"The fault, dear Brutus, is not in our stars, but in ourselves."

The newest panic over artificial intelligence is that chatbots are politically biased. A recent Washington Post test of several leading AI models found that many of them, including ChatGPT, tended to give more left-leaning answers to political questions than right-leaning ones. Conservatives who have suspected this for years felt vindicated. AI companies, predictably, insisted they are working toward neutrality.

But the deeper lesson is not that AI is biased.

The deeper lesson is that we are.

Artificial intelligence is not born in the wilderness, innocent and untouched by human culture. It is built by people, trained on human writing, judged by human evaluators, refined by human institutions, and released into a human marketplace. It is not an alien intelligence looking down at mankind from a cloud of pure logic. It is more like a mirror held up to the people and institutions that created it.

And what does it reflect?

It reflects the internet, which is not neutral. It reflects corporate media, which is not neutral. It reflects higher education, which is not neutral. It reflects Hollywood, New York, Silicon Valley, Washington, the Ivy League, Wikipedia, nonprofits, government agencies, professional-managerial culture, and the white-collar class that produces much of the “authoritative” text online.

That does not mean every answer is wrong. It does not mean experts have nothing to teach us. It does not mean conservatives are always right when they cry bias. But it does mean that the so-called neutral knowledge environment from which AI learns is already saturated with assumptions.

Those assumptions tend to be elite, credentialed, progressive, cautious, therapeutic, institution-trusting, and highly concerned with reputational safety. They are not always partisan in the bumper-sticker sense. Often the bias is subtler than that. It is the bias of the HR department. The bias of the faculty lounge. The bias of the nonprofit grant proposal. The bias of the “responsible” corporate memo.

That is why a chatbot may sound less like a campus radical than an Ivy-trained assistant deputy undersecretary for inclusive stakeholder engagement. It is not exactly left-wing populism. It is elite institutional liberalism.

The problem is not only political bias. It is also incentive bias.

A chatbot is trained to be helpful. But “helpful” and “correct” are not the same thing. If a user asks for a chapter-by-chapter summary of a book the model cannot actually verify, the truthful answer may be, “I do not have enough information.” But the useful-seeming answer is a polished outline. The model gets rewarded for producing, not necessarily for knowing when silence would be more honest.

That is a design flaw.

A wrong answer often looks more impressive than an honest limitation. A confident political answer often looks more satisfying than a careful map of competing values. A clean “both sides” answer may look balanced even when there are five sides, or when one side is asking the wrong question entirely.

This is where artificial intelligence becomes useful, not as an oracle, but as a sparring partner. The best use of AI is not to ask it to tell us what to think. The best use is to force it to argue with itself. Give me the conservative case. Give me the progressive case. Give me the libertarian case. Give me the religious objection, the populist objection, the constitutional objection, the working-class objection, the technocratic objection. Now tell me which facts are known, which are disputed, and which assumptions are doing the hidden work.

That is how AI can help us become less biased — not because it is unbiased, but because it can be made to expose bias from multiple angles.

The answer is not to pretend neutrality is easy. It is not. Nor is the answer to dismiss AI as useless. It is already too powerful for that. The answer is humility: from the companies building it, from the users relying on it, and from the institutions whose worldview it absorbs.

AI did not invent our blind spots. It learned them from us.

The fault is not in our models, but in ourselves.

(Contributing writer, Brooke Bell)