Six Decades of Pop, Measured by Machine
Open your phone and scroll to the top songs of the week. Read a handful of choruses. Now go back and read lyrics from 1965. The difference isn't just synthesizers versus drum machines. The moral temperature dropped.
That's the headline claim of a large study out of the Centre for Digital Music at Queen Mary University of London, and it's one of the clearest illustrations I've seen of the AI impact on human psychology told through culture rather than through a chatbot window. The researchers — Vjosa Preniqi, Andreas Kaltenbrunner, Kyriaki Kalimeri and Charalampos Saitis, publishing in Scientific Reports — didn't listen to songs. They fed 380,000 popular tracks released between 1960 and 2023 into artificial intelligence and computational language models, then watched what the words were doing over sixty years.
The verdict lands like a slow door slamming shut. Expressions tied to care, decency and social connection steadily fell. Language tied to harm, cheating, subversion and degradation climbed, more or less in a straight line. Joy in the lyrics waned. Anger and disgust moved in. The authors call popular song a "cultural barometer," and right now the needle is pointing somewhere cold.
What the Study Actually Measured
The method matters more than the soundbite, so let's be precise. The team didn't scrape every song ever written. They leaned on two established corpora, the WASABI dataset and Billboard chart data, which is itself a revealing choice — this is a study of what sold, what got radio play, what a generation actually absorbed. A 2024 GPT-4 model served as their semantic baseline, tagging and categorizing vocabulary and short phrases according to moral foundations: care versus harm, loyalty versus subversion, and so on.
Why bother with a model at all? Because nobody can listen to 380,000 songs. That's the quiet point here. Large-scale AI text analysis uncovers macro-trends that are flatly invisible to a human reading lyrics by hand. You cannot feel a sixty-year slope toward disgust by ear. You can only see it at scale.
And that scale is exactly where the AI impact on human psychology becomes legible. The model isn't diagnosing anyone. It's a measuring instrument pointed at the collective output of decades of songwriters, and what it records is a drift in the moral vocabulary a culture hands its young people.
Six Decades, One Direction: Virtues Down, Vices Up
The decline of the virtues is the part that sticks with me. Markers tied to care, loyalty and decency — the linguistic glue of social cohesion — waned across the period. Not vanished. Waned. Meanwhile the vice vocabulary expanded, and the study describes the rise of themes like harm, cheating and degradation as a sharp, fairly linear escalation rather than a blip.
I want to resist the easy moral panic. A rising trend in a lyric corpus is not evidence that today's teenagers are worse people. But it is evidence that the stories popular culture rewards changed. Decency used to be a safe bet in a chorus. Now subversion sells. That's a real shift in what a society finds worth singing along to, and it shows up regardless of who's holding the microphone.
The Emotional Darkening Nobody Planned
Alongside the moral shift sits an emotional one. The researchers document a macro-level darkening: joy in popular lyrics declined while anger and disgust rose. The two curves — moral and emotional — track each other, which is the finding I find most unsettling. Values and feelings aren't separate ledgers. As the language turns cynical, the mood follows.
This is where I apply a mild correction to my own first reading. The darkening doesn't only mean "people got meaner." Rising disgust in lyrics could equally reflect artists finally saying out loud what older generations swept under the rug — economic stress, mental-health strain, social fragmentation. Older broadcasting standards kept a lot of pain off the air. That isn't the same as pain not existing.
The AI Impact on Human Psychology, in Plain Terms
So what is AI in psychology, actually, when it looks like this? Strip away the lab coat and it's the discipline of using computational models to read patterns in human behavior and expression at a scale no person could manage. Not therapy robots. Not sentient machines. A tool that quantifies what people write, say, click and sing, then looks for the shape underneath.
That's the version of psychology where AI genuinely shines: the pattern-recognition layer. It can read the mood of half a million songs and tell you the slope. It can surface that a generation's soundtrack got bleaker before any cultural critic could put the feeling into a paragraph.
If you've spent any time thinking about why your brain treats old songs differently, this is the same machinery running in reverse. Music is an emotional instrument, and this study is essentially a sixty-year electrocardiogram of the culture it lives in.
Can AI Understand Human Psychology? Not Quite
Here is the thing everyone over-explains, so let me under-explain it honestly: no. The model in this study understands that a certain word clusters with the vice "degradation." It does not understand degradation. It has no felt sense of betrayal, no memory of a song that wrecked it at seventeen. It reads the trace of psychology, not the thing itself.
The study's authors are upfront about this, which is why I trust them. They flag that the system can't always parse metaphor, sarcasm or regional context, and they note a corpus bias from drawing mostly on chart data. They also refuse, in the strongest terms available to them, to claim causation. It is explicitly a correlation — and the authors describe culture and music as a feedback loop. Songs reflect the world and shape the next generation of listeners at the same time. You cannot tell, from a corpus, which one is doing the pushing.
That gap is the honest answer to can AI understand human psychology. It can model the fingerprints of human psychology with growing precision. It cannot, on this evidence, read the mind that left them. The same limitation governs every serious application, including the algorithmic matchmaking reshaping how people find love: the system reads behavior so fluently that we keep mistaking the reading for understanding.
A Mirror, Not a Verdict
What this study really delivers is a mirror held up by a machine. A culture that spent six decades slowly swapping care for cynicism in its most intimate art form learned it had done so — and learned it only because a model could hold 380,000 songs in its head long enough to notice.
I'd take the finding seriously and the doom lightly. The lyrics got darker; whether that's a diagnosis or a confession is still an open question, and that ambiguity is, I think, the most human part of the whole result.
