Latest / AI Ethics with Fexingo: Bias, Safety, and Responsible Artificial Intelligence / When Your AI Dating App Rejects You Based on Your Selfies
Transcript
- Lucas: Luna, have you ever wondered why you swipe right on some people and left on others? Luna: I mean, sure, it's usually a gut thing—photo, bio, vibe. But lately I've been hearing that the apps themselves are ranking us before we even get a chance to decide. Lucas: Exactly. And that ranking isn't just based on your bio or your interests. It's increasingly based on your face. We're talking about AI systems that analyze your profile photos and assign an attractiveness score—and that score determines who sees you, and who you see. Luna: So the algorithm is basically playing matchmaker with a beauty filter baked in? That sounds like a recipe for bias. Lucas: It is. And it's not hypothetical. A 2025 study out of Stanford looked at how AI models predict dating success—meaning, whether you'll get a match—based purely on profile photos. They found that the AI could predict success with 68 percent accuracy just from the images. No bio, no interests, nothing else. Luna: Sixty-eight percent—that's not trivial. But what does 'success' mean here? A match? A conversation? A date? Lucas: In the study, success was defined as receiving a match within the first week. And the AI's predictions were based on features like symmetry, skin texture, even the perceived warmth of a smile. The problem is, those features are heavily tied to race, age, and conventional beauty standards. Luna: So the AI is essentially encoding society's existing biases—and then amplifying them at scale. Lucas: Right. And it gets worse. Some apps, like Tinder and Hinge, have admitted to using a 'desirability score' internally—though they're cagey about the details. A 2023 investigation by The Markup found that users with higher scores were shown to more people, while those with lower scores were effectively shadowbanned. Luna: Shadowbanned from love. That's brutal. And the user has no idea it's happening. Lucas: Exactly. You're not being rejected by humans—you're being filtered out by an algorithm that's never been audited for fairness. And unlike a job application or a loan application, there's no recourse. You can't appeal your attractiveness score. Luna: It reminds me of that old Black Mirror episode, 'Nosedive'—where everyone is rating each other's social interactions. Only now it's real, and it's happening in the most intimate part of our lives. Lucas: Speaking of intimacy, there's another layer: privacy. These apps are collecting thousands of facial data points. And some of them, like Bumble, have faced criticism for using that data to train their models without explicit consent. Your face becomes a data point in a system you can't control. Luna: And if that data gets leaked or sold? Now your face is tied to your dating preferences, your location, your rejections. That's a goldmine for advertisers—and a nightmare for privacy. Lucas: It's a big part of why we keep this show ad-free, by the way. When your data isn't the product, you can have these conversations without worrying about whose interests are being served. If today's episode gave you something to think about, consider supporting us at buy me a coffee dot com slash fexingo. It keeps us independent. Luna: Absolutely. Every little bit helps us dig into stories like this without any strings attached. Lucas: So back to the dating algorithms. Some researchers argue that these systems could actually be designed to reduce bias—if they were transparent. But right now, they're black boxes. Luna: What would a more ethical dating AI look like? Lucas: A few things. First, opt-in scoring—not automatic. Let users decide if they want their photos analyzed. Second, regular audits for demographic bias. And third, a clear explanation of why someone is being shown or hidden. The EU's Digital Services Act actually requires platforms to explain their recommendation systems. But in the US, there's no such mandate. Luna: And until there is, users are left trusting that an algorithm they don't understand is making fair decisions about their love life. Lucas: Trust is a big word here. Because the incentives for the platforms are not aligned with fairness. They want engagement—more swipes, more time on app. An algorithm that surfaces conventionally attractive people might keep users hooked, even if it's not fair. Luna: So it's not just bias—it's an economic model that rewards bias. Lucas: Bingo. And that's the core tension. These companies are private, for-profit entities. They're not public utilities. But dating apps have become the primary way people meet partners. In 2024, over 40 percent of new relationships in the US started online. So these algorithms are shaping society's mating patterns. Luna: That's a huge responsibility. And one that's currently unregulated. Lucas: There are some efforts. A coalition of researchers called the Algorithmic Dating Accountability Project—ADAP for short—is pushing for transparency standards. They want apps to disclose if they use attractiveness scoring, and to allow users to opt out. But so far, no major app has signed on. Luna: Because why would they? If they tell users they're being ranked by their looks, people might leave. Lucas: Exactly. Ignorance is profitable here. But there's also a potential upside. Some smaller apps, like OkCupid, have experimented with showing users their own 'desirability score' as a way to gamify self-improvement. Critics say that's just adding anxiety to an already stressful process. Luna: I can imagine someone obsessing over their score, trying to take better photos, maybe even editing their face to fit the algorithm's preference. That's a slippery slope. Lucas: It is. And it raises the question: should we even be applying machine learning to something as subjective as human attraction? The Stanford researcher who led that 2025 study, Dr. Angela Park, said in an interview that 'attraction is a deeply human, irrational, and contextual thing. Reducing it to a set of facial features is not just inaccurate—it's a category error.' Luna: I love that phrase—category error. Love isn't a classification problem. Lucas: Right. And yet, these systems are being deployed on millions of people. The question is: what do we do about it? Regulation is one path. Another is user education—helping people understand that the algorithm is not a neutral arbiter of desirability. Luna: Maybe the most powerful thing is just knowing it exists. So next time I get a slow trickle of matches, I can wonder: is it me, or is it the algorithm? Lucas: Probably both. But at least now you know to ask the question.