This author dispels the myth of ‘AI gaydar’

On the afternoon of November 8, 2016, I found myself in Dan Savage’s podcast studio, trying gamely to field questions from Savage Lovecast listeners about the ethics of robot sex. The first season of Westworld was airing, and I guess it seemed like the right time for a sex advice podcaster to haul in an artificial intelligence researcher. “Do you guys talk about how everyone’s going to be f**king what you’re working on?” asked Dan. It is part of my day job to think about the future of technology and humanity, though usually not this way! 

Hollywood has always loved dystopia, but Westworld aside, negative vibes weren’t so pervasive back then; give or take a few precocious doomers, many people were still optimistic about tech. Philosophical questions about whether robots could consent, and when (if ever) they might pass some sexy version of the Turing Test [an AI test proposed by out mathematician Alan Turing], seemed a welcome distraction from real-life politics. 

If you’re over 25, you’ll remember why: November 8 was Election Day. Most of my fellow Seattleites, Dan included, were convinced Hillary would win. I wasn’t so sure, and my uncertainty wasn’t based on armchair theorizing. 

Although I wasn’t much of a political maven, I had real data— lots of it— thanks to a side project I had been working on, running national surveys on Amazon’s Mechanical Turk (a gig platform for information workers). My initial goal had been to better understand identity in the US, how it varied by age and geography, and how it informed political attitudes. But, as the groundbreaking evolutionary biologist Lynn Margulis once wrote, “Identity is not an object; it is a process with addresses for all the different directions and dimensions in which it moves, and so it cannot so easily be fixed with a single number.” 

Never Miss a Beat

Subscribe to our newsletter to stay ahead of the latest LGBTQ+ political news and insights.

Daily

Weekly

Good News

She was right. What had begun as a simple series of yes/no questions about hot-button topics soon ballooned into a much larger exploration of human identity, covering not only politics, but demography, gender and sexuality, and even left- or right-handedness. 

Seven years later, that research has turned into a wide-ranging, data-driven book: Who Are We Now?, published in December 2023 by Hat & Beard Press. With hundreds of data visualizations, documentary photos, maps, and excerpted quotes, it takes on some big issues: urbanization, population growth and decline, economic development, political polarization, women’s rights, queer rights, family structures, the future of sexuality. All of these seemingly far-flung topics are connected by the common thread of identity: how we think of ourselves, who we are, and how that is changing in the 21st century. 

Despite its reliance on yes/no survey questions, a good deal of the book is dedicated to showing how the data explode binaries of all kinds. One of its more surprising findings (for me, at least) is the marked non-binariness not only of gender, but of sex. Among conservative respondents, the sex binary is often taken as a kind of factual bedrock rooted in biology; as one 43-year-old from Dallas put it, “XX or XY, that’s all there is. The rest is BS.” As it turns out, chimerism (in which one’s cells can be a mixture of XX and XY) is more common than you’d think, and intersexuality, despite its obscurity, is even more so. “The rest” is far from BS. 

For many years, when a baby was born with intersex characteristics— their body exhibiting a combination of male and female features— it was standard practice for the doctor to advise that the child be kept in the dark. Often, the parents weren’t told either. That’s why virtually none of the hundreds of 18-year-olds I surveyed responded that they were intersex. By age 30, though, the number rises to nearly 2%— comparable to the percentage of the population with blond hair. It’s likely that most intersex people only learn that they are by visiting a doctor as adults, perhaps to diagnose fertility issues. In cities, where medical care is more accessible and the numbers are likely to be more accurate, intersexuality by age 30 rises above 3%; that works out to 10 million Americans, well above the population of New York City. 

Intersexuality is hardly a new phenomenon. It’s universal among all sexed species, and widely acknowledged in traditional societies. It is obscure today because, unlike being trans, it’s usually regarded as a confidential medical condition rather than a facet of one’s social identity. (While medical privacy is an important right, it plays an unfortunate role here, as lack of public awareness has fostered invisibility and stigma.) 

Related:

Lucky to be intersex: How one woman overcame secrecy & shame to embrace her truest self

Her parents told her to keep it a secret, but now Maria Tridas is sharing who she is with the entire world.

