Why does ‘I read it in a book’ sound smarter than ‘I asked AI’?

There is a strange social asymmetry.

Someone says, “I once read about this in a book.” Often, nobody asks which book, who wrote it, what evidence supported the conclusion, or whether the person remembers it correctly. The word “book” already sounds respectable.

But say, “I asked AI, requested sources, and checked the answer,” and some people react very differently: right, so you cannot think for yourself.

The actual quality of the result could easily be the opposite of what that reaction assumes.

This is more than a personal impression

In 2025, Scientific Reports published a series of five preregistered experiments. In one of them, participants saw the same piece of advice written by ChatGPT. One group was told it came from ChatGPT, while another was told it came from another person. When the advice was labelled as ChatGPT’s, the identical text received lower ratings for quality, effectiveness, and authenticity.

Another 2025 study in PNAS ran four experiments with more than 4,400 participants. People who used AI at work were judged as lazier, less competent, and less diligent. Participants anticipated this reaction, which gives them a reason to conceal their AI use.

So the dismissive phrase “ChatGPT wrote that for you” is not limited to comment sections. It already carries a measurable social cost.

Perhaps we are judging the cost of the answer, not the answer itself

The old chain looked roughly like this:

knows a lot → has read a lot → invested years → must be competent.

AI breaks that chain. A first overview of a subject, competing explanations, and a list of sources can now take minutes rather than days. To an observer, the result may feel too cheap.

This resembles what psychologists call the effort heuristic. In classic experiments, people rated poems, paintings, and other works more highly when they believed those works required more time and effort. The effect became especially strong when quality was difficult to judge directly.

Information may be treated in the same way. When we cannot quickly evaluate the argument itself, we reach for indirect signals: the thickness of the book, years of education, a confident voice, or the amount of time spent. AI removes a familiar signal of effort, so it can appear to remove the signal of quality as well.

But that is a poor substitute for verification.

If person A spends four hours finding a correct answer, while person B reaches the same answer in ten minutes with AI, checks the original sources, and understands the reasoning, the informational value of the result does not fall merely because the route was shorter.

A book is also a status symbol

“I read it in a book” communicates more than a source. It carries a cultural trail: education, discipline, seriousness, comfort with long texts.

“I asked ChatGPT” still carries another stereotype: did not know, did not investigate, pressed a button, copied the output.

Both stories can be false.

Someone can read nonsense twenty years ago and remember it badly. They can choose the first book they find by an author who has simply packaged personal beliefs with confidence.

Another person can use AI as a research tool: ask for competing explanations, find contradictions, request primary sources, open them, and verify what they actually say.

And of course, the reverse is possible: a person can copy the model’s first confident hallucination without thinking at all. The delivery mechanism alone does not tell us whether intellectual work took place.

“AI said so” is still a weak argument

It is easy to defend a new tool so enthusiastically that we fall into the opposite mistake. AI does not become an authority merely because it answers quickly, at length, and with confidence.

A model can invent a fact, confuse a source, omit a crucial exception, or follow the bias built into a question. Alongside algorithm aversion, there is an opposite problem: automation bias, or excessive reliance on automated advice even when contradictory information is available. Where such reliance leads once it becomes the norm and not obeying the system becomes irrational is covered in the article on invisible control.

So “it is true because ChatGPT said it” is a bad argument. Yet “it is true because I read it in a book” is not logically better.

A book, a search engine, a teacher, and a language model can all help us move along that chain. None of them removes the need to understand what we are repeating. Where normal delegation ends and loss of control over the result begins is the subject of “Delegation is not degradation”.

What is becoming valuable now

Access to information is gradually becoming less scarce. The scarce skill is the ability to frame a useful question, detect a contradiction, choose a criterion, verify a source, and apply a conclusion to a real situation.

Remembering hundreds of facts still matters, but it no longer provides the same advantage by itself: AI has made search, first-pass synthesis and information compilation cheap. That is precisely why it provokes so much irritation. It makes cheaper a set of actions that long served not only as tools for work, but also as signals of intellectual status.

The useful question is no longer whether someone used AI. It is what they did after receiving the answer.

Did they understand it? Check it? Compare it? Find its weak points? Or did they simply paste polished text?

That is what deserves judgment.