Artificial intelligence has been justifiably under fire in recent weeks for the myriad dangers it represents.
At around the same time that a researcher for Anthropic predicted the odds of human extinction from AI were “greater than 10 percent,” the company admitted it had blocked research that could lead to AI-engineered bioweapons. China’s top spy chief has warned his country’s leaders of AI-powered security threats, like cyberattacks, misinformation, and widescale espionage. Even Elon Musk, an otherwise reckless techno-maximalist, has backed the calls of top AI CEOs to slow down AI development due to its risks.
But as policymakers get serious about AI safety, they shouldn’t neglect another fundamental problem with AI: bias—and in particular its bias against women.
Perhaps this problem sounds trivial compared to the existential threats posed by AI that are dominating news coverage of late. But it’s exactly its low-key pervasiveness that makes bias so insidious and so damaging.
Don’t take my word this problem exists. Just ask AI.
When I asked ChatGPT if it’s biased against women, its answer was an unequivocal “Yes.” “Not because AI inherently ‘hates’ women,” ChatGPT said, but because “AI systems can reproduce, or amplify gender biases that exist in their training data, design choices, and societies that produce that data.”
Claude as the ultimate tech bro? This isn’t a harmless quirk.
Bias shapes societal attitudes by perpetuating stereotypes that limit women’s freedom. Here’s what Gemini had to say about that: “Text-generation models frequently associate men with career success, business, and leadership, while linking women to domestic roles, family, or supportive positions.”
Here’s an obvious example (from ChatGPT):
Bias can also sabotage women’s future opportunities, as more employers and institutions adopt AI. For instance, according to Gemini, automated hiring systems tend to favor male candidates: “Studies have shown that when Al evaluates identical qualifications, it often rates female candidates—particularly older women—as less experienced or qualified than male peers.” Algorithms also tend to assign women lower credit scores and higher interest rates, which means women will pay more to their mortgage lender chatbots if they ever want to own a house.
AI chatbots may even treat users differently, depending on their gender. A new study from Johns Hopkins University finds that when women ask AI to draft emails and job applications, its responses are less sophisticated than what it generates for men. “You’ll get back a response that’s less complex, at a lower grade level, and less formal,” said senior author Anjalie Field in JHU’s HUB.
As ChatGPT notes, this bias isn’t intentional but a product of the internet’s overall tilt toward men. As Shubhi Rao, CEO of the AI startup Uplevyl explains, AI is simply the sum of the internet, and the vast majority of its content is male-centric. For example, “if you look at any publications, whether it’s published journals in medicine or engineering or psychology or any field for that matter, men publish more than women do,” Rao says. “Of course the internet reflects that.” Add to that thousands of hours of Joe Rogan podcasts, “looksmaxxer” videos, and billions of posts from Reddit. (Reddit is a principal source of AI training data, and 60 percent of its users are men). The manosphere isn’t just part of the Internet; the Internet is the manosphere.
“If you are pulling information from two baskets, and one basket is filled to the brim with data that is very skewed towards men, and the basket is only a third full of data that truly addresses the needs of women, we know that it’s going to be over-indexing toward one gender over the other, statistically speaking,” says Rao.
The balance of content is likely to become even more skewed over time, thanks to the Trump administration’s policies. It has systematically sought to erase government data on women’s health and well-being as part of its campaign against “DEI.” The nonprofit Dataindex.us has identified more than 400 federal datasets that have been deleted so far, including data on maternal and reproductive health, wages, and other information relevant to women.
The loss of this data means less research, analysis, and secondary content; less information for women to guide their lives; and less raw material for AI to ingest, thereby reinforcing existing biases. This dearth of data will be especially problematic as developers created “vertical” LLMs that specialize in particular domains, like law or medicine.
Rao, however, is doing her part to combat AI’s gender bias. Her company, Uplevyl, is the world’s first female-forward AI startup, and one of her goals is to build vertical LLMs, such as for financial planning, that take women’s needs into account. You can learn more about her approach and about the nature of AI bias in this week’s interview on our Substack page.
