Why the Planes Will Not Come
I spoke to Richard Feynman about AI adoption, and this is what he said.
TL;DR: This post is a thought experiment. I built a research base from Richard Feynman’s primary sources and used Claude to reconstruct an interview with him about AI adoption. What follows is that interview. Every answer is grounded in something he actually said. Skip to the interview if you already know who Feynman was.
I was talking to my son recently when the subject of science came up. I mentioned Richard Feynman. He’d heard the name but wasn’t sure he knew his work. “I keep coming back to Feynman”, I told him.
Here is why.
One of the youngest senior scientists Oppenheimer recruited as he assembled the free world’s greatest minds for the Manhattan Project, Richard Phillips Feynman was 24 when he arrived at Los Alamos in 1943. He was already causing problems: picking the locks on filing cabinets containing classified documents, leaving notes inside to demonstrate they weren’t secure.
He went on to win the Nobel Prize in Physics in 1965 for his work on quantum electrodynamics. And in January 1986, aged 67, he sat on the Rogers Commission investigating the Space Shuttle Challenger disaster and did something no one else on the panel had thought to do. He dropped a piece of O-ring material into a cup of ice water, left it for a few minutes, then held it up for the cameras. It had lost its resilience. The engineers had known about this. The temperature on launch morning had been well below the material’s safe operating range. The organisation had looked at the evidence and decided not to see it.
Two years later, on the day he died, there was something written on his blackboard at Caltech. Nobody is certain who wrote it first. It said: “What I cannot create I do not understand.”
Feynman is consistently ranked amongst the greatest scientific minds of the 20th century. Most personal accounts of him settle on his charm and his genuine love of the bongo drums. The question beneath all his public work was the same one: not what do we call this thing, but what is it actually doing?
If you want to hear him think out loud, two BBC documentaries are good starting points: The Pleasure of Finding Things Out (1981) and Fun to Imagine (1983). Both are on YouTube. Watch the magnets exchange in Fun to Imagine specifically: it is the clearest demonstration of how he refuses to answer a question without first insisting that the question is honest.
Many of us use his thinking methods in our own practice. You’ll find these distilled down into productivity hacks in business newsletters across the internet. His public-facing work was about knowledge and understanding. Being able to say what you know about something without using someone else’s words. The idea being, if you cannot explain it without the jargon, you do not truly understand it. You just have the name for it.
When AI became the dominant conversation in every boardroom and leadership conversation I encountered, I found myself reaching for Feynman again. Not for answers. For the questions.
So I asked myself: what if I could actually speak to the man himself?
He died on 15 February 1988. But his documented thinking is consistent enough and rich enough to be interrogated. So I used Claude to build a research base from Feynman’s actual words: transcripts, speeches, interviews, and his autobiography, not third-party articles about him.
Then I asked him.
What follows is a reconstruction grounded entirely in his documented words. Every position he takes is drawn from something he actually said or wrote. Some connections are inferred; where I have inferred, the inference follows a consistent pattern across his published work.
The Interview
Des: There is a great deal of activity in organisations right now around AI. Pilots, training programmes, transformation roadmaps. Most leaders would say they are embracing it. What do you make of it?
Feynman: Well, I would have to ask: what do you mean by “embracing”? Because there is a difference between doing something and understanding what you are doing. I gave a talk to science teachers once, in 1966. There was a first-grade science book with a picture of a toy dog and the question, “What makes it move?” The answer in the teacher’s edition was: “Energy makes it move.” Now, that is not science. That is a mystic formula for answering questions. You could just as well say “God makes it move.” Nothing is learned. I suspect a great deal of what is happening in organisations with AI is something like that. They have learned the words.
Des: In your 1974 Caltech commencement address you gave this a name: Cargo Cult Science, after the South Pacific islanders who built bamboo runways and carved wooden headphones trying to bring back the cargo planes that stopped coming after the war. Is that the pattern you are describing?
Feynman: Yes. And I want to be careful here, because the people who built those runways were not stupid. They were doing something quite rational given what they could see. They had watched the planes land during the war. They saw the airfields, the radio towers, the rituals. The planes came. Then the war ended and the planes stopped. So they built wooden headphones, bamboo antennas, their own runways. They did everything correctly, from the outside. And the planes did not come. What they could not see was the supply chain. They had access to the surface of the thing, not the mechanism underneath. That is what I think you are describing. The organisations have seen what AI looks like from the outside. They are building the runways.
Des: How would someone know whether they actually understand something, rather than just being able to describe it? Is there a test?
Feynman: There is a very simple one. You say: without using the new word you just learned, tell me what you know. Without using the words “transformation,” “capability,” “AI-powered,” tell me what your organisation can do now that it could not do before. What problem can you solve? Can you point to it? You see, knowing the name of something does not mean you understand it. I had a boy show me a bird once, a brown-throated thrush. In Germany they call it a Halzenfugel. He knew the name in English, in German, in every language you like. But even if you know all those names in every language, you still know nothing about the bird. You only know something about people: what they call it. Now that thrush sings, and teaches its young to fly, and nobody knows how it finds its way. That is the thing itself.
