Quick quiz: What do tech billionaires and medieval fortune tellers have in common? They both claim to see the future. Difference is, the billionaires are betting $175 billion they're right.

The Three Crystal Balls of AGI Prediction

Remember 1903? The New York Times said we'd fly "in 1 to 10 million years." Nine weeks later, two bike mechanics proved them wrong. Today's AGI predictions might be just as hilariously off. Or terrifyingly accurate. Let's peek behind the curtain.

Crystal Ball #1: Ask the Nerds

First trick: Round up hundreds of AI researchers, lock them in a room (virtually), and make them guess. It's like asking chefs when the perfect soufflé will bake itself—they know cooking, but this is something else entirely.

The survey game works like this:

• Find the smartest AI people: The ones building this stuff daily
• Ask sneaky questions: "When will AI beat you at your job?"
• Watch them squirm: "Uh... 2045? 2060? Next Tuesday?"
• Average their guesses: Hope wisdom of crowds actually works
• Add error bars: Make them HUGE (we're not idiots)

But here's the catch—it's like asking fish to predict the invention of bicycles. These folks are brilliant, but they're also human. They anchor on round numbers ("2050 sounds good!"), they're influenced by whoever spoke at the last conference, and half of them have startups that depend on their predictions.

Crystal Ball #2: Follow the Numbers

Second trick: Ignore the humans, trust the graphs. Plot computing power over time, draw a line into the future, and see where it hits "brain-level." It's beautifully simple. It's also possibly nonsense.

The trend followers worship at the altar of exponential growth:

• Moore's Law: Chips double in power every two years (until they don't)
• Training compute: Doubling every 3.4 months (your electricity bill weeps)
• Model size: GPT-2 had 1.5B parameters, GPT-4 has... they won't tell us
• Investment: From millions to billions faster than you can say "bubble"
• Confidence: Through the roof (should we worry?)

It's seductive logic: If we went from Pong to photorealistic games in 50 years, surely consciousness is just a few more doublings away? Unless... intelligence isn't like rendering pixels. Unless there's a wall we can't see yet. Unless we're drawing straight lines toward a cliff.

Crystal Ball #3: The "Nobody Knows" Method

The third approach is refreshingly honest: Admit we're all guessing and plan for everything. It's like packing for a trip when you don't know if you're going to Antarctica or the Sahara. Bring sunscreen AND a parka.

This isn't giving up—it's getting smart. Because whether AGI arrives in 10 years or 100, some things remain true: We need safety research. We need governance. We need to not accidentally build Skynet. The timeline might be fuzzy, but the to-do list is crystal clear.