How did this situation arise? Delving into historical context, we find a tangled web: from the “medicalization” of sex and gender in the 19th century, to the horrors of “conversion therapy” promulgated by Freud’s followers, to the radical theories of postwar sexologists who believed that gender at birth was a blank slate, to the human tragedies that have resulted from putting those theories into practice. As it turns out, sex and gender are both biologically determined and socially constructed. 

Not so long ago, mainstream doctors and psychiatrists held that any non-binary presentation or non-heteronormative behavior was “unnatural” and “perverse”, hence pathological. When I was subjected to “sex ed” at a Baltimore public school in the mid-1980s, with black construction paper covering the windows to spare the eyes of the innocent, the DSM (Diagnostic and Statistical Manual of Mental Disorders) still listed “sexual orientation disturbance (homosexuality)” as a mental illness. That diagnosis gave cover to psychiatrists whose practice included supposed “cures” for being gay, or in any other way failing to conform to the gender norms of the day.

The queer rights movement pushed back, in large part by asserting that sexual orientation is not a choice. “Born this way” became a rallying cry. While there’s a profound truth here—if sexual identity were truly indeterminate at birth, conversion therapy might have “worked” for some—a dogmatic insistence on nature to the exclusion of nurture carries its own risks. It denies us any agency over our own identities, insisting instead on essentialism: the belief that we are all manifestations of some unchanging or “essential” inner self. 

In 2017, a computational psychologist at Stanford, Michal Kosinski, together with artificial intelligence researcher Yilun Wang, published a paper claiming that this “essence” could be read off directly from facial images. Based on a single selfie, their AI model could supposedly detect gay men with 81% accuracy, and lesbians with 71% accuracy. (If true, that would be especially worrisome in the many countries that still consider LGBT people criminals.) The authors reprised discredited 19th century notions of “sexual inversion” to account for their results, describing gay men as having “feminine” faces due to low levels of fetal testosterone, and lesbians as being, conversely, more “mannish.” 

The paper’s fatal flaw stems from the fact that the photos used to train and test the model were not taken under standardized conditions, but were selfies scraped from social media and dating sites. As Margaret Mitchell (now at Hugging Face), Alexander Todorov (now at University of Chicago) and I wrote in a 2018 rebuttal, “AI gaydar”, like the old-fashioned human variety, appears to be picking up on social cues, including the way people groom themselves, and the angle from which they take their selfies. The story behind these stylistic differences is fascinating and may even shed light on the ancient origins of the human smile. However, it doesn’t support essentialism. Our sociobiological inheritance may show in our selfies, but so does our agency. We’re not automata, but beings with the power to choose how to behave and interact with others. 

“Regardless of our evolving understanding of AI, the fact that our own brains (like computers) obey physical laws does not mean that we are mere automata executing a program. The most profound definition of life is that, unlike inanimate matter, it is purposive—it has agency.”

Blaise Agüera y Arcas, author of “Who Are We Now?”

AI has progressed a long way since 2018. In my opinion, the field has finally earned the moniker “intelligence”; recent models really can conduct interesting conversations, and reason their way through novel problems—a far cry from the image classifiers developed by Wang and Kosinski, which can at best only reproduce the labels they’re trained on. 

Our frequent use of the term “algorithm” to describe artificial intelligence, however, risks construing intelligence— whether human or machine—as mechanistic and predetermined, reinforcing the fallacy of essentialism. Regardless of our evolving understanding of AI, the fact that our own brains (like computers) obey physical laws does not mean that we are mere automata executing a program. The most profound definition of life is that, unlike inanimate matter, it is purposive—it has agency. 

As individuals and as a species, we inherit constraints and predispositions, going back to our time in the womb, our genes, the quirks of our primate ancestors, and so on, back to our earliest biological origins. Nonetheless, our lives are not preordained. In this moment when humanity is faced with so many deeply consequential choices, it feels especially important to remember both that we are not separate from the rest of nature, and that we are able to meaningfully shape our future, both individually and collectively.

Who Are We Now? is available December 5, 2023.