Opinion
What Would Richard Feynman Ask ChatGPT?
If Richard Feynman were alive in 2026, what would he ask an AI? 25 questions about quantum mechanics, probability, scientific thinking, uncertainty, and the limits of human knowledge.
If Richard Feynman were alive in 2026, he probably wouldn't begin by asking AI for answers. He would ask it to explain things simply, expose its assumptions, challenge accepted explanations, and show exactly where the uncertainty begins.
Feynman had little patience for explanations that sounded impressive without actually making something understandable. He was deeply interested in quantum mechanics, electromagnetism, computation, and probability — but equally interested in whether people genuinely understood what they were talking about.
An AI capable of instantly searching enormous bodies of knowledge would be an extraordinary tool for him. It would also be an extraordinary target for skepticism. Feynman might spend less time asking "What do you know?" and considerably more asking "How do you know?"
The Short Answer
What would Richard Feynman ask ChatGPT?
If Richard Feynman were alive today, he would likely ask AI to explain difficult ideas simply, then identify exactly what those explanations leave out. His defining question was never "What do you know?" but "How do you know?" The 25 questions below are speculative, grounded in the intellectual habits documented across his lectures and writing — above all his insistence that a convincing explanation is not the same thing as a correct one.
Feynman's reputation rests partly on his scientific achievements and partly on his unusual approach to understanding. He worked on quantum electrodynamics, participated in the Manhattan Project, helped introduce ideas that became foundational to quantum computing, and became one of the most influential science communicators of the twentieth century.
But his most enduring intellectual habit was simpler: don't fool yourself. If an explanation could not survive careful questioning, it wasn't good enough. If a calculation produced a surprising result, investigate it. If everyone agrees, check the evidence.
Give Feynman access to a modern AI system, and the machine might discover very quickly that having an answer is not the same thing as having an explanation.
Part One
What do we actually understand?
Feynman was famous for insisting on genuine understanding rather than memorization. A person can repeat an equation without understanding the physical idea behind it. And an AI can generate an elegant explanation without necessarily possessing the kind of understanding humans mean when they use the word. That would be a problem worth investigating.
Explain quantum mechanics without using any equations. Then tell me exactly what your explanation leaves out.
The second sentence is the important part. A simplified explanation is useful, but simplification also hides the difficult parts. Feynman would want both.
Give me an explanation of something you think you understand, then identify the weakest assumption in it.
This turns explanation into self-criticism. AI systems are exceptionally good at producing coherent narratives — but coherence is not proof. A convincing explanation can still be wrong.
What is the simplest experiment that would prove you wrong?
That might be one of Feynman's favorite questions. A theory becomes scientifically interesting not because it explains what we already know, but because it makes itself vulnerable to evidence.
Take the most widely accepted explanation of a fundamental physical phenomenon and explain what remains unexplained.
Science is often presented as a collection of solved problems. Feynman knew better. Every successful theory also draws a boundary around what it does not explain.
What do physicists say they understand that they actually don't?
Deliberately provocative — which is exactly why it is useful. Scientific progress depends partly on recognizing the difference between "we can calculate it" and "we understand what it means."
Part Two
What is really happening in the quantum world?
Few subjects would be more irresistible to Feynman than modern quantum physics. He helped develop quantum electrodynamics and introduced the diagrammatic language physicists still use to reason about particle interactions. He also had a complicated relationship with attempts to turn quantum mechanics into an intuitive story. The theory works extraordinarily well; understanding what it means is another matter.
What is the most misleading intuitive picture people use when explaining quantum mechanics?
Every scientific theory eventually acquires metaphors. Some help. Others become obstacles that have to be unlearned before real understanding starts.
If quantum mechanics is complete, what exactly is reality doing before measurement?
That opens the door to the measurement problem and competing interpretations. Feynman wouldn't necessarily demand a philosophical answer — he might first ask what experimental evidence could distinguish the possibilities.
What experiment would most seriously challenge our current understanding of quantum mechanics?
Not "what experiment confirms it?" The more interesting question is always what could break it.
Show me a phenomenon quantum mechanics predicts perfectly but that no human explanation of it fully satisfies.
This gets at a persistent tension in physics: prediction can be extraordinarily precise even when interpretation remains contested.
Do you actually understand quantum mechanics, or can you merely manipulate descriptions physicists recognize as correct?
Particularly interesting coming from Feynman, because it applies the problem of understanding directly to the machine.
Part Three
Can AI actually do science?
