Maya Chen: Nathan, I've been angry all week and I've been trying to figure out if it's — mm — the world, or my phone.
Dr. Nathan Hayes: Ha. Possibly not a distinction you can make cleanly.
Maya Chen: Which is sort of the whole episode, isn't it. Because today we're getting into algorithmic feeds — why they reward outrage, how deliberate that is, and what we actually know about whether it changes how people think. And I keep getting stuck on this one internal Meta decision, which is that an angry-emoji reaction was worth five times a like inside their ranking system.
Dr. Nathan Hayes: Five to one. And the mechanism behind that choice is straightforward once you understand engagement-based ranking — content gets surfaced according to predicted user interactions, clicks, comments, shares, watch time. Anger just happens to drive those signals at a disproportionately high rate.
Maya Chen: Wait, so — it's not that Meta invented outrage, it's that they measured it, valued it, and then built a system that treats it as the most useful data point?
Dr. Nathan Hayes: Correct. And the same logic applies across Meta's properties, YouTube, TikTok, X — anywhere you have engagement-based ranking, you have a structural preference for emotionally intense content. That's the design.
Maya Chen: And the popular story is that this is what's tearing us apart politically. The algorithm is the villain. But you've been — I can already see you, you're going to complicate that.
Dr. Nathan Hayes: So — what I'd say is the mechanism is real. The downstream effects on polarization? That's where the evidence gets genuinely murky. And that gap deserves its own careful look.
Maya Chen: Hmm. Yeah. That's the tension that made me want to do this one.
Dr. Nathan Hayes: The amplification mechanism and the attitude-change effect — they're not the same thing. Conflating them is where most of the popular narrative goes wrong.
Maya Chen: Okay — so let's start at the beginning. The five-to-one weight. What does that actually tell us about how this whole system was built?
Dr. Nathan Hayes: What it tells you is that someone sat down and explicitly assigned a numeric value to emotional intensity. That's not an accident of machine learning — that's an engineered preference. The ranking function was literally told: treat anger as five times more informative than a neutral tap on a like button.
Maya Chen: Which means the algorithm didn't — mm — discover outrage. It was handed it.
Dr. Nathan Hayes: Exactly. Now here's the plain version of how the whole system works. Imagine a vending machine that learns you buy chips every time you're stressed. So eventually it starts putting chips at eye level the moment you walk up looking tense. That's engagement-based ranking. It doesn't judge the chips. It just noticed they move.
Maya Chen: Oh. Yeah. That's — yeah, that lands.
Dr. Nathan Hayes: And the reason outrage specifically is the chips — not joy, not curiosity — that's where the neurobiology comes in. There's a well-documented cognitive tendency called negativity bias. Negative information gets processed faster, retained longer, attended to more reliably than positive information. The brain's threat-detection system is primed for it.
Maya Chen: Wait — so the algorithm didn't even have to invent the exploit. The exploit was already in us.
Dr. Nathan Hayes: Right. That's the more precise claim. The algorithm found a vulnerability in how our nervous systems actually work and — now, this is what I want people to hold — it mapped onto negativity bias without anyone necessarily intending that. The five-to-one weighting may have been intentional. The neurological exploit? Probably just... fell out of the optimization.
Maya Chen: Hmm. That's almost more unsettling, actually. That it didn't require a villain in a room deciding 'let's use their brains against them.'
Dr. Nathan Hayes: Correct. And it shows up in something very concrete. Picture someone opening Instagram at seven in the morning while their coffee brews. A post appears — something about a political issue that already bothers them. Not random. They interacted with similar content before, angrily, and the system flagged them as a high-interaction user for that content type. The feed learns. It serves.
Maya Chen: And they haven't even had their first sip yet.
Dr. Nathan Hayes: That's the engagement feedback loop — emotionally intense content performs, gets promoted, generates more signal, trains the algorithm to prefer more of it. It compounds.
Maya Chen: It's almost too elegant. Like — the system found a cheat code in human psychology and just... kept pressing it.
Dr. Nathan Hayes: That's actually the right frame. Though I'd push back on 'cheat code' slightly — only because it implies a workaround. Negativity bias isn't a bug in human cognition. It evolved for good reasons. What the algorithm did was take something adaptive and monetize the excess.
Maya Chen: But — wait, I want to stay in the loop for a second. Because 'it compounds' is doing a lot of work there. What does that actually look like, mechanically? Like, each successful outrage post isn't just rewarded once, right?
Dr. Nathan Hayes: No. No, that's exactly the part people miss. Every interaction on a post that performed — the angry comment, the share — that becomes training data. The algorithm recalibrates. Next round, it doesn't just surface similar content. It surfaces content that's slightly more intense, because that's what the data said moved people.
Maya Chen: It's... escalating each time.
Dr. Nathan Hayes: Ratcheting, yes. And Northwestern and University of Chicago researchers documented this specifically — engagement-based algorithms don't just amplify emotional content generically, they preferentially amplify intergroup, moralized, toxic political content. That's the category that keeps winning the optimization.
Maya Chen: Mm. So the system isn't just learning 'outrage works.' It's learning 'outrage about the other group works best.'
Dr. Nathan Hayes: Correct. Which matters for — right, for where Frances Haugen fits in. Because she's not describing an accident.
Maya Chen: Yeah — this is the part that I find hardest to sit with, actually. Haugen testified to Congress in 2021. She'd been a product manager inside Facebook. And she said — and I mean, she had the internal research to back this — that Facebook knew its algorithms amplified outrage and that leadership kept choosing profit over doing anything about it.
