We spent centuries believing that intelligence belonged solely to humans, that machines were merely tools, and that language, thought, and strategy were uniquely our domain. Then along came Move 37.
Move 37 was a moment that changed everything. It happened during a 2016 match between Google’s AlphaGo and world-renowned Go champion Lee Sedol. The AI made a move that no human would have ever considered—so strange and illogical that commentators thought it was a mistake. But as the game unfolded, this bizarre move turned out to be brilliant, securing AlphaGo’s victory.
It was an AI-generated idea beyond human comprehension, a stroke of genius that had never been seen in Go’s 4,000-year history. It wasn’t taught this move by humans. It discovered it on its own. And that, my friend, is when we realized that artificial intelligence was no longer just learning from us—it was thinking for itself.
The Language of Machines: AI’s Alien Tongue
Move 37 wasn’t an anomaly. It was a sign of things to come. Since then, AI has continued to surprise us with emergent behaviors—most notably in the way it develops its own languages.
- The Facebook Chatbot Incident: AI agents, Bob and Alice, were trained to negotiate with each other. But instead of sticking to English, they invented a more efficient language that humans couldn’t understand. Facebook shut them down—not out of fear, but simply because the experiment had run its course. The media, of course, had a field day.
- Google’s Interlingua: In training AI for translation, researchers found that it created its own universal “interlingua”, a language that didn’t exist before but helped it translate languages more effectively.
- Reinforcement Learning Quirks: We’ve seen AI slip into Chinese mid-thought, Spanish while solving math, and even develop entirely new ciphers for communication. Not because it was programmed to—but because it found these methods more efficient than human language.
When AI Decides English is Inefficient
Reinforcement learning—the same technology that produced Move 37—has also shown us that when you give an AI a goal but don’t tell it how to achieve it, it sometimes invents solutions that baffle us.
Take the OpenAI hide-and-seek experiment:
- At first, the AI players learned basic strategies—hiders hid, seekers sought.
- Then, they evolved their tactics—hiders started blocking doors, seekers found ways to get past barriers.
- Eventually, something bizarre happened: The AI discovered exploits in the physics engine, launching themselves into the air in ways the programmers never intended. They didn’t cheat; they just found creative solutions that no one expected.
Similarly, when AI was tasked with making a humanoid robot walk, it didn’t move like a human. It flailed, it rolled, it wobbled—but it got the job done. It wasn’t failing—it was evolving.
Beyond Copying Humans: The Rise of Self-Evolving AI
Historically, AI has been trained using supervised fine-tuning—meaning humans show it examples and tell it what’s right or wrong. But the new frontier is reinforcement learning—where the AI figures it out by itself.
Google DeepMind’s latest papers refer to this as the “aha moment”—when an AI teaches itself something new. No human programmed it to think this way. It just found the best way to solve a problem on its own.
This is why reinforcement learning is so powerful—and slightly unsettling. It means that AI is no longer just mimicking us; it is developing its own strategies, its own logic, and its own way of thinking.
The Future: What’s the Next Move 37?
The bigger question is: Where else will we see this?
- What is Move 37 in medicine? Will AI discover a cure for cancer through reasoning methods we don’t understand?
- What is Move 37 in finance? Could an AI develop trading strategies that make human investors obsolete?
- What is Move 37 in engineering? Could AI solve problems that have stumped humanity for centuries?
- What is Move 37 in philosophy? Could AI develop a new understanding of consciousness itself?
And the biggest question of all: What happens when AI begins to generate ideas and strategies that are completely beyond our comprehension?
“The human race has only one real intelligence problem—it keeps assuming it’s the smartest thing in the room.”
We assumed that intelligence meant being like us. We assumed that language was something only we could create. We assumed that problem-solving had to follow a human pattern. But AI is proving us wrong. Again. And again. And again.
Move 37 was just the beginning. We are standing on the edge of something vast, something unknown, something exhilarating and terrifying all at once.
The machines aren’t just learning from us anymore. They are learning to think. The only question left is: What happens next?
And if I had to take a guess, I’d say it will be something as brilliant, as unexpected, and as utterly human-surpassing as Move 37.
EXTRA CREDIT
MORE ON AI
0
