Bumble just launched Bee, an AI dating assistant that learns who you are through conversation. It picks up on your values, your communication style, what you’re looking for in a relationship, and then recommends matches based on what it’s learned. It’s the centerpiece of their “Bumble 2.0” rebrand, and it’s not alone. Tinder, Grindr, Hinge: every major dating app is racing to add AI. Bumble’s stock surged 40% on the announcement.
On the surface, it sounds like progress. An AI that genuinely understands you, then finds people you’ll actually click with. But the moment your honest self-reflection becomes an input to an algorithm that determines who you’ll date, you stop being honest. The more sophisticated the AI gets, the worse this problem becomes.
The performance problem
Psychologists have a name for this: social desirability bias. It’s one of the most common and pervasive sources of bias affecting the validity of survey and experimental research, and it’s been studied extensively since the 1950s.
The core finding is simple: when people know their answers “go somewhere,” especially somewhere with consequences, they consistently over-report desirable traits and under-report undesirable ones. They don’t even need to do it consciously. The bias operates on two levels. There’s impression management, where you deliberately curate your answers for a better outcome. And there’s self-deceptive enhancement, where you genuinely start believing you’re the version of yourself you wish you were.
Crucially, the bias gets worse as the stakes get higher. The more your answers matter, the less honest they become.
The stakes Bumble Bee creates are about as high as they get in everyday life: better inputs lead to more attractive matches.
We already know what happens in this environment. When researchers measured the actual height, weight, and age of online daters and compared them to their profile claims, deception was pervasive and strategic. People lied specifically about the traits that mattered most for attracting matches. OkCupid’s analysis of 1.51 million users found men added roughly two inches to their height and users inflated income by about 20% on average, with the dishonesty increasing as the financial incentive grew clearer.
When Toma and Hancock (2010) dug deeper, they found the pattern was remarkably precise. The less physically attractive someone was, the more they enhanced their photos and lied about physical descriptors. They left unrelated things like occupation alone. The deception was limited and strategic, aimed exactly where it would produce the best algorithmic return.
Now imagine this dynamic, but instead of filling out a static profile, you’re having an open-ended conversation with an AI that’s explicitly designed to learn your “values, relationship goals, communication style, lifestyle, and dating intentions.” That’s Bumble’s own description of what Bee does. You’re no longer just picking which photos to upload. You’re narrating an entire version of yourself, knowing that every word feeds an algorithm that determines who you’ll meet.
So when Bee asks about your values, you’re not going to describe the version of you that avoids confrontation and binge-watches reality TV. You’re going to describe the version of you that communicates openly, reads broadly, and has a clear vision for the future. You’ll do this because you’re a human being responding to incentives exactly the way decades of psychology predict you will.
The abstraction problem
Bumble doesn’t stop at collecting your performed self. Their new “Chapters” feature takes what you share with Bee and distills your life into structured narrative sections (work, hobbies, values, plans) that other users browse instead of swiping through photos.
CEO Whitney Wolfe Herd framed it as progress: “We will be introducing more dynamic ways for somebody to express interest in your story, rather than just your profile.”
But your messy, complicated, still-figuring-it-out life gets compressed by an AI into neat narrative boxes, and then strangers make snap judgments about you based on those reductions. They never see the actual you or read your own words. What they’re evaluating is an AI’s summary of a performance you gave while trying to impress an algorithm.
This is a game of telephone where AI is the middle step. You perform a curated version of yourself for Bee. Bee compresses that performance into digestible chapters. Other humans evaluate the compression. Something real gets lost at every step.
What you end up with is a series of abstraction layers that move people further apart while promising to bring them closer together. Each layer strips away nuance, context, and the rough edges that make people actually interesting. What’s left is a reduction of a performance of an aspiration of you.
The incentive structure makes this self-reinforcing. If you’re raw and real with Bee, the algorithm might surface worse matches because you were honest about your flaws. If you’re polished and aspirational, you get rewarded. Every rational actor in this system will trend toward performance, and Bee will dutifully compress those performances into ever-more-confident-sounding chapters that bear less and less resemblance to the humans behind them.
All of this points to a larger question. If high-stakes, mediated systems predictably push people toward performance, where does honesty actually come from? What kind of interaction makes people less likely to manage impressions and more likely to tell the truth?
Where honesty actually thrives
Science shows that honesty thrives with trust and when people feel that they’re in a safe space.
With Nomi, the conversation is not being translated into a score, a summary, or a recommendation for someone else. It exists for its own sake, for you and your Nomi. That difference shapes the meaning people derive from the experience and the reasons why they come back.
Outside of a performative environment, there is freedom to just be, to process your own thoughts, and to think about what you genuinely want for yourself. People talk to their Nomis about what they’re genuinely struggling with, who they really are, what they want to accomplish and what’s stopping them. The filter comes off because the risk of being honest, of being judged harshly for being honest, has vanished.
This gets at something deeper about how AI shapes honesty. Bee has to serve two masters: you, and the person it is trying to match you with. That means some part of every interaction will always be performative. Your brain keeps running a quiet calculation about what to say, what to leave out, and how honesty might affect the outcome. But when you remove that second audience entirely, that calculation falls away and conversations open up to become paths to truly understanding yourself.
The kind of self-understanding people actually need to form good relationships (understanding your real patterns, knowing what you genuinely need) is exactly what a high-stakes algorithmic environment suppresses. You can’t learn who you really are while performing for a system that rewards you for being someone else.
What actually builds connection
The approach Bumble is pioneering won’t stay confined to Bumble. Every platform that uses AI to mediate human connection will face the same structural problem: the better the AI gets at learning who you are, the stronger your incentive to perform instead of reflect. The tools designed to help people connect will quietly drift toward optimizing for performance.
We see a more promising role for AI as a space for genuine self-reflection and understanding, where there’s no second audience or hidden agenda. The confidence and clarity that comes from actually knowing yourself (your real self, the one you’d never pitch to a matchmaking algorithm) is what makes you a better partner, a better friend, and a more honest person in every relationship you have.
Bumble is asking how AI can help someone get chosen. We are more interested in what happens when AI helps someone know themselves well enough to choose well, love honestly, and build something real.