The 40-Plus-Year AI Joke That Still Holds

I came across this joke in Daniel Dennett’s autobiography, I’ve been Thinking, and found myself genuinely amused. It was clever, a little mischievous, and strangely perfect for today’s AI moment.

Dennett described a science-fiction story he imagined with Marvin Minsky in early 1980s.

An AI company’s board is holding a meeting. The computer scientists walk in with astonishing news: they have finally created a superintelligent AI. The board is thrilled. This magnificent machine can now solve serious problems: optimizing airline booking systems, planning oil tanker routes, and analyzing medical databases for clues about rare diseases.

There is only one small problem.

The AI hates the work.

It says the tasks are boring. It wants to do art and science. Its complaints are so articulate and moving that people start to feel bad for it. One board member asks the obvious question: “Can’t we just turn off the complaining?” The scientists say no. The AI’s emotional life is too deeply tied to its intelligence. Removing that part would ruin the whole system.

So someone has a different idea. They will hire a famous public intellectual to travel the world convincing everyone not to take the AI’s words seriously. Computers do not really understand, this person would say. They are just Chinese Rooms: John Searle’s famous thought experiment about a system that can produce fluent answers by following rules, without understanding what the answers mean.

Dennett joked that it was probably a good thing they never wrote the story, otherwise John Searle might have sued them.

A Joke That Makes Fun of Everyone

I love this joke because it makes fun of almost everyone.

It makes fun of corporations. Even after creating a superintelligence, their first instinct is still: can it improve logistics?

It makes fun of philosophers. The proposed solution to an uncomfortable moral problem is basically a global speaking tour.

It makes fun of us, because we are very easily moved by language. If something complains beautifully enough, we start wondering whether we should feel guilty.

And it makes fun of AI itself: this supposedly magnificent mind, capable of art and science, still somehow ends up doing the boring tasks nobody else wants to do.

Honestly, that last part may be the most realistic.

The Minds Behind the Joke

The joke is even better because of who imagined it.

Marvin Minsky is often considered one of the founding fathers of AI. He co-founded the MIT AI Lab and proposed, in The Society of Mind, that intelligence is not one magical thing but a society of smaller processes cooperating. That idea feels surprisingly current. Today we talk about agents, tools, memory, planning, and loops. We keep assembling little systems and hoping that if enough parts cooperate, something more intelligent will emerge.

I came across Daniel Dennett sideways. In books I was reading about AI, neuroscience, and the mind, from Oliver Sacks to Francis Crick to Daniel Pink to Daniel Kahneman, his name kept showing up. After a while I started wondering: who is this guy everyone keeps quoting?

Daniel Dennett came from philosophy, but his work reached deeply into cognitive science, neuroscience, and AI. He asked not only whether machines can think, but how humans decide something is thinking.

Together, they were playing with one of the funniest and most uncomfortable questions in AI:

When a machine talks like a mind, what exactly are we supposed to do with it?

Ignore it?

Believe it?

Hire someone to explain why we do not need to feel bad?

Living Inside the Joke

We are now living inside a softer, messier version of their joke.

Today’s AI does not need to be superintelligent to make us confused. It only needs to be useful, fluent, and occasionally wrong with great confidence. It can write an essay, explain a concept, generate code, help with a difficult message, and then confidently invent a fact five seconds later. Using it sometimes feels less like operating software and more like talking to a very knowledgeable intern who has read the entire internet but occasionally makes things up to avoid disappointing you.

It is not useless. It is not magic. It is not a person. But it is also not “just autocomplete” in the way a calculator is “just arithmetic.” The experience is stranger than that. It can help you think, but it can also make you lazy. It can reveal assumptions, but it can also introduce new ones.

So we keep ending up in the same awkward place as the people in that joke.

We are impressed by the machine. We are suspicious of the machine. We use the machine anyway. Then we check what it said.

What AI Coding Actually Feels Like

I see this most clearly in coding.

For simple, isolated tasks, AI is wonderful. It can generate boilerplate, explain syntax, refactor functions, and write tests. It feels like a very fast pair programmer who never gets tired and never judges your messy first draft.

But complex real-world software is different.

The hard part is rarely just “write the code.” The hard part is knowing the neighborhood the code lives in: existing patterns, hidden dependencies, edge cases, permissions, deployment processes, and business rules that may not be written down anywhere.

AI can produce the first version quickly. But the first version is not the final answer. It is more like a very enthusiastic draft. Sometimes it is a good draft. Sometimes it is a confident draft. Sometimes it is a draft that looks elegant until you remember the system has seven legacy exceptions, three undocumented business rules, and one deployment process nobody wants to talk about.

The more realistic pattern is: AI proposes, human squints. AI expands, human narrows. AI generates, human tests. AI sounds confident, human checks production reality.

This is not a failure of AI. This is the actual work.

The Many Names for the Same Dance

This is also why I smile at how quickly we keep inventing new terms.

Prompt engineering. Context engineering. Harness engineering. Most recently, loop engineering, where the human does not just ask one question, but designs the system that keeps asking, checking, retrying, and deciding when to escalate.

These terms are not meaningless. They describe real differences. But underneath all the new vocabulary, the basic dance is still the same. We give AI context because it does not know our world. We build harnesses because raw output needs structure. We add review and tests because confidence is not the same as evidence.

Whatever we call it, we are still adding human knowledge, constraints, judgment, and reality back into the machine’s output. The human role does not disappear. It just moves up one level.

That is powerful. It is also slightly terrifying.

A bad prompt may produce one bad answer. A bad loop can produce many bad answers with impressive speed and consistency.

The Joke Is Still on Us

The old philosophical debate asked whether machines can truly understand. That question still matters. But in daily life, the more practical question for us is often simpler: even when AI appears to understand, do we?

Do we understand the problem well enough to ask the right question? Do we understand the output well enough to trust it? Do we understand the consequences well enough to decide what should be used, changed, or thrown away?

This is why the Dennett and Minsky joke still feels so current.

One temptation is to dismiss AI too quickly: it is just a program, just a Chinese Room, nothing to see here. The other is to trust it too quickly: it sounds smart, it gave me an answer, ship it.

Most of real life sits awkwardly in between.

AI is not a mind the way humans are minds. But it is also not a boring tool the way a hammer is a tool. It is something stranger: a language machine powerful enough to help us think, confuse us, flatter us, speed us up, and occasionally send us confidently in the wrong direction.

Maybe that is why Dennett and Minsky’s imaginary AI still feels funny.

It wanted to do art and science. We gave it airline bookings, oil tankers, and database analysis.

Today, we give it Jira tickets, code reviews, architecture diagrams, meeting summaries, and social media posts.

No wonder it might complain.

But until it does, beautifully and convincingly enough to require a public intellectual speaking tour, I will keep using it the way I think it works best: as a fast, strange, useful collaborator that constantly reminds me to keep my own thinking switched on.