The Constitution Was a Move 37
When I first asked myself “What’s the human equivalent of Move 37?”, I went straight to individual genius. Einstein. Darwin. Gödel.
Wrong frame.
Move 37 didn’t come from a single brilliant mind. It came from massive parallel exploration—millions of games played simultaneously, searching solution space at a scale no individual could reach.
So the human equivalent shouldn’t be a lone genius either. It should be coordinated parallel processing across multiple brilliant minds.
That’s the Constitutional Convention.
Philadelphia, 1787
Fifty-five delegates. Four months. A room full of people who fundamentally disagreed on almost everything.
They weren’t discovering physics. They weren’t climbing a known mountain. They were exploring possibility space—trying to find a stable configuration for governance that had never existed before.
Each delegate brought different frameworks:
- Madison with his deep study of failed republics
- Hamilton with his vision of centralized power
- Sherman with his practical legislative experience
- Franklin with his diplomatic pragmatism
They argued. They deadlocked. They nearly walked out multiple times.
But here’s what made it Move 37: the parallel processing architecture.
Daily sessions where insights combined. Committee reports synthesizing different approaches. Informal conversations over dinner. Written proposals circulating among delegates. Each mind operating at peak capacity, with outputs feeding into a shared problem-solving space.
What emerged wasn’t any single person’s vision. It was a configuration that no individual would have reached alone—a set of interlocking compromises and innovations that seemed strange to many at the time but proved remarkably robust.
Separation of powers. Federalism. The Electoral College. Bicameral legislature. Each of these was controversial. Some were called unworkable. Several delegates refused to sign because they thought the whole thing was wrong.
That’s what Move 37 looks like. A solution that violates the conventional wisdom of the experts in the room—but proves brilliant in application.
The Test of Time
Here’s what separates the Constitution from other examples of human collective achievement.
The Manhattan Project was parallel processing. Thousands of scientists, coordinated effort, breakthrough result. But they were discovering physics that already existed. They were mastering known rules, not finding unexploited possibilities.
The Moon landing was parallel processing. Massive coordination, brilliant engineering, historic achievement. But again—climbing a difficult but visible mountain within known physics.
The Constitution created something different. It found a stable configuration in possibility space that humans had never discovered despite millennia of trying. And unlike a scientific discovery that might be superseded, this configuration has remained functional for 250 years.
That’s the other attribute of a true Move 37: it stands the test of time.
AlphaGo’s Move 37 wasn’t just surprising—it was sound. It revealed something true about Go that remained true afterward. The Constitution wasn’t just innovative—it was durable. It revealed something true about governance that has remained true for a quarter millennium.
The Domain Problem
Go has roughly 10^170 possible board positions. That’s an incomprehensibly large number, but it’s finite. AlphaGo could explore it through self-play.
The domain of “possible governance systems” is larger. Maybe infinitely larger. The rules aren’t fixed. The win conditions aren’t clear. The feedback loops take decades or centuries.
As the domain grows wider, more parallelization is required.
This is why individual genius couldn’t find the Constitution. Madison alone, even with his extraordinary intellect and preparation, couldn’t explore enough of the space. The problem required multiple minds processing in parallel, sharing insights, building on each other’s partial solutions.
And it’s why finding Move 37 moments in open-ended domains—business, climate, urban planning, social systems—will require parallelization beyond what any individual or small team can achieve.
What This Means for AI
The Constitution suggests something important about where AI breakthroughs will come from.
Not from making individual models smarter in a human-like way. Not from achieving “PhD-level reasoning” on benchmarks designed for sequential human minds.
From massive parallel exploration of possibility spaces too large for humans to search—looking for stable configurations that have always existed within the rules but that we’ve never found.
The constitutional convention had 55 minds working for 4 months. Modern AI can simulate millions of parallel explorations in hours. The question is whether we’re pointing that capability at the right problems and whether we can recognize Move 37 when it appears.
Because when it does appear, it will probably look wrong. The experts will say it violates best practices. The pattern recognition of experienced practitioners will reject it.
Just like many delegates thought the Constitution was unworkable.
Just like the commentators thought Move 37 was a mistake.
Move 37 = finding and successfully exploiting a previously unexploited possibility that has always existed within the domain’s rules, discovered through parallel exploration at a scale that sequential processing could never achieve.
The Constitution is proof that humans can do this too—under the right conditions, with the right architecture, when the stakes are high enough to force genuine collaboration.
The question for AI isn’t whether it can match human intelligence. It’s whether it can find the Move 37s hiding in domains we haven’t fully explored yet.
They’re out there. They’ve always been out there.
We just haven’t had enough parallel processing to find them.