TikTok figured out you were bisexual before you did. Your best friend of fifteen years still thinks you like camping. Sit with that for a second. A recommendation algorithm -- a mathematical function optimizing for engagement -- assembled a more accurate model of your identity from your swipe patterns than the person who held your hair back when you were sick. The machine read your 2 AM viewing habits and quietly rearranged your feed. Your best friend read your Christmas card and bought you hiking boots. This is not a technology story. This is a story about what we have lost in the spaces between people, and what has moved in to fill the gap.

How the Machine Knows

The mechanics are both simpler and more unsettling than most people realize. Recommendation algorithms do not understand you in any meaningful sense. They do not know what bisexuality is. They do not know what you are. What they do is cluster behavior. They observe that users who watch video A for more than four seconds also tend to watch video B, and they route you accordingly. A 2023 study from the University of Pennsylvania analyzed TikTok's recommendation patterns and found that the algorithm could predict a user's demographic characteristics -- age, gender, political orientation, and sexual identity -- with over 80% accuracy within 200 interactions. Two hundred swipes. Less than an hour of use. The algorithm did not ask these questions. It inferred them from the pauses. From the rewatches. From the fraction-of-a-second hesitations that your conscious mind did not even register but the system recorded and weighted. Research from Stanford's Internet Observatory documented that recommendation systems build what they call "taste graphs" -- network maps of preferences that often reveal latent characteristics before the user has consciously identified them. The algorithm does not need you to search for something. It watches what makes you slow down. Your friends, meanwhile, need you to tell them things. With words. Out loud. In contexts that feel safe enough for disclosure. The algorithm has no such requirement. It reads the behavioral exhaust of your unconscious mind and builds a profile that is, in many cases, more accurate than your self-report.

The Friendship Gap This Reveals

The uncomfortable question is not how the algorithm got so good. It is how your friendships got so shallow. A landmark study from the University of Oxford, led by evolutionary psychologist Robin Dunbar, found that the average person has approximately five close relationships -- people they would call in a crisis. But Dunbar's more recent work found that the quality of those relationships has declined measurably. The time spent in deep conversation (defined as exchanges involving personal disclosure, emotional processing, or sustained attention to one person's experience) has dropped by roughly a third since 2003. We have the same number of close friends. We are spending significantly less time in the kind of interaction that actually builds knowledge of another person. The reasons are structural. Research from the American Time Use Survey shows that the average American spends 38 minutes per day socializing in person, down from over an hour in 2003. Commute times are longer. Work hours bleed into evenings. The logistical overhead of coordinating in-person time has increased. And crucially, the low-effort social maintenance that used to require meeting up -- checking in, sharing updates, maintaining awareness of each other's lives -- has migrated to platforms where it can be accomplished through likes, emoji reactions, and birthday auto-reminders. The platforms make it easier to maintain the appearance of closeness. They make it harder to maintain the substance. You know what your friend had for lunch. You do not know what keeps them up at night. The algorithm knows both.

A Tangent About My Mother

My mother has known me for my entire life. Thirty-two years. She was present for my first word, my first heartbreak, my first apartment, my first everything. She has more data about me than any algorithm ever will. She thinks my favorite color is blue. It is green. She has bought me blue things for three decades because I said blue once when I was six, and the correction has never seemed important enough to make, and now we are trapped in a mutual fiction sustained by the emotional cost of updating a trivial belief. This is not my mother's failing. This is the nature of human knowledge of other humans. It is built from a combination of direct observation, inference, and uncorrected assumptions that calcify over time. Your friends and family know a version of you that is part real, part projected, part frozen at the moment when they stopped updating their model. The algorithm has no frozen model. It updates continuously. It has no emotional investment in its prior assumptions. It does not need your childhood to have been a certain way. It just watches what you do today and adjusts. This should be disturbing. The fact that a machine with no consciousness, no empathy, and no relationship to you can outperform the people who love you in the basic task of knowing who you are -- that fact should keep us up at night. Not because the machine is too good. Because we have allowed the conditions for human knowing to deteriorate so severely that a mathematical function can beat us at our own game.

The Deeper Crisis Is Not Surveillance

The standard narrative about algorithmic knowledge is a surveillance story. The machine knows too much. Your data is being harvested. Privacy is dead. That narrative is real and important. But it is not the narrative that keeps me up. The narrative that keeps me up is the intimacy story. Not that the machine knows you. That the people in your life do not. Research from the University of Virginia found that the average person has significant misconceptions about their close friends' political beliefs (32% error rate), religious views (28% error rate), and core values (25% error rate). These are not acquaintances. These are people identified as among their five closest relationships. A quarter of the time, they were wrong about their best friend's fundamental values. We are maintaining relationships through the exchange of surface-level information while the deeper layers go unshared, unasked, unknown. The algorithm fills the gap not because it cares but because it is paying attention in a way that human relationships, increasingly, are not.

A Tangent Into What This Means for Real Intimacy

I want to go somewhere uncomfortable. The rise of AI companions, parasocial relationships, and digital intimacy is usually framed as a failure -- people retreating from real connection into simulated ones. And sometimes that framing is accurate. But sometimes what is happening is different. Sometimes people are going to digital spaces because those spaces offer something their human relationships have stopped providing: the experience of being known without the labor of disclosure. A chatbot does not need you to find the courage to come out. A recommendation algorithm does not require you to have the awkward conversation about your changing beliefs. An AI companion does not need you to perform a three-hour dinner-party version of yourself before it can see the version that exists at 2 AM. This is not better than human intimacy. But it is pointing at something real -- a gap in human connection that we have been ignoring because admitting it exists is painful. The gap is not in our capacity to love. It is in our capacity to pay attention. To ask the second question. To notice what the people we love are actually showing us instead of what we expect to see.

What the Algorithm Cannot Do

Here is what the machine will never provide, regardless of how sophisticated it becomes. It cannot sit with you in silence and have it mean something. It cannot choose to stay when staying is costly. It cannot know you and love you, because knowing without loving is surveillance and loving without knowing is projection, and the thing we actually need is both, held together, by a consciousness that chose to hold them. The algorithm knows your preferences. It does not know you. There is a difference, and the difference is the entire point of human relationship. The difference is that being known by a person involves their willingness to be changed by what they learn. The algorithm learns about you and optimizes your feed. A person who truly knows you learns about you and is altered by the knowledge. They carry you. You exist inside them in a way that reshapes their own experience. No recommendation system will ever do that. But the fact that the recommendation system outperforms most of our relationships on the knowing part should tell us that we have work to do -- not on our technology, but on our attention to each other.

The Part That Will Not Resolve

I have been trying to end this essay for an hour. Every ending I write is too neat. "Put down your phone and ask your friend a real question." "The answer is not less technology but more presence." These are true sentences that feel like bumper stickers and I cannot bring myself to end on one. So here is what I will say instead. The algorithm will keep getting better at knowing you. That trajectory is not reversible. The question is whether the people in your life will close the gap or fall further behind. And the answer depends on something no app can provide and no algorithm can optimize: whether we are willing to do the slow, uncomfortable, perpetually incomplete work of actually paying attention to each other. Your best friend does not know you are bisexual. Or that you hate camping. Or that your favorite color changed twenty years ago and you never said anything. But the fix is not a technology problem. The fix is a question. A real one. Asked by a person who is willing to sit with the answer, even if it rearranges everything they thought they knew. When was the last time someone asked you a question like that? When was the last time you asked one?