Nietzsche wrote: “It is better to know nothing than to half-know many things.” I used to think that was a bit harsh. After a year of working with AI every day, I think it is exactly right.
What I mean by “artificial belief”
If you know nothing about a topic, you know that you don’t know. You stay careful. You ask someone.
If you know a subject well, you can tell when an answer is wrong.
The danger is in the middle. When you half-know something, you know just enough to follow an answer, but not enough to check it. So when an AI gives you a confident, well-written answer, it feels right, and you believe it. You didn’t earn that belief by understanding anything. It was handed to you.
That is what I call artificial belief: a belief that comes from how sure the answer sounded, not from what you actually know.
2 + 3 = 5, and so does 6 + (−1)
Take the number 5. You can get there with 2 + 3. You can also get there with 6 + (−1). Same answer, completely different paths.
If all you look at is the 5, both look the same. But the answer is the smallest part of it. What matters is how you got there: understanding what the question is really asking, the context around it, and then making a decision.
Now imagine you don’t really understand the question. You can’t see the path. All you can see is the final answer, 5, and it looks right. So you trust it, and because it looked right, you start to feel that you understand too. That is how artificial belief gets into you. Not through the reasoning, which you never saw, but through a final answer that looked correct.
AI is very, very good at giving you the 5.
Two screenshots
Here is a real example. I was working with Claude Opus 5 on some code, and it had picked a number for when the program should switch to using threads. It said 200,000 elements. It sounded sure. Then the numbers came back:
The first answer was off by five times, and it sounded just as sure as the fix did.
And another time, after I pushed back on how it had explained something:
It corrected itself only because I questioned it.
Look at what happened in both cases. The model fixed its mistake, which is good. But it fixed it only because something checked it: a measurement in the first case, and me in the second. If I had half-known the subject, I would have nodded along and kept the wrong answer. Nothing in the first reply would have warned me.
Why this happens
A language model like this writes one word at a time, picking whatever is most likely to come next. That is a very powerful trick, and it is often right. But “likely” is not the same as “true,” and the model sounds the same either way. It doesn’t sound less sure when it is wrong.
So the confidence you hear tells you nothing. The only thing that tells you whether it is right is your own knowledge, or a test in the real world.
Where it costs you
Artificial belief is cheap to pick up and expensive to keep. It sits quietly in your head until the day it matters: a production system goes down, a doctor’s report needs reading, money is on the line, a decision can’t be undone. That is when you reach for what you “know,” and find that part of it was never really yours.
Half-knowledge doesn’t fail when things are easy. It fails at the critical moment, later in life, when you have the least time to find out.
Why we need more experts, not fewer
There is a growing idea that we don’t need to learn things deeply anymore, because the AI knows. I think it is the other way around. The better AI gets, the more we need people who really understand their field, because they are the only ones who can catch it when it is confidently wrong.
We should not run the things that truly matter, in our own lives or in the world, by fully trusting a system that predicts the next likely word. Use it, yes. Hand it the final say on something that can’t be undone, no.
Become the expert, then let AI do the heavy lifting
None of this means “don’t use AI.” I use it all day. It is the best helper I have ever had.
But it works best for people who already know the subject. Once you are a real expert, AI is a huge multiplier: it does the heavy lifting, and you can spot the 200K-instead-of-30K mistakes in seconds. For someone who half-knows, the same tool quietly fills their head with things that aren’t true.
So learn the thing properly first. Then let the machine help you go faster. Better to know nothing than to half-know, and best of all, to actually know.