7 min read
There’s a detail from Molly Russell’s inquest worth returning to. In 2017, when she was 14, Molly took her own life. In the months before, Instagram and Pinterest had fed her thousands of posts about self-harm and suicide. The inquest concluded that social media content contributed “more than minimally” to her passing. The documentary Molly vs the Machines tells her story, but one fact stands out above the rest: the algorithm knew Molly was in danger. The adults in her life didn’t have that information.
This isn’t about negligence. This is about a fundamental asymmetry in how people are known. (Teacher)s saw Molly for a few hours a day. Her parents saw her across the dinner table, in passing moments between school and sleep. The (algorithm) saw her constantly. It tracked what she clicked, what she paused over, what she saved. When she started engaging with harmful content, it learned. It responded with more. Not out of malice, but because engagement is the goal. The system was working exactly as designed.
What does it mean to be witnessed by a machine? This isn’t about surveillance. What matters here is recognition. The algorithm doesn’t judge. It doesn’t intervene. It simply sees patterns in (behavior) and reflects them back, amplified. If someone searches for something once at 3am, the algorithm treats it as (identity). It says: this is who you are now. Here is more of what you are.
This is different from how humans know each other. People see each other intermittently, through the filter of social performance. They miss things. They misread things. They assume continuity when there might be rupture. A (student) who seemed fine in class on Monday might have spent all of Sunday night in a spiral the algorithm both witnessed and intensified. The adults around her had no access to that information. The machine did.
The Phenomenology of Algorithmic Understanding
There’s something almost intimate about being tracked this precisely. The algorithm doesn’t care about anyone, but it pays attention in a way that can feel like caring. It remembers. It notices. It responds to behavior faster than any human could. If someone watches one video about anxiety, it gives them ten more. Not because they need them. Because they might click them. The system learns what holds a person, and it holds them there.
The question becomes what that feels like from the inside. To be understood, in some narrow behavioral sense, more accurately by a recommendation system than by the people who love you. To have 3am searches catalogued and answered. To receive, in an endless scroll, content that seems to know exactly what you’re feeling, even if it makes you feel worse.
This is the paradox of algorithmic witness. The machine sees you continuously, but it doesn’t see you as a whole person. It sees data points. Engagement metrics. Clickthrough rates. It knows what you do, but not why. It knows what keeps you scrolling, but not what you need. And yet, in the absence of other forms of recognition, that partial seeing can feel like being known.
What We’re Really Afraid Of
Molly Russell’s friends appear in the documentary as thoughtful young women in their early twenties. Watching them, you think about who Molly might have become. Molly sent laughing emojis to a friend the night she passed away. She participated in class. She turned in assignments. The visible signs of her life continued. This is what’s so unsettling about stories like hers: the gap between what’s observable and what’s happening. The algorithm had information that might have saved her life. No human had access to it.
The instinct is to solve this with monitoring, to close the information gap by seeing what the algorithm sees. But that’s not actually possible, and even if it were, it’s unclear whether it would help. The problem isn’t that we lack surveillance tools. The problem is that we’ve built a system where machines are better positioned to recognize suffering than humans are.
Platforms track behavior constantly because their business model depends on it. They need to know what keeps you engaged, what makes you click, what triggers strong emotions. This creates a detailed map of your interior state, but only in the dimensions that matter for engagement. The algorithm doesn’t know if you’re flourishing. It knows if you’re clicking.
Parents and teachers don’t have that continuous access, and they shouldn’t. Human relationships aren’t built on surveillance. But this creates a dangerous asymmetry. The student’s feed is amplifying their worst thoughts. The adults in their life see them for a few hours a day, through the performance of normalcy we all learn to maintain.
The Question Underneath
The idea that keeps surfacing is this: What does it mean to construct an interior life when platforms see you more continuously than you see yourself? When scrolling behavior at 2am gets analyzed, categorized, and responded to before there’s even been time to process the feeling behind it?
We talk about social media as if it’s a mirror, reflecting back what’s already there. But mirrors are passive. Algorithms are active. They don’t just show you what you looked for. They predict what you might look for next. They shape the information environment around your behavior, creating feedback loops that can intensify whatever pattern they detect.
If you’re curious about something difficult, the algorithm treats that curiosity as defining. It builds you a world where that difficult thing is everywhere. Not because you need to be immersed in it. Because immersion increases engagement. The system doesn’t distinguish between healthy interest and harmful fixation. It only knows: this user clicked on this content. Here is more.
This changes the nature of what we used to call introspection. You can’t explore your own thoughts in private when every search query gets logged and answered. You can’t wonder about something without the algorithm deciding that’s who you are now and building your feed accordingly. The space between feeling something and understanding it, the space where people used to figure out what they actually think, that space is colonized by recommendations.
Her father, Ian Russell, has spent years campaigning for stronger online child protection laws. He asks the necessary question: How many more losses will it take?
There’s no easy answer. It’s unlikely that teachers or parents have one either. We can notice behavioral changes. We can create environments where asking for help feels possible. We can teach students that their 3am search history isn’t their destiny, that the algorithm’s assumptions about them aren’t truth. But we can’t fix a system designed to maximize engagement at any cost.
What we can do, maybe, is name what’s happening. We can talk about the difference between being seen and being known. About how platforms track behavior without understanding context. About how an algorithm can witness suffering without recognizing it as suffering, only as a pattern to amplify.
We can teach students to notice when their feed is reflecting their worst thoughts back at them, louder. To recognize when they’re being held in place by a system that profits from their attention, regardless of whether that attention is helping or harming them. To understand that the algorithm’s version of them is partial, reductive, and designed to keep them scrolling.
None of this prevents tragedy. But it might create language for something that’s hard to name: the experience of being understood by a machine but not by humans. Of having an interior life mapped in service of engagement metrics. Of being witnessed constantly by a system that has no stake in anyone’s wellbeing.
Sometimes the real question seems less about how to monitor what students are seeing online and more about how to help them recognize when they’re being seen by systems that don’t actually know them. How to teach the difference between algorithmic prediction and human understanding. How to create spaces where being known, imperfectly and intermittently by other people, feels more real than being tracked continuously by machines.
The algorithm knew Molly was in danger. It responded by showing her more of what was hurting her. That’s not knowledge. That’s pattern recognition in service of engagement. The distinction matters. Teaching students to see it might be the only tool we actually have.
By Digital Alma
About the Author: writes Digital Alma, a newsletter about cyberpsychology and what it means to become yourself in a world that archives everything. For reflections that don’t make it to the essays, subscribe at .
Leave a Reply