Dr. Nathan Hayes: And importantly — she wasn't speculating. She took the internal documents. That's the Facebook Files. The company had the research. They had it.
Maya Chen: And then in March 2026, there's this BBC investigation — more than a dozen whistleblowers from Meta and TikTok — and they describe it as an 'algorithm arms race.' Platforms started loosening what they'd allow, specifically because TikTok's growth was terrifying them. Internal research showed outrage fueled engagement. They leaned in.
Dr. Nathan Hayes: That framing — arms race — is actually precise. It names the competitive mechanism. The harmful content tolerance wasn't ideological, it was a response to market pressure.
Maya Chen: Which is somehow worse? Because you can't even argue they believed in it. They just — they just needed the numbers.
Dr. Nathan Hayes: Now — and this is the distinction I want to hold — Haugen proves intent. She proves the mechanism was known and tolerated. What she doesn't settle is the magnitude of the downstream harm. Those are separate evidentiary questions.
Maya Chen: Okay but — if they knew, and they chose profit anyway, and then more whistleblowers are saying the same thing five years later... does the question of magnitude even change who's responsible for fixing it?
Dr. Nathan Hayes: It changes who's responsible, yes. But — and I want to be precise here — responsibility and causal magnitude are separate questions. And the magnitude question is where I'd push. Because we actually have an experiment. A real one.
Maya Chen: The chronological feed study.
Dr. Nathan Hayes: 2020. US election. Researchers took consenting Facebook and Instagram users and switched them off the default algorithm entirely — reverse-chronological feeds. You see posts in time order. No ranking. No amplification. Time on platform dropped substantially. Exposure to uncivil content on Facebook shifted. And polarization — measured issue polarization — did not significantly change.
Maya Chen: Wait. They turned the algorithm off and it... didn't move the needle on how people actually think?
Dr. Nathan Hayes: Correct. And this is what I'd call the exposure-versus-attitude gap. The algorithm demonstrably changes what you see. What it does not straightforwardly change is what you believe. Those are — actually no, let me put it this way — we've been assuming those two things move together. The experiment says they don't, at least not cleanly.
Maya Chen: Hmm. That's — that's strange. Because the whole narrative rests on them moving together, right? Algorithm feeds you outrage, outrage shifts your worldview.
Dr. Nathan Hayes: Which is intuitive. But the data says exposure changed and attitudes didn't follow. So either the exposure was never doing the work we assumed, or the people whose attitudes could still move — they're not the ones the algorithm is primarily reaching.
Maya Chen: Oh — wait, is that where the fringe piece comes in?
Dr. Nathan Hayes: Yes. And this is the part that surprised me when I first read it. A Nature review found that average user exposure to misinformation and extreme content is actually low — and heavily concentrated in a narrow, highly motivated fringe. Not the typical user. The person already seeking that content.
Maya Chen: So the algorithm isn't radicalizing the masses. It's... amplifying people who were already headed somewhere extreme.
Dr. Nathan Hayes: That's where the evidence points. And the Science Advances review on YouTube found the same thing — the radicalization rabbit hole hypothesis, successive recommendations pulling ordinary users toward extreme content, the effect sizes there are modest where they've actually been studied. The fears were overstated. And I want to be careful — YouTube hosts extremist content, profits from it. That's real. But the pathway of ordinary user gets recommended toward radicalization? Empirically contested.
Maya Chen: So the villain might be — mm — less the algorithm turning people, and more the algorithm finding people who are already turned and giving them a megaphone.
Dr. Nathan Hayes: That's the uncomfortable recalibration. The mechanism is real. The harm at the edges is real. But if the algorithm is primarily amplifying existing fractures rather than creating new ones — that's a harder problem. Because you can't just fix the ranking function and expect the fractures to heal.
Maya Chen: I keep tripping on that word — amplifying. Because it changes where you point. If the algorithm is amplifying fractures that already exist... then fixing the algorithm is sort of like, I don't know, turning down the volume on something that's still playing.
Dr. Nathan Hayes: And blaming the speakers for the song.
Maya Chen: Yeah. Which — mm — I find genuinely uncomfortable. Because it's easier to point at Meta, at the five-to-one weighting, at Frances Haugen standing in front of Congress with documents. That's a system. You can regulate a system. But if what's underneath is... cultural fractures that predate any platform, that's just — that's a much lonelier problem.
Dr. Nathan Hayes: Now, I want to be precise about one thing — the harm is still real. The BBC investigation, the whistleblowers from Meta and TikTok describing an arms race, loosening content tolerance to chase TikTok's numbers. That happened. The margins matter. I'm not saying nothing to fix.
Maya Chen: No, I hear that. I think what I'm sitting with is — we went into this assuming the chronological feed experiment would confirm the story. Switch off the algorithm, polarization drops. And it didn't. Exposure shifted, time on platform dropped, and people's actual views on issues? Unchanged.
Dr. Nathan Hayes: The exposure-versus-attitude gap. Yes. That's the result that should give everyone pause.
Maya Chen: Maya Chen: Seven in the morning. Coffee brewing, Instagram open. We started there. And now I'm thinking — the algorithm served them something that already lived in them. The anger was already there. The feed just... knew where to knock.
Dr. Nathan Hayes: Five times a like. Still think that number is damning. I just — I'm less certain now what it's damning of.
