are asking the wrong questions. A large language model is a statistical machine trained on human text; it can produce fluent paragraphs but it does not have beliefs, goals, or intentions. It is a parrot, not an oracle. The real questions are about who builds these systems, what harms they cause while being built and deployed, and who is accountable when something goes wrong. The word “reckless” is fair, she thinks, but not because a robot might one day decide to kill us. It is reckless because companies are releasing powerful tools into society without adequate testing, without meaningful regulation, without consent from the people who will be affected.
Her central analogy is devastatingly simple: “Can a bridge decide to collapse?” When a bridge falls, no one investigates whether the bridge was sentient, whether it had moral beliefs, whether it chose to fail. No one asks whether the bridge felt oppressed by its load-bearing duties or whether it somehow became conscious and developed a grudge against the city. Instead, we ask: who designed this bridge? Who approved the plans? Who skipped the inspections? Who used cheap concrete? Who ignored the warning signs? We look for human responsibility, because a structure does not fail on its own; it fails because people made decisions that made failure likely. Bender says the same should be true for AI. When a model discriminates against Black people, invents fake legal cases, helps someone build a dangerous weapon, or spreads harmful lies, we should not ask whether the model “went rogue.” We should ask who built it so flimsy, who rushed it to market, who failed to test it, who continues to use it even after the harms become visible. The rogue-AI story is not just technically wrong; it is morally lazy. It turns accountability into a mystery. It turns engineering failures into something that looks like a bad science fiction movie. While executives and politicians debate whether a future machine might decide to kill humanity, the actual machines already in our lives are making decisions that hurt people. That is not a predicted apocalypse. It is a current emergency.
What, then, is really existential? Bender is very direct about this. There are autonomous weapons powered by AI that are already being used in warfare, killing actual people in actual conflicts. There is the climate catastrophe, which is not a distant hypothetical but a reality measured in fires, floods, and heatwaves, and tech companies are building more and more energy-hungry data centers, training ever-larger models, and treating the environmental consequences as an externality they can ignore. There is the way bosses are using AI as an excuse to get rid of pesky workers, to surveil them, to deny them benefits, to grind them down. There is disinformation, which undermines democratic institutions and creates a fog in which people cannot agree on basic facts. There are face recognition systems that misidentify people and cause real harm. If we are worried about humanity’s future, these are the places to look. The machine-god narrative, in her view, is meant to distract us from all of this. It tells us that we should fear the machine, not the people who control it, not the investors who profit from it, not the ideologies that shape it. It treats AI as an inevitable force, like a tornado or an asteroid, rather than a product that could be regulated, contested, delayed, or redesigned. She describes the industry as a mix of cults, a mix of true believers, people who talk like preachers warning that the end times are coming. The singularity has been coming for decades now. It is always promised, always imminent, never actually arrives. And as a result, the real, concrete, present dangers get shoved aside.
The absurdity of “AI extinction” becomes clear if you ask for a step-by-step story. Bender cannot really envision it, and she is far from alone. Imagine a post that says: does an angry GPU just show up at my house or something? How exactly would a machine, even a superintelligent one, decide that the only way to accomplish a task is to kill everyone? Would it hypnotize you so that you do not unplug it? Would it somehow escape into the internet and take control of every system? People claim it could infect your machines. Yes, it could, but we already have malware, and we already know that infection is less a magical threat and more a question of security and vulnerability. The hard part is not a model choosing to do harm; the hard part is the very ordinary risk that things built by humans, with flaws and blind spots, will be misused or will fail in predictable ways. Meanwhile, the world is full of real threats that are already here: floods, fires, major disruptions to the global food supply chain, diseases, chemical weapons. Bender notes that not long after her conversation with the interviewer, Anthropic itself revealed that it had blocked several attempts by scientists to use its models to build biological weapons. That is a real safety concern. It deserves attention. But notice what it is: human beings using a tool. It is not a machine deciding to eliminate humanity. It is people doing dangerous things with technology, and the answer to that is oversight, security, regulation, accountability, and democratic control. None of those things are helped by the apocalyptic myth that the machines will rise up. That myth makes us look away from the people who build the weapons, create the conditions, and profit from disaster.
There is also a question of privilege. Bender says that there is another element where people are so privileged that they cannot imagine other things harming them. If you are not worried about police brutality, about floods, about warfare, about poverty, about being displaced from your job or your home, then you have the luxury to obsess over the machine-god. But most people living in the world do not have that luxury. They already live with concrete threats that can hurt them today. They do not have the time or the safety to spend years debating whether an artificial general intelligence will one day decide to unplug us. They are vulnerable now to the effects of automated decision-making, to the concentration of power in a few tech companies, to the spread of propaganda, to the violence that states and corporations can deploy with the help of surveillance and weapons systems. When someone in a privileged position says that AI will cause the end of civilization, they are often projecting their own fears onto a future they will never experience. The end of civilization is already here, or almost here, for people in the path of climate disasters, for communities targeted by autonomous weapons, for workers being replaced and abandoned. If you are safe from all of that, it might be easier to believe the problem is a rogue machine rather than the social and economic choices of human beings. But the choice to focus on a hypothetical apocalypse is itself a kind of blindness, a refusal to face the world as it is.
Why do we keep talking about this idea? People have been predicting the arrival of intelligent machines since the 1940s and 1950s. Bender has a game she likes to play: she gathers quotes about artificial intelligence from different decades and reads them to people, asking them to identify when each statement was made. Sometimes people get it right, but a lot of times they do not, because it all sounds the same. Every generation has its own version of the same warning: machines are getting too powerful, computers will think like us, technology will escape our control. The repetition is important. It shows that this is not a new insight or a a sudden discovery; it is a recurring cultural panic. It is also an excuse for a certain kind of behavior. It lets companies position themselves as saviors, racing to build AI while also promising to keep it safe. It allows them to draw attention away from the ordinary harms they are creating right now, and to present existential safety as a mysterious technical problem that only they can solve. The question the rest of us should be asking is not “When will the machines decide to kill us?” but “Who is building these systems, whose interests do they serve, and who is being harmed today?” The answer, Bender suggests, is human. The bridge did not decide to collapse. The people who built it flimsily are responsible. The same is true of AI. The machine is not gambling with our lives. Some people are. And the sooner we stop worshipping or fearing the machine-god and start asking uncomfortable questions about the humans behind it, the sooner we might actually do something about the real risks, real harms, and real power that are already in our midst.