In Brazil, I found students who could recite the definition of Brewster’s Angle without hesitation. Perfectly correct. Then I took them outside and pointed at the ocean. Polarised light, right there. They had never connected the definition to the thing in front of them. An organisation that can describe its AI strategy fluently but cannot point to a changed capability is in exactly that position. It knows what people call the thing. It does not know the thing. You see?
Des: What should a leader actually do? Not strategise or hire a vendor. Do?
Feynman: Get closer to the machine. Not the polished interface, not the dashboard. The actual thing. I remember at Princeton, the cyclotron was in the basement, wires everywhere, physicists physically climbing around the equipment. At MIT they had a beautifully engineered setup: a remote control room, very proper. Princeton was getting results. MIT was not. The people who had built the instrument knew where everything was. They had a kind of knowledge that only comes from being in contact with the actual thing when it breaks, when it does something unexpected.
And I think, though this is perhaps a less obvious thing: there is no value in making yourself too serious about it. I spent an afternoon once watching a plate someone had tossed in the cafeteria, watching how it wobbled, working out the ratio of the wobble to the spin. No importance whatsoever. I was just doing it for the fun of it. And that is where the work came from that eventually won me the Nobel Prize. I get a kick out of this. I cannot stop. If you want your organisation to actually understand something, protect the time when people are allowed to be curious about something that has no immediate return. That is not a luxury. That is where the real learning lives.
Des: Is there reason to be hopeful?
Feynman: Oh, certainly. The cargo cultists were not beyond hope. What they needed was access to the mechanism. Not more training. Proximity to how it actually works. I saw this at Los Alamos during the war. There was a team running calculations on IBM machines: capable people, working painstakingly slowly, because the military had given them tasks but no context. No why. When I got permission to explain what they were actually calculating, the whole thing transformed. They began inventing their own processes. They went roughly ten times faster. Same people. Same machines. What changed was that they understood what they were doing and why.
That is where I would start. You see?
This experiment uses primary sources only: actual transcripts, documented speeches, verbatim recordings, and snippets from his autobiography. Every exchange above is grounded in something Feynman said, applied to a challenge he never encountered but would have recognised immediately.
The methodology is worth explaining because it is the point.
When you build a sufficiently rich model of how a thinker actually thinks, grounded in what they genuinely said and believed across decades of documented work, you can ask them questions they never answered and arrive at positions they would likely have held. The constraint is the research. Without it, you get fluent nonsense. With it, you get something the thinker might actually have said.
I started by building a model of how he actually spoke: his sentence structures, his characteristic word choices. Or, as I learned from Rebecca, The Strategic Linguist, I mapped his syntax, his semantics, and his pragmatics: how he built sentences, what meanings he reached for, and how he used language as a social act.
The research requirement is the discipline. The interview only works if you can ground every answer in something the thinker actually said or believed. That is what separates this from asking an AI tool to “think like Feynman.” The difference between a reconstruction grounded in primary sources and a simulation is exactly the difference Feynman identified in Brazil: one understands the thing and can point to a real-world example, while the other knows the name.
I do realise that this is not the same as interviewing the great man himself. That would be me fooling myself. And as Feynman famously said:
“The first principle is that you must not fool yourself, and you are the easiest person to fool.” Richard P Feynman
Feynman would also tell me not to worry about whether it matters.
This is an invitation. Pick someone whose thinking you find interesting. Pull the thread. See what comes out. Attach no importance. Just have fun with it.
If you want to spend time with Feynman on his own terms, start with Surely You’re Joking, Mr. Feynman! It is a collection of stories from a life lived in permanent curiosity. The physics barely appears. What you get is a portrait of someone who treated learning as genuinely enjoyable, and who found that the most interesting things happened when he stopped caring whether what he was investigating had any immediate use. It is the best account I know of what it looks like when someone refuses to accept the name of a thing as a substitute for understanding it.
References
Most of the primary sources used in this article are freely available:
“What is Science?” (NSTA address, 1966). Full text.
“Cargo Cult Science” (Caltech commencement address, 1974). Full text on Caltech’s own site.
The Pleasure of Finding Things Out (BBC Horizon documentary, 1981). Free full documentary on Archive.org.
Fun to Imagine (BBC series, 1983). Full series on YouTube. Magnets exchange only (7 mins).
Interview transcripts at Farnam Street: several key exchanges in full.
The Feynman Lectures on Physics: freely available online.
Surely You’re Joking, Mr. Feynman! is available wherever good books are sold.
Great experiment, Des! Very creative and informative. I once had a C‑suite client interrogated by AI and borrowed some ideas from the Big Three firms, but your approach is even better.
For the record this post confused me for at least 3 minutes until my brain slowed down enough to comprehend the premise. 😁🤣