Feynman wasn't simply interested in calculating answers. He was interested in discovering how nature works. Modern AI can already assist with literature analysis, mathematical reasoning, simulation, and hypothesis generation — but assistance is not the same as discovery, a distinction that also shows up in the measured data on where AI actually improves expert work.
Give me a hypothesis humans have not seriously considered — then every reason it might be wrong.
The second request matters more than the first. Generating hypotheses is easy. Finding the ones worth testing is hard.
Design an experiment that could distinguish two competing explanations of the same phenomenon.
This is closer to real scientific reasoning. Instead of summarizing existing knowledge, identify where the knowledge branches.
What scientific discovery is hiding in existing data because nobody has asked the right question?
The world's scientific databases contain enormous quantities of information. The limitation may increasingly be not data, but imagination.
If you could perform only one experiment to discover new physics, what would you choose?
No infinite computing. No unlimited laboratory. One experiment. The constraint forces prioritization.
What would you investigate if you were willing to look completely ridiculous for ten years?
Scientific progress sometimes begins with ideas that initially sound unreasonable. The distinction that matters is between being unconventional and being unsupported.
Part Four
Can we trust the numbers?
Feynman had a deep appreciation for probability and uncertainty — and for how easily numbers create an illusion. A precise number is not necessarily a reliable one. A sophisticated model can rest on incorrect assumptions. An enormous dataset can carry systematic bias.
Give me a number you are highly confident about, then show me everything that could make that confidence wrong.
Confidence needs context. Where did the number come from? What assumptions produced it? What data was missing?
What is the difference between uncertainty caused by randomness and uncertainty caused by ignorance?
A fundamental distinction. Sometimes nature itself produces probabilistic outcomes; sometimes we simply don't know enough. Those are not the same thing.
Show me an example where more data would make our conclusion worse rather than better.
More information doesn't automatically produce better understanding. Correlated observations, selection effects, and poor experimental design can all manufacture the illusion of certainty.
What statistical result would look convincing to an intelligent person but be almost meaningless?
A question modern science badly needs. AI can generate polished charts, summaries, and correlations — but presentation isn't validation.
If a result is statistically significant but physically meaningless, what have we actually learned?
Feynman would insist on returning to the physical world. What changed? What can we predict, test, build, or observe?
Part Five
What are we missing?
Feynman was fascinated by the boundaries of knowledge and entirely comfortable saying "I don't know." Uncertainty could be the beginning of a much better question. AI creates an interesting paradox: it can produce answers about almost anything, which makes it increasingly important to identify what it shouldn't pretend to know.
What is the most important thing about the universe that we currently don't know?
Not the most obscure. Not the most difficult calculation. The most important.
What question in physics are we asking because of our current technology rather than because it is fundamental?
Scientific questions are shaped by what humans can measure. A future instrument could completely change which questions seem worth asking.
What assumption about reality is so deeply embedded in physics that we barely notice we are making it?
This is where paradigm shifts often begin — not by adding another fact, but by questioning something everyone thought was obvious.
If everything we know about physics had to be reduced to three principles, what would survive?
Strip away the terminology, the history, the notation. What remains?
What is nature trying to tell us that our current language is incapable of expressing?
Perhaps the answer is mathematical. Perhaps computational. Perhaps we haven't invented the right language yet.
The Question Feynman Might Ask Last
“How do you know when you're wrong?”
After asking AI about quantum mechanics, probability, experiments, and the limits of knowledge, Feynman might eventually turn the conversation around — not toward the universe, but toward the machine.
That question cuts directly into the problem of artificial intelligence. An AI can generate explanations, estimate probabilities, compare evidence, and even be instructed to criticize its own reasoning. None of that guarantees it knows when it has crossed from uncertainty into error. The most dangerous system isn't the one that knows nothing. It is the one that sounds certain when it shouldn't.
Feynman's greatest lesson may not be about physics at all. It may be about intellectual honesty: complicated subjects do not become more profound simply because they are hard to explain. Sometimes complexity is real. Sometimes complexity is camouflage. The job of a good scientist is to tell which is which.
That principle becomes remarkably relevant in the age of AI. A system can produce an explanation in seconds, derive the mathematics, and summarize an entire field — none of which removes the responsibility to ask whether it is true, how we know, and what would prove it wrong.
More in this series on Opinion, and related coverage across artificial intelligence.
Liyam Flexer
Founder & Editor, The Best Blog Ever — writes opinion on AI, physics, and the questions we’ve stopped asking.