AI Gives You the Answer First, and That Gets Thinking Backward
Innovation theorist John Nosta says polished chatbots tempt us to skip the confusion and friction that build real understanding.
Business Insider · AI isn't making us smarter — it's training us to think backward, an innovation theorist saysJohn Nosta, an innovation theorist and founder of NostaLab, worries less about what artificial intelligence knows than about the order in which it hands it over.
Human thinking, as he describes it, usually moves through confusion, then exploration, then tentative structure, and finally confidence. People get lost before they get sure.
AI flips that sequence, he argues. "With AI, we start with structure," he said. "We start with coherence, fluency, a sense of completeness, and afterwards we find confidence."
By that, Nosta means the system is not working through a problem the way a person does. It is assembling the words that are statistically most likely to come next. The result arrives polished, complete, and confident sounding, and that creates what he calls a powerful illusion.
The problem, in his view, is that people accept that polish and stop too soon. Because AI-generated answers sound authoritative, he said, people often skip the slower effort of doubting, testing, and fully grasping them. "Coming to the answer first is an inversion of human cognitive process," Nosta said. "That's antithetical to human thought."
Nosta does not argue that machines will fail to beat humans at raw computation. He said that is inevitable. His concern is that people will outsource the most valuable part of thinking itself.
"It’s the stumbles, it’s the roughness, it’s the friction that allows us to get to observations and hypotheses that really develop who we are," he said.
Questions people ask
If AI just predicts language patterns, why does it sound so sure of itself?
Large language models are trained to produce the most likely next words, and likely wording often sounds smooth and authoritative. Confidence in tone is part of fluency, not proof that the model checked facts or reasoned. That is why a false answer can read just as cleanly as a true one.
Is there a better way to use AI without skipping the thinking part?
Use it as a sparring partner rather than an answer machine. Ask it to question your draft, offer counterarguments, or show its steps and sources, then verify those yourself. The extra back and forth restores some of the exploration that a single polished answer removes.
Does confusion actually help learning?
Yes, within limits. Psychologists find that struggling with a problem, making guesses, and getting feedback helps people remember and transfer ideas better than simply being told. The key is productive confusion followed by guidance, not endless frustration.