Especially on that cesspit we call Twitter, people hold pseudo-scientific debates with each other, and when they need backup, they call in bot accounts and AI. “Hey Grok, is this true? Answer properly, no beating around the bush.”
The way AI gets treated like an all-knowing god is funny. Because contrary to popular belief, AI doesn’t know everything. During training it learns patterns from text; when it searches the internet, it draws on whatever sources come up. But comparison or verification rarely enters the picture.
If a “study” claiming “42% of Turkish DNA is microplastic” sits at the top of the results, AI tends to pick it. AI doesn’t verify every source. It can’t, either, because it’s a system limited by the patterns it learned and the sources it uses. And visibility in search results doesn’t prove a claim is true. SEO, popularity, bot accounts and false information republished over and over shape which claims become visible.
Asking AI about contested topics, especially history, is both funny and dangerous. Funny because people pick and choose their information. If AI says something that doesn’t support their view, it gets cursed at; if it says what they want, it gets praised. The “just give me a straight answer” line says it all.
Dangerous because it’s not just history. People even ask AI for health advice. And as you know, the internet is crawling with “alternative” medicine types. Those who legally can’t call themselves doctors use unscientific titles like “healer” or “wellness practitioner” and write prescriptions anyway.
“Is this true?” has become a reflex, especially on Twitter. Fast internet searches may look convenient, but given the depth of the questions being asked and the fact that AI still makes mistakes, it’s actually doing us harm.
I’ve even seen AI mess up a recipe. The recipe it pulled from another site had extra stuff in it, my flour ran out, and I ended up with runny dough. Luckily, it firmed up a bit in the oven.
Historiography
History is not a list of facts. The date of a war may be certain; why it started, who was right and how many died are a clash of sources, archives and interpretations.
The historian’s job is to manage that clash. Who wrote the document, for whom, when, and what would they have wanted to hide? If a chronicler’s number contradicts an archival record, which one do you trust, and why?
AI doesn’t ask these questions on its own. It produces the statistically most likely text. In other words, it doesn’t resolve the clash; it takes the internet’s average of it and hands it back as a single-voiced narrative.
Isn’t the “internet average” true? The majority believing something doesn’t make it true. Besides, this is the internet age. There are bots everywhere and more disinformation than ever.
An anonymous post racking up tens of thousands of likes out of nowhere looks suspicious to me now. And thanks to Twitter, the free-speech platform, we can’t even see who liked it.
“asd123 liked this!”
What’s inside the machine
The machine declared a god has three invisible authors. The first is training data: what language the internet is written in, by whom, and how much. Behind the answer to a history question asked in Turkish, there’s likely the perspective of English-language sources.
The second is post-training tuning: what the model will and won’t say, and where it treads carefully, is shaped by people.
The third is system instructions: directives the company gives the model that the user never sees.
In February 2024, a diversity setting made Google’s Gemini depict historical figures and eras in ahistorical ways; the company had to pause image generation. In 2025, Grok first started slipping the “white genocide” claim about South Africa into unrelated answers, and xAI blamed an unauthorized change to its instructions. A few months later, after another update, it called itself “MechaHitler.”
The lesson here isn’t “bad model.” Neutral, all-knowing gods are products whose direction can change with a few lines of instructions.
You can now see the sources behind their answers. But that doesn’t mean everything’s fixed. You still have to check whether the sources are accurate, even whether they exist, because it may have made them up. Yes, AI can even cite books that were never written. Cases in the US where lawyers submitted AI-fabricated case law and were sanctioned are documented examples (Mata v. Avianca).
Not sure if they’ve updated it, but there was a time when you asked AI for a number between 1 and 10 and it said 7. The human tendency to pick 7 seeped into the texts, and the model copies it. Seven layers, seven colors of the rainbow, etc.
AI isn’t a machine that takes all of humanity’s knowledge and distills “the truth” from it. Like I said: training data and internet searches. And when choosing among them, it weighs factors like patterns in the training data, token distributions, how the prompt is phrased and sampling settings. That’s what it serves you. Even to give a random answer, it imitates human choice patterns.
Why we worship it
The first reason is authority inflation. Everyone presenting themselves as an authority these days has bred distrust. Biased research, sadly, does exist. Not to mention disinformation. All of this has planted an “everyone’s biased” feeling in people. The machine is perfectly suited to the role: no face, no party, and you can’t see who pays its salary.
The second reason is fluency. The human brain tends to accept a well-built, confident text as true. AI writes every answer with the same confidence. On contested topics, if the first answer is wrong, the following ones usually are too. Ask “Are you sure?” and it’ll apologize and claim to correct itself. What it actually does is give you another wrong answer.
The reverse happens too. If the first answer is right but the user pushes back, models tend to side with the user and change their answer. The literature calls this “sycophancy.”
The third reason is escaping effort. No one wants to read dozens of sources. No one wants to deal with contradictions. Saying “we don’t know” feels like losing the argument. AI removes all of that. But in return, it asks one thing: that you stop questioning.
I’ve already covered the ironic part. If AI gives the wanted answer instead of the true one, it’s a god; if it says otherwise, it’s the devil. But that doesn’t last. In the next argument, people consult AI again.
1984
It’d be rude to Orwell not to tie this in.
AI used in a factory isn’t much of a danger. If its output falls outside spec, someone notices. Even if another AI wrote the standard and defective parts somehow got made, physical reality still catches the error. If a factory that makes 5-meter cables is turning out 5-centimeter cables, something’s wrong. 5 m versus 5 cm isn’t relative information. And checking it is simple and easy.
The real issue is which institutions use the information these systems produce, who audits it, and whether people still have access to alternative sources.
What happens when the systems that produce information and the systems that verify it end up in the same hands?
AI has taken over many professions. And it keeps going. Imagine this spreading further. AI historians, say. Or sociologists. Or biologists. Imagine AI priests, even. Then manipulate internet searches. And create a society that bows to AI without question. Congratulations! Now you can rule people however you like.
This isn’t impossible. In 1984, the Ministry of Truth rewrote old newspapers. Today the job has gotten much easier. Online news outlets can delete old posts or, with an “edit” button, claim they’d been saying the opposite all along. I don’t think people use their memory much while doomscrolling. So the thought “Wait, wasn’t this story different?” never forms. They react in the moment and get on with their lives.
Not a god, an assistant
Instead of “is this true?”, you need to ask: “what’s the source of this claim, which historians dispute it, where’s the primary document?” A good assistant quickly gathers sources, summarizes them and shows you where to start. But it doesn’t sign the report.
We created AI. To support us. To make our work faster and easier. Worshipping something we created ourselves, entrusting everything to it, is ignorance. And the ignorant are easy to exploit.
Leave a Reply