Don’t miss at the Monthly…
Follow the money; a Blue Wave is brewing. Pre-election polling is often just so much noise. That’s why Politics Editor Bill Scher counsels watching to see where the parties are spending if you want a true picture of the upcoming midterms. The evidence shows, Bill says, that a Blue Wave may indeed be on its way. “Democrats are forcing Republicans to defend considerably more territory than Republicans are forcing Democrats to defend,” Bill writes. “When we take into account states where other independent expenditure operations are spending at least $1 million on an underdog, the map expands to include Kansas and—maybe the biggest surprise—Mississippi.” Read here.
Justice Roberts’ flawed embrace of the unitary executive theory. The Supreme Court has relied on the so-called “unitary executive theory” to justify granting Trump ever more discretion and power. The Court’s conservatives have also prided themselves on their understanding of constitutional history to justify their advocacy of this view. But as Stanford University historian Jack Rakove argues, these justices are flat out wrong. “Neither the Constitutional Convention nor the First Congress presented a view of executive power as coherent or expansive as the apologists for the UET have been proclaiming,” he writes. The Framers were, in fact, deeply ambivalent about the scope of power that should be vested in the presidency. Read the full analysis here.
Is reading really dead? Those who despair of Americans’ declining literacy often point to falling reading scores among the nation’s school kids or to survey data indicating that about four in 10 Americans failed to pick up a single book in 2025. James Marriott’s new book, The New Dark Ages, warns of a “post-literate” era when reading is an obsolete skill. But reviewer Christoph Irmscher questions Marriott’s premise. “Evidence of decline is not proof of imminent extinction,” he writes. “In fact, some of the media that seem destined to destroy reading may in fact be helping to revive it.” (Christoph also notes the irony of a highly literate book on the decline of reading that by its very existence assumes it will be read.) Read here.
Plus…
- Clara Bingham reviews Gloria Steinem’s last book, An Unexpected Life. “In Steinem’s absence, we can take solace in this jewel of a book whose pages spill over with her generosity, big-hearted wisdom, and true-north moral compass. Given today’s misogynist political climate and backlash against feminism, An Unexpected Life couldn’t come at a better time,” Clara writes.
- Merrill Goozner notes that the Census Bureau’s relatively rosy numbers on the uninsured, released last week, don’t reflect the devastating cuts to health care access included in the “One Big Beautiful Bill” last summer.
- Carter Dougherty condemns a potential settlement over credit card “swipe fees” that will bring consumers little relief from high costs.
- Matt Krupnick finds that—as predicted—the federal government’s new limits on graduate student borrowing are limiting opportunities for aspiring health care workers (this piece comes to us courtesy of The Hechinger Report).
- In his second column of the week, Bill Scher writes about another reason a Blue Wave might be coming: Americans’ disposable incomes are down. History shows that when Americans’ feel worse off, they tend to punish the party in power.
Parting note…
Both political parties have been paying a lot more attention to child care—a positive development overall, given the decades of neglect on this topic.
Much of the focus, however, has been on young moms with babies (in part a consequence of the right-wing’s obsession with tradwifery and “pro-natalism”). Still overlooked are the needs of older mothers—especially those bearing the double burden of caring for aging parents.
As a new AARP survey of “sandwich generation” caregivers finds, nearly 17 million Americans—average age 42—are both raising kids and caring for an adult with a complex medical condition or a disability. They’re putting in an average of 27 hours a week, often at significant cost to their physical, financial, and emotional health. According to the AARP, 67 percent report a “work disruption due to caregiving,” and 22 percent have experienced “high financial strain.”
Help with child care costs would be a boon to this group, but policymakers should also explore expanding aid for eldercare. In the short term, this includes restoring protections for immigrants, who comprise a significant share of home health care workers and whose deportations are causing a staffing crisis. In the long term, the nation must tackle the problem of affordable long-term care, which many Americans may be surprised to learn that Medicare does not cover.
While the likelihood of Congress creating a new “Medicare LTC” is remote at best in the coming two years, the graying of America means the problems faced by the sandwich generation will soon be too big to ignore.
For those of you who are new to the Monthly, we’re a long-running (57 years!) nonprofit, independent publication. We elevate ideas and advance good governance (yes, we still believe in such a thing).
If you like what you’ve read, please consider subscribing, following, and sharing with your network. We depend on readers like you to be part of our community and keep us afloat.
Have a great week!
Anne Kim, Senior Editor
