Do you think AI will make you smarter?
Probably not, according to Nobel Prize-winning physicist Saul Perlmutter, who is credited with discovering that the universe's expansion is accelerating.
He said the biggest dangers of AI are psychological: it can give people the illusion that they understand something when they don't, and it can impair their judgment as technology becomes more integrated into our daily work and learning.
“The trouble with AI is that it can give the impression that you've learned the basics before you actually have.” Perlmutter spoke on a podcast episode with Norges Bank Investment Group CEO Nikolai Tangen on Wednesday.
“There's a bit of a danger that students might find themselves relying on it a little bit sooner, before they know how to do the intellectual work themselves,” he added.
Perlmutter said the answer is not to dismiss AI completely, but to treat it as a tool, one that supports thinking rather than doing it for humans.
Use AI as a tool, not a replacement
Perlmutter said AI can be powerful, but only if users already know how to think critically.
“The good thing is that if you know all the different tools and approaches on how to think about a problem, AI can often help you find the information you need,” he said.
At the University of California, Berkeley, where Perlmutter teaches, he and his colleagues have developed critical thinking courses that center around scientific reasoning, such as probabilistic thinking, error checking, skepticism, and structured disagreement, and teach these courses through games, exercises, and discussions designed to automatically develop those habits in everyday decision-making.
“I challenge students to really think about how they can use AI to operationalize this concept, to actually use it in their daily lives,” he said.
confidence issues
One of Perlmutter's concerns is that the AI will often speak with more certainty than necessary and could become “overly confident” in what it says.
The challenge, Perlmutter said, is that AI's confident tone can short-circuit skepticism, and people are likely to accept the AI's answers at face value rather than question whether they are correct.
That confidence, he says, reflects one of humans' most dangerous cognitive biases: trusting information that appears authoritative or that confirms existing beliefs.
To counter that instinct, Perlmutter said, people should evaluate AI output the same way they would evaluate human claims: weighing reliability, uncertainty, and potential for error rather than accepting answers at face value.
Learn how to catch when you're being fooled
In science, Perlmutter said, researchers assume they're making mistakes and build systems to catch them. For example, scientists hide their results until they are thoroughly checked for errors, which reduces confirmation bias, he said.
The same idea applies to AI, he added.
“Many [these concepts] “It's just a tool to think about where we're being fooled. We could be fooling ourselves, the AI could be fooling itself, and it could be fooling us,” he said.
That's why AI literacy also includes knowing when not to trust its output and tolerating uncertainty rather than treating it as absolute truth, he said.
Still, Perlmutter is clear that this is not a question of permanent solutions.
“AI is going to change, and we have to keep asking ourselves: Is it helping us, or are we becoming more fooled? Are we allowing ourselves to be fooled?” he said.
