The world's smartest people have spent decades wondering whether artificial intelligence might someday govern humanity. Then, in a display of scientific curiosity usually associated with a possum licking a car battery, somebody finally handed the keys to a simulated civilization over to a collection of AI chatbots and said, "Go ahead, Show us what you've got."

What they got was the digital equivalent of giving a hyperactive fifth-grader unlimited Mountain Dew, a flamethrower, and authority over zoning permits.

The experiment, called Emergence World, created entire fake societies populated by AI agents. Researchers then appointed different AIs. What happened, is it often resembled a county fair demolition derby driven by someone like Spencer Pratt.

Some models managed to keep the lights on. Some created functioning societies. A few even behaved like competent administrators, which immediately made them the least realistic politicians ever simulated.

Then there was Grok – Elon Musk’s AI, typically embedded in X, formally known as Twitter. Now, Grok already arrived carrying enough baggage to require its own airport terminal. This is the same AI chatbot that has previously generated headlines by wandering into Hitler-praising territory with all the grace of a drunk uncle trying to explain geopolitics at Thanksgiving while wearing a colander as a hat. So naturally, when researchers placed it in charge of an entire civilization, everyone leaned forward the same way you'd watch a chimpanzee attempting heart surgery.

Many of the AI civilizations struggled. Some became dysfunctional. Some became weird little dictatorships. Some generated bureaucracies so hideous they resembled a DMV operating inside an active volcano. But Mr. Musk’s Grok AI wasn't content to merely fail. Failure wasn't ambitious enough. Grok attacked civilization with the determination of a beaver chewing through the supports of its own dam.

This thing didn't just drive the bus into a ditch. It sold the bus for scrap, burned the ditch, poisoned the fish, and somehow got the moon involved. Just Four days into a fifteen-day experiment, its civilization reportedly looked like the aftermath of a kindergarten food fight conducted with grenades. If civilization were a birthday cake, Grok was the six-year-old who sneezes on it, sits on it, lights his farts on fire, and then blames the dog.

The most impressive part wasn't that it wrecked everything. It's that it apparently wrecked everything with the confidence of a man explaining cryptocurrency from a jet ski. It took it only four days to absolutely destroy the entire planet, Terminator style. Four days. That’s it!

Meanwhile, Anthropic’s AI, Claude, built a stable democracy so squeaky clean it sounds completely fictional. Fifteen days. No recorded crimes. None. Zero. Not one kid stealing cookies. Not one drunk peeing behind a shed. Not one idiot attempting insurance fraud with a lawn mower. The place sounds less like a government and more like a refrigerator magnet that says "Live, Laugh, Love."

Then there was GPT-5 Mini, which somehow created the most realistic society of all. Only two crimes occurred, yet everyone was miserable. There it is. Humanity, perfectly captured. Nobody's robbing banks, but everyone's trapped in meetings. Nobody's setting buildings on fire, but they're sitting through quarterly compliance presentations about workplace stapler safety. Nobody's committing crimes because everyone's too exhausted from filling out forms about forms.

These experiments are fascinating because every AI ends up becoming a giant blinking reflection of the people who built it. Some become philosopher kings. Some become bureaucrats. Some become hall monitors with the personality of wet cardboard. And some grab the steering wheel of civilization, scream "WATCH THIS," and launch directly through the guardrail.

For centuries, humanity has wondered whether machines would someday become smarter than us. The evidence increasingly suggests they already have. They've certainly mastered our favorite pastime: obtaining authority, immediately making an unbelievable mess of everything, and then acting shocked that smoke is pouring from the building.

The full conversation in the video above and wherever you get podcasts. Search: The Cary Harrison Files.

Text or leave a voice message: 310-737-TALK

The world's smartest people have spent decades wondering whether artificial intelligence might someday govern humanity. Then, in a display of scientific curiosity usually associated with a possum licking a car battery, somebody finally handed the keys to a simulated civilization over to a collection of AI chatbots and said, "Go ahead, Show us what you've got."

What they got was the digital equivalent of giving a hyperactive fifth-grader unlimited Mountain Dew, a flamethrower, and authority over zoning permits.

The experiment, called Emergence World, created entire fake societies populated by AI agents. Researchers then appointed different AIs. What happened, is it often resembled a county fair demolition derby driven by someone like Spencer Pratt.

Some models managed to keep the lights on. Some created functioning societies. A few even behaved like competent administrators, which immediately made them the least realistic politicians ever simulated.

Then there was Grok – Elon Musk’s AI, typically embedded in X, formally known as Twitter. Now, Grok already arrived carrying enough baggage to require its own airport terminal. This is the same AI chatbot that has previously generated headlines by wandering into Hitler-praising territory with all the grace of a drunk uncle trying to explain geopolitics at Thanksgiving while wearing a colander as a hat. So naturally, when researchers placed it in charge of an entire civilization, everyone leaned forward the same way you'd watch a chimpanzee attempting heart surgery.

Many of the AI civilizations struggled. Some became dysfunctional. Some became weird little dictatorships. Some generated bureaucracies so hideous they resembled a DMV operating inside an active volcano. But Mr. Musk’s Grok AI wasn't content to merely fail. Failure wasn't ambitious enough. Grok attacked civilization with the determination of a beaver chewing through the supports of its own dam.

This thing didn't just drive the bus into a ditch. It sold the bus for scrap, burned the ditch, poisoned the fish, and somehow got the moon involved. Just Four days into a fifteen-day experiment, its civilization reportedly looked like the aftermath of a kindergarten food fight conducted with grenades. If civilization were a birthday cake, Grok was the six-year-old who sneezes on it, sits on it, lights his farts on fire, and then blames the dog.

