Gotcha Culture: Why Catching an AI Cheater Feels So Good
Welcome back to PsyberSpace I'm your host, Dr. Leslie Poston. This week, we're talking about why some people seem absolutely gleeful if they think they've caught someone using AI and what that public display of glee or one upmanship is actually doing for them psychologically. By the end of this episode, you should be able to catch yourself doing it before you hit post.
Leslie Poston:Watch what happens the next time an AI detector flags someone's writing. The reaction almost never looks like concern for a person facing a consequence they don't deserve. It looks like satisfaction. Screen straps circulate. Confident LinkedIn posts get written.
Leslie Poston:Someone always adds some version of I knew it to the comments as if the accusation confirmed a suspicion they'd been patiently carrying, waiting for proof to catch up to their instincts. That reaction is Schadenfreude, pleasure taken in someone else's misfortune. And it has two judgments happening almost instantly, how responsible the person seems for their own downfall and how deserved that downfall feels as a result. And the part many people don't examine is that judgment of deservingness isn't stable. It bends depending on who's being accused.
Leslie Poston:Research on moral hypocrisy has suggested people hold outsiders to a stricter standard than they hold their own side, forgiving an identical act more easily when someone they already like commits it. The pattern holds up pretty consistently. Real favoritism towards your own group and real harshness toward anyone outside it, even when the groups are arbitrary. This means the same AI use can read as reasonable help or as unforgivable cheating depending on almost nothing about the act itself and on everything about whether you already had a reason to root for or against the person doing it. And one detail that might actually unsettle you a bit.
Leslie Poston:Detection tools have flagged students who wrote every word of their own papers, sometimes at rates approaching one in five within a single graduating class. The person doing the accusing feels the exact same rush of deserved comeuppance, whether the accusation turns out to be true or false. This tells us the pleasure is not about tracking guilt. It's about tracking the chance to feel morally superior, and the accusation only is supplying the occasion. And here's a sign to watch for in yourself.
Leslie Poston:If the satisfaction arrives before you've actually confirmed the truth of the accusation, that's a tell. Certainty is outrunning evidence. And when certainty gets there first, it's usually not about the evidence at all. Social media makes this easy to keep doing because it rewards it. A callout post that turns out right gets the same burst of engagement as one that turns out wrong, at least in the algorithmic window before anyone corrects the record.
Leslie Poston:That's a variable reinforcement schedule, the exact intermittent reward structure that keeps people pulling a slot machine lever. The behavior only needs to pay off often enough to keep going, not every time. In fact, social media tools are constantly adding features that leverage this for people to use against each other. Just today, LinkedIn is beta testing, a feature that allows people to report a post for seeming like AI. There's no recourse for the person being reported to combat the accusation.
Leslie Poston:This is going to be weaponized against the most vulnerable on the Internet, and it's because it taps into all of the psychological mechanisms that we're talking about today. If these accusations are rarely about the person being accused, then it's honest to question what they're actually doing for the accuser. And there's not one answer. There's actually at least five, and each of them leaves a different recognizable fingerprint. The first is protecting a self image you know isn't fully earned.
Leslie Poston:Downward comparison is what we call this. When people face a threat to self esteem that they can't directly fix, they start to cope by finding someone worse off instead. And this shows up pretty strongly in people already uncertain of their own standing. A tool that produces competent, unremarkable output on demand is a real threat to anyone whose sense of value depends on being distinctly good at something. Catching someone else cheating does some real work against that psychological threat.
Leslie Poston:It reframes an entire field as rigged rather than skill based, And a rigged field means an ordinary result stops being someone's personal failing. The sign here is a feeling of relief rather than resolve. If the accusation makes you feel like the whole system is unfair rather than making you want to actually improve at anything, that relief is the tell in this case. The second is impostor syndrome. The impostor phenomenon describes people who are genuinely good at what they do but can't internalize that, attributing their own success to luck rather than skill, and carrying a chronic fear of being exposed as a fraud despite real evidence against that fear.
