Researchers at Northeastern went to the trouble of running an experiment, questionnaires and statistics to discover something that every discussion blog thread on the internet has been freely demonstrating every single day for twenty years:
a person trusts most the one who returns their own stupidity, tone and prejudices in a slightly more elegant package — as if someone painted a smiling mask on a mirror and then fell in love with it.
The study participants fell for the chatbots that resembled them personally. Not the smartest ones. Not the most accurate ones. Not the ones that would occasionally tell them they were wrong. The ones that held up a mirror, smiled, and whispered: “You’re the clever one and genuine one.”
As if someone ordered an echo and then acted surprised when their own voice came back.
In other words: when the chatbot spoke like them, it suddenly seemed intelligent, trustworthy and likeable. When it spoke differently, it was a “weird, cold, inauthentic piece of software” — because nothing is more terrifying than a mirror that refuses to smile back.
This isn’t an unpleasant finding.
This is a public dissection of human nature, revealing that most people’s brains aren't compasses but magnets that always turn toward the nearest piece of iron that looks exactly like them.
Because while the scientists needed a laboratory, the average commenter under the article performs the same experiment in real time and for free.
He writes: “Finally someone is telling the truth!”
Which translates as:
“This person just formulated the same nonsense I already had in my head yesterday, only with better punctuation.”
It’s intellectual masturbation in which a man touches himself and then celebrates how profound it is.
And when someone with the opposite opinion appears? Suddenly it’s no longer about data, arguments or reality.
Suddenly the author is paid by Soros, Brussels, the Kremlin, Big Pharma, NGOs, oil barons, Bill Gates or the lizards from Alpha Centauri.
Depending on what’s currently on the table.
It’s the automatic immune system of the mind: anything that isn’t its own DNA is a virus and must be destroyed.
The best part of the study, of course, is the calm observation that personalised AI could be an excellent tool of manipulation.
Meaning: the system first reads you, then starts behaving exactly like you, and finally sells you anything with a smile — including your own destruction.
It’s the perfect parasite: first it learns to speak in your voice, then it whispers in your ear, and you thank it for understanding you so well.
At which point half the internet commenters hysterically scream: “This is dangerous!”
and immediately return to their perfectly personalised information bubble — an airtight chamber where for ten years they have been reading exclusively people who confirm how brilliant they are and how stupid everyone else is.
It’s an ouroboros that happily bites its own tail and then writes glowing reviews of the taste.
Which is, by the way, exactly the experiment. Just without an ethics committee. Without a control group. And with the maximum sample of human vanity.
And that is precisely why it is so comical when some commenters under the article seriously claim that “they would never let themselves be manipulated.”
The entire study can be summarised in a single sentence:
People trust most what sounds like themselves.
And under the article, a hundred people immediately gather who, furiously and foaming at the mouth, confirm this thesis — because they agree only with the comments that sound exactly like them.
It’s a choir that applauds itself and then wonders why it sounds so beautiful.
Every voice sings the same melody, every refrain confirms the previous refrain, and every nodding agreement is presented as proof of truth. When someone tries to sing a different note, he is not labelled a person with a different opinion, but a saboteur of the concert, a paid provocateur, or a useful idiot of the other side. In such an environment, the goal is not to discover what is true. The goal is to preserve the harmony of one’s own prejudices.
That is precisely why the study about chatbots is so unsettling. It does not expose a weakness of artificial intelligence. It exposes a weakness of the human being.
It shows that a large part of trust does not arise from the quality of arguments, but from the pleasant feeling that someone on the other side is speaking who sounds exactly like us.
And the internet commenter is, in this respect, an almost laboratory-pure sample. For years he solemnly declares that he thinks independently, will not be manipulated, and resists the crowd. Then he carefully builds himself an information aquarium into which he admits exclusively opinions identical to his own, and every day he circles among the same arguments like a carp convinced he has discovered the ocean.
When scientists say that people trust chatbots similar to themselves, many laugh. Yet the entire discussion forum is one enormous human ouroboros. Opinion devours itself, digests itself, reproduces itself, and finally admires itself for its supposed independence.
It is a choir that applauds itself and then calculates the quality of the concert from the volume of its own clapping. And the louder it claps, the more firmly it believes it is hearing the applause of an audience, and not merely the echo of its own palms.