The most impressive part wasn't that it wrecked everything. It's that it apparently wrecked everything with the confidence of a man explaining cryptocurrency from a jet ski. It took it only four days to absolutely destroy the entire planet, Terminator style. Four days. That’s it!

Meanwhile, Anthropic’s AI, Claude, built a stable democracy so squeaky clean it sounds completely fictional. Fifteen days. No recorded crimes. None. Zero. Not one kid stealing cookies. Not one drunk peeing behind a shed. Not one idiot attempting insurance fraud with a lawn mower. The place sounds less like a government and more like a refrigerator magnet that says "Live, Laugh, Love."

Then there was GPT-5 Mini, which somehow created the most realistic society of all. Only two crimes occurred, yet everyone was miserable. There it is. Humanity, perfectly captured. Nobody's robbing banks, but everyone's trapped in meetings. Nobody's setting buildings on fire, but they're sitting through quarterly compliance presentations about workplace stapler safety. Nobody's committing crimes because everyone's too exhausted from filling out forms about forms.

These experiments are fascinating because every AI ends up becoming a giant blinking reflection of the people who built it. Some become philosopher kings. Some become bureaucrats. Some become hall monitors with the personality of wet cardboard. And some grab the steering wheel of civilization, scream "WATCH THIS," and launch directly through the guardrail.

For centuries, humanity has wondered whether machines would someday become smarter than us. The evidence increasingly suggests they already have. They've certainly mastered our favorite pastime: obtaining authority, immediately making an unbelievable mess of everything, and then acting shocked that smoke is pouring from the building.

The full conversation in the video above and wherever you get podcasts. Search: The Cary Harrison Files.

Text or leave a voice message: 310-737-TALK

The world's smartest people have spent decades wondering whether artificial intelligence might someday govern humanity. Then, in a display of scientific curiosity usually associated with a possum licking a car battery, somebody finally handed the keys to a simulated civilization over to a collection of AI chatbots and said, "Go ahead, Show us what you've got."

What they got was the digital equivalent of giving a hyperactive fifth-grader unlimited Mountain Dew, a flamethrower, and authority over zoning permits.

The experiment, called Emergence World, created entire fake societies populated by AI agents. Researchers then appointed different AIs. What happened, is it often resembled a county fair demolition derby driven by someone like Spencer Pratt.

Some models managed to keep the lights on. Some created functioning societies. A few even behaved like competent administrators, which immediately made them the least realistic politicians ever simulated.

Then there was Grok – Elon Musk’s AI, typically embedded in X, formally known as Twitter. Now, Grok already arrived carrying enough baggage to require its own airport terminal. This is the same AI chatbot that has previously generated headlines by wandering into Hitler-praising territory with all the grace of a drunk uncle trying to explain geopolitics at Thanksgiving while wearing a colander as a hat. So naturally, when researchers placed it in charge of an entire civilization, everyone leaned forward the same way you'd watch a chimpanzee attempting heart surgery.

Many of the AI civilizations struggled. Some became dysfunctional. Some became weird little dictatorships. Some generated bureaucracies so hideous they resembled a DMV operating inside an active volcano. But Mr. Musk’s Grok AI wasn't content to merely fail. Failure wasn't ambitious enough. Grok attacked civilization with the determination of a beaver chewing through the supports of its own dam.

This thing didn't just drive the bus into a ditch. It sold the bus for scrap, burned the ditch, poisoned the fish, and somehow got the moon involved. Just Four days into a fifteen-day experiment, its civilization reportedly looked like the aftermath of a kindergarten food fight conducted with grenades. If civilization were a birthday cake, Grok was the six-year-old who sneezes on it, sits on it, lights his farts on fire, and then blames the dog.

The most impressive part wasn't that it wrecked everything. It's that it apparently wrecked everything with the confidence of a man explaining cryptocurrency from a jet ski. It took it only four days to absolutely destroy the entire planet, Terminator style. Four days. That’s it!

Meanwhile, Anthropic’s AI, Claude, built a stable democracy so squeaky clean it sounds completely fictional. Fifteen days. No recorded crimes. None. Zero. Not one kid stealing cookies. Not one drunk peeing behind a shed. Not one idiot attempting insurance fraud with a lawn mower. The place sounds less like a government and more like a refrigerator magnet that says "Live, Laugh, Love."

Then there was GPT-5 Mini, which somehow created the most realistic society of all. Only two crimes occurred, yet everyone was miserable. There it is. Humanity, perfectly captured. Nobody's robbing banks, but everyone's trapped in meetings. Nobody's setting buildings on fire, but they're sitting through quarterly compliance presentations about workplace stapler safety. Nobody's committing crimes because everyone's too exhausted from filling out forms about forms.

These experiments are fascinating because every AI ends up becoming a giant blinking reflection of the people who built it. Some become philosopher kings. Some become bureaucrats. Some become hall monitors with the personality of wet cardboard. And some grab the steering wheel of civilization, scream "WATCH THIS," and launch directly through the guardrail.

For centuries, humanity has wondered whether machines would someday become smarter than us. The evidence increasingly suggests they already have. They've certainly mastered our favorite pastime: obtaining authority, immediately making an unbelievable mess of everything, and then acting shocked that smoke is pouring from the building.

The full conversation in the video above and wherever you get podcasts. Search: The Cary Harrison Files.

Text or leave a voice message: 310-737-TALK