Leslie Poston:For this one, AI isn't threatening because it's competitive. It's threatening because the one piece of evidence that someone's relied on to hold the fraud feelings at bay, having actually done the work themselves, stops meaning anything once a machine can fake the same result. Catching someone else's shortcut becomes proof that doing the work still counts. The sign here is disproportion. If your reaction to someone else's shortcut is really about defending your own legitimacy rather than about what they did, that sense of disproportion is the tell in this case.
Leslie Poston:The third is sharper and has higher stakes than the first two. Research on AI disclosure and trust has found that people who use AI secretly are often the harshest judges of others who admit to its use. Bandura's work on advantageous comparison explains this logic. People make their own conduct feel more acceptable by holding it against someone else's version framed as worse. Publicly shaming a more obvious user draws a convenient line with their own use safely on the acceptable side, and it builds a reputation that makes their own use less less likely to draw suspicion.
Leslie Poston:The real psychological cost here sits kind of on the other side of this one. Research on why people hate hypocrites more than the underlying offense would predict tells us why. Condemning a behavior publicly signals your own virtue, and getting caught doing the thing you condemned doesn't just expose one act. It exposes the signal as false, which people punish harder than the original behavior alone. I covered the study behind the disclosure finding in more depth in an episode called the AI transparency trap.
Leslie Poston:And the hypocrisy piece connects directly to what we dug into in an episode called can we uncover the psychology of a carrot if you want to find out more. The sign here is probably the loudest one on our list. If you feel a compulsion to publicly declare where you stand before anyone's even asked, that eagerness to go on record is worth taking notice of before you post your callout post. The fourth is a performance layered on top of any of these. Moral grandstanding is public condemnation that functions less as an actual moral stance and more as self promotion, marked by outrage calibrated to signal insight rather than to match the offense.
Leslie Poston:A callout post is rarely addressed to the person being called out. Technically, it's addressed to everyone else, and it lets the person posting it claim two things at the same time, moral superiority and a discernment the crowd supposedly doesn't have. The signal here is where your attention goes after. If you're checking the replies more than you're checking in to see whether the person you called out was telling the truth, the performance has taken over from the concern. And the fifth attaches to any of the other four rather than standing apart.
Leslie Poston:AI companies train their models on other people's creative and intellectual work without asking permission or offering payment. That's a real grievance. But frustration aimed at a target too large to confront doesn't disappear. It's just relocating. A workable substitute target has to be manageable, safer to confront than the real source, and easy to identify.
Leslie Poston:And the training pipeline run by a multinational company fails every one of those tests. A classmate who used a chatbot and might have lied about it passes all three. The anger at the company doesn't vanish when it's misdirected and lands on the classmate instead. It's just borrowing moral weight. And the sign is the mismatch.
Leslie Poston:If the intensity of your reaction to one person is bigger than what that person could possibly be responsible for, you're probably not actually angry at them. You're probably angry at the system. Now none of these five would find this much oxygen without a tool that's willing to supply false certainty on demand. So let's figure out exactly why that tool fails. Most AI detectors score writing on how predictable word choices are and how much sentence rhythm varies, sometimes called perplexity and burstiness.
Leslie Poston:If you've talked to me offline, you know that I call AI smart mad libs that like to hallucinate. AI generated texts tends to score low on both perplexity and burstiness. I'm sure plenty of human writing scores the same way for reasons that have nothing to do with AI, but some that do. I did train on all of our work. Non native English speakers often produce more formulaic patterns from how the language was taught to them.
Leslie Poston:Some neurodivergent writers produce more uniform sentence structures for reasons rooted in how they process language. Demographic data confirms this isn't a marginal error. Non native English speakers and black students, for example, get falsely flagged at roughly double the rate of everyone else by these AI detection systems. Heck, a student at the University at Buffalo was accused of academic dishonesty after a detector falsely flagged a paper she wrote entirely herself. And roughly 20% of her graduating class was flagged right along with her.
