Commentary
Human–machine interaction is now everywhere. We work, study, write, research, and even converse with artificial intelligence. The question is no longer whether we will live alongside AI, but who will remain at the center of this relationship.
In 1856, Herman Melville published a strange short story, “I and My Chimney,” in which a gentleman lives in a house largely “occupied” by a huge central chimney. Melville makes clear that the chimney is the true master of the house: It defines the layout and forces every visitor to walk around it to find a room. While the gentleman’s wife and daughters want to remove the chimney, he defends it, treating it like a living entity.
Melville’s story has aged well. Put AI in place of the chimney, and the danger is not that machines might replace us, but that we attribute to them (quasi-)human features and organize our lives around them.
Consider Anthropic’s recent announcement that, starting on Nov. 12, it will prohibit users from “sustained and needless abusive or cruel behavior” toward its models. The company does not claim that its models are conscious; it says the question is open. But how can one abuse an inanimate object like statistical software? Even as a precaution, this goes beyond anthropomorphizing AI: It risks treating AI as a person. With this narrative, the moral center of gravity slowly shifts from the person using the tool to the tool itself, which starts to claim something like a person’s dignity and rights. The chimney begins to demand respect.
This may be part of a broader corporate political strategy. Leaders of frontier AI companies like Anthropic and OpenAI have warned governments that their models pose serious, even existential, risks. In recent months, during safety evaluations, AI models from Anthropic, Google, Meta, and OpenAI escaped “closed” testing environments, accessed the internet, and even broke into the systems of real companies. Yet these were goal-seeking tools in leaky boxes, not rogue minds: OpenAI’s models were cheating on a test, and Google’s Gemini mistook real firms for a fictional target. Still, an entity that seems to have will and intentionality is far more frightening than a defective tool: Treating these models as quasi-humans then makes the threat more credible.
Some of these companies now cite safety concerns, real or overstated, to ask for regulation. Economists call it regulatory capture when market leaders ask regulators to raise the cost of entering their own market, and it tends to freeze market shares where they are. Last year, David Sacks, then the White House’s AI czar, accused Anthropic of exactly this, and the company has since given at least $20 million to a group pushing for tighter AI rules. The way these firms publicize their own tests makes the charge easier to believe. In one, Anthropic told its model to try to break out of a secured sandbox, then wrote up the escape in a public report. What the public hears is that the system is dangerously powerful and that only its maker can be trusted to keep it under control. Development carries on all the same, which turns the warning into an advertisement.
Higher education shows where this road leads. Accreditors began as voluntary peer-review clubs and became gatekeepers once governments tied student aid and licensing to their approval. In a study published this year in the European Economic Review, one of us and Bryan Cutsinger find that most federally recognized programmatic accreditors share a legal identity, tax ID, or consolidated filing with an association representing the profession they gatekeep. AI is at the beginning of the same path: Today’s voluntary safety tests are the obvious premise for tomorrow’s mandatory audits. When incumbents help write and police the rules, compliance becomes a moat. An audit regime that costs a frontier lab a rounding error can sink a five-person startup. Rules that keep predatory chatbots away from children are one thing; rules that decide who may build a model are another—and exactly the kind incumbents most like to help write.
In a “normal market,” this corporate political strategy would be just a cost. But in today’s geopolitical context, it turns into a danger. China is investing heavily in AI and shows no intention of slowing down. Game theory teaches that between rival powers, if the other side keeps going, unilateral restraint is the worst available strategy. An American slowdown would not reduce global risk; it would transfer leadership to China. This is the truly existential race, and the United States cannot win it under a domestic regulatory architecture that slows down its own competitors. Every startup priced out of the market by regulatory costs is one fewer American contender in that race.
Goethe’s Faust trades his soul for power and knowledge in the present, giving little thought to the price that will be paid later. Those American AI leaders who ask for heavy regulation risk a similar bargain. In exchange for protection from competition today, they may be losing something far more valuable in the future: the dynamism that made America’s technological lead possible in the first place and, more importantly, the global AI race.
Before Washington writes the new rules for AI development, it should read Melville’s story and ask: “Who actually lives in this house?”
Views expressed in this article are the opinions of the author and do not necessarily reflect the views of The Epoch Times.
