Can AI Be Wrong?

AI Does Not "Know," It Only "Predicts"

Our generation is growing up alongside AI, so it is easy to assume "whatever AI says must be right." But the truth is: AI makes mistakes often, and when it does, it is especially sneaky about it. It will say something completely made up in a voice so confident and so textbook-like that you would never suspect a thing. This phenomenon has a name: "hallucination." To understand why it happens, you first need to grasp one fact. AI does not "know" any facts at all. It is only "predicting" the next most likely word. So when it is not sure, it does not say "I do not know." Instead it strings words together that sound smooth and produces a paragraph that just sounds plausible.

Real Case One: Books That Do Not Exist

Here is a real example that actually happened. Someone asked AI: "Please recommend 5 classic books on time management." AI produced a beautiful list. Each book had a title, author, publisher, year of publication, and even a short summary. It looked extremely professional. But when this person went to libraries and online bookstores to find these books, two of them did not exist at all. The titles were AI "stitching together" pieces of real book titles; the authors were well-known scholars it had "guessed"; the publishers sounded plausible but were made up. This kind of thing is especially common when you ask AI for "citations," "reading lists," or "historical details," because AI is so good at imitating the shape of academic language.

Real Case Two: Wrong Math

Here is another example. Someone gave AI a not-particularly-hard word problem. It produced a long, rigorous-looking solution (formulas, steps, units, all in place) but arrived at a wrong final answer. Why? Because a large language model is not a calculator at heart. It is a "language predictor." It is "guessing" how the math symbols should be arranged, not actually performing the calculation. So when content requires precise computation, statistics, dates, names of people or places, or legal clauses, anything that has to be exactly right, the chance of AI errors goes up noticeably. It is not that AI "got dumber." It is just how the thing works by design.

Why Will AI Not Just Say "I Do Not Know"?

You might wonder: when it is not sure, why can it not just say "I do not know"? It has to do with how it was trained. During training, AI saw enormous amounts of text, and most of those texts were sentences that "give an answer." Sentences like "I do not know" appeared rarely. So the habit it picked up is: no matter what, generate a response that looks reasonable. Add to that the fact that AI has no real self-awareness, it does not know how much it knows or does not know, and naturally it will not stop itself. Here is the key takeaway: AI confidence does not equal correctness. The smoother it sounds, the more you should keep your guard up.

A Simple Mindset to Stay Safe

For today, build the most important mindset first: treat AI as a "very articulate classmate who occasionally talks nonsense with a straight face," not as an encyclopedia. Every specific fact it gives you (names, numbers, dates, quotes, URLs) should be considered "to be verified," not "confirmed," until you check it yourself. In the next lesson, we will learn three concrete methods to help you quickly tell whether information from AI is true or false. Use AI with a healthy dose of skepticism. That is the only way to truly get the most out of it.