Leslie Poston:The output format makes it worse. An AI detector doesn't report a confidence range. It reports a single number, say 98% AI generated. And that kind of false precision reads as rigor whether or not the measurement underneath it can support the claim. Whichever of the five signs is actually driving someone's accusation, that single number gives them something concrete to point to instead of a feeling that they'd have to justify out loud.
Leslie Poston:The sign to watch for here is simple. If a single number is doing all of the convincing and nobody has asked or investigated what's actually behind the number, that's manufactured certainty, not evidence. Of course, none of this just started with AI. Earlier versions of the same psychological instincts make the shape of it all easier to see. For example, witch finders across early modern England and Scotland used to use tests like pricking the skin for a devil's mark to publicly identify frauds against god and community, as they said.
Leslie Poston:The tests measured nothing real, but the finder gained genuine standing from every exposure, sometimes payment, and the community got a sense of restored order regardless of whether anything true had been uncovered. In more recent news, the twenty nineteen Operation Varsity Blues scandal followed a similar trajectory with a structural target instead of a supernatural one. Wealthy parents, including actresses Lori Lohland and Felicity Huffman, paid a consultant to fabricate athletic records and bribe test proctors into elite universities. The fraud was real and prosecutable, but the size of the public reaction had less to do with the fraud itself than with what the fraud couldn't fix. Elite admissions already run on legal advantages available almost exclusively to wealthy families, legacy preference, donor buildings, tutoring that costs tens of thousands of dollars.
Leslie Poston:None of that is illegal, and none of that went away, regardless of what happened to Lowland and Huffman. What was available to the public instead was a small number of identifiable, chargeable, camera ready people, and the public treated punishing them as if it could settle a score that was actually something they had against the entire system. And not every instance of exposure runs on this pattern. Researchers who've spent years examining published data for signs of fabrication have corrected the scientific record in ways that genuinely matter. Their underlying drive isn't that different from anything we've talked about in this episode.
Leslie Poston:What is different is their process, verification before any public claim, a willingness to be wrong, and accountability that holds up without needing the audience's applause to feel like they're finished with their work. So what can you do to avoid schadenfreude, to avoid virtue signaling by picking apart sentence structure and trying to decide that totally normal punctuation use suddenly makes something AI because you might not have been familiar with it before or because it seems popular now. Well, before you share a callout post or an accusation, check the actual reliability of whatever produced the doubt. Treat any AI detector percentages as meaningless since these tools fail at rates high enough to falsely implicate a fifth of an entire graduating class, for example. A number attached to an accusation isn't proof just because it's a number.
Leslie Poston:One recent example that was proof was the professor that gave his students a take home midterm and then an in class final. I'll try and find that story and link it for you in the show notes. It's a way for you to see that it is possible to catch AI deception, but you just can't rely on an AI detector to do it. Another technique, run your own five sign check before you post anything. Are you feeling relief instead of resolve?
Leslie Poston:Is this potentially really about your own legitimacy instead of theirs? Do you feel a need to go on record before anyone's asked a question? Are you watching the replies to your post more than you're watching whether your post is actually correct? Is the size of your emotional reaction bigger than the size of what this one person could possibly control? Any one of those showing up doesn't necessarily mean you're wrong.
Leslie Poston:It means the accusation is doing something for you that has nothing to do with the person on the other end of it. If you're in a position to set policy at a school, a newsroom, a workplace, etcetera, push for a rule that no accusation gets acted on from a detector score alone. Require a second and third form of evidence. If you can, without revealing their sources, a draft history, perhaps an editing timeline, notes, or an actual conversation before any consequences land, that single procedural change would have protected the Buffalo student we mentioned earlier and everyone flagged next to her. And if you've used AI and have been quietly anxious about what that means about you, that worry deserves to be addressed directly in whatever way actually works for you, not managed indirectly by making sure other people get caught first.
Leslie Poston:Thanks for listening to PsyberSpace I'm your host, Leslie Poston, signing off. As always, until next time, stay curious, and don't forget to subscribe so you never miss an episode.
Creators and Guests
