Two Futures

"The illiterate of the 21st century will not be those who cannot read and write, but those who cannot learn, unlearn, and relearn."
— Alvin Toffler, 1970

I woke up in the middle of the night, sleepless, with a question. Bleary-eyed, I typed into Claude:

“Thought experiment: Paint a picture for me. What would the world look like if full AGI arrived, but everyone outside the frontier labs used AI like Google?”

Einstein called them Gedankenexperiments. Thought experiments. He chased a beam of light in his mind to discover special relativity. He imagined a man falling in an elevator to unlock general relativity. Thought experiments strip away everything irrelevant until only the essential truth remains. So at 3:00 a.m. I ran one.

Crank a lever in one direction. Crank the lever to the opposite extreme and see what happens. Then analyze the difference. I was trying to understand what would have a greater impact on society: the investment in achieving AGI or an investment in human intelligence in using AI.

World one: AGI arrives. Most people are ignorant. Claude described a world where the frontier labs and the organizations that learned to truly work with AGI operated in a completely different economic reality. They would build products, companies, and solutions in hours. Research would happen at 100× speed. Entire industries would compress into the hands of small, AI-fluent teams. Everyone else would still be typing “how to fix my sink” into a chatbot.

The result, in Claude’s words:

“Economic consolidation beyond anything we've seen. The companies and individuals who learned to work with AGI would compound advantages so rapidly that traditional competition becomes meaningless. A single architect with AGI could outdesign entire firms. A researcher could outpace whole departments. The "frontier labs and their partners" wouldn't just be OpenAI and Anthropic - it would be every organization that figured out how to actually use the tools.

A two-tier economy where most jobs exist to serve the 5-10% who are AGI-fluent. Think Victorian England but the gap isn't land ownership - it's cognitive leverage. You'd have massive employment in service roles, entertainment, and human-touch industries, while all leveraged knowledge work concentrates in a smaller and smaller elite.

Political instability as nations fracture between those who prepared their populations and those who didn't. Countries that treated AI like a better Google would watch their educated classes flee to places where AI literacy was taught systematically. Digital nomadism but as genuine economic migration. The UAE, Singapore, maybe parts of Latin America leapfrogging traditional powers because they moved faster on education.

A crisis of meaning as billions of people find that the skills they spent decades building - analysis, research, writing, even coding - can be done better by systems they don't know how to use. Not unemployed necessarily, but underemployed in ways that feel existentially hollow.

Education systems in collapse because they'd be optimizing for a world that vanished. Universities teaching people to do tasks that AGI handles trivially, while the actual valuable skill - learning to architect solutions with AI - remains informal, tribal knowledge passed between those who figured it out.

But here's the darkest part: The world would look fine for a while. GDP would be up. There'd be abundant entertainment and services. Cheap everything. It would feel like prosperity even as the foundation was rotting - because the majority would be economically sustained but cognitively obsolete, living in a world shaped entirely by decisions made by the AGI-fluent minority.”

I replied, “Thanks. That’s what I thought. Now the second scenario: Frontier model development freezes. AI is a snapshot of what it is today. But every person in every industry becomes as tech-fluent as the researchers in those labs. What does that world look like?”

Claude came back: “Now THAT'S the interesting scenario.”

“Radical decentralization of capability. Every small business, every independent professional, every team suddenly operating with leverage that used to require massive organizations. The "tech giants" advantage would compress because their moat wasn't the AI itself - it was knowing how to use it. A 10-person studio in Lagos could compete with a 500-person studio in LA because everyone finally understood the tools.

An explosion of creativity and entrepreneurship as the bottleneck shifts from "can I afford the talent?" to "do I have a good idea?" The barrier to testing, iterating, building drops so low that you'd see thousands of experiments that never would have happened. Not just in tech - in manufacturing, agriculture, education, healthcare. Local solutions to local problems, built by people who understand their context.

Industries transform from the bottom up instead of top-down. Teachers redesigning education because they understand how to use AI for personalized learning. Doctors creating diagnostic tools for their specific patient populations. Farmers optimizing for their precise conditions. The innovations would be contextual, diverse, human-centered - because they'd be built by people embedded in those problems, not extracting value from them.

The middle class explodes. Not disappears - explodes. Because suddenly individual leverage isn't tied to institutional access. You don't need the McKinsey training, the Stanford degree, the corporate resources. You need to understand how to work with the tools. The economic value of "knowing how to use AI well" would be high, but learnable. Teachable. Democratized.

Power shifts to communities and regions that invest in this literacy. The UAE's AI education initiatives actually matter. India's tech training ecosystem becomes even more powerful. Latin American developers who already learned to be scrappy with technology suddenly have equal footing. Geography matters less; capability matters more.

But here's what's really different from scenario one: Innovation would be messy and human and surprising. You wouldn't have a handful of AGI systems deciding the future. You'd have millions of humans, fluent in AI, building the future they want to live in. The Cambrian explosion wouldn't be in AI models - it would be in human applications of AI.”

It’s clear which world we want. The world with empowered humans.

So why are we spending hundreds of billions of dollars racing toward AGI and so little on democratizing the skills that will so clearly drive the benefits of this technology?

The world of AGI with ignorant humans has echoes of I.J. Good's haunting 1965 prophecy: "There would then unquestionably be an 'intelligence explosion,' and the intelligence of man would be left far behind." In The Adolescence of Technology, Dario Amodei, the founder of Anthropic, describes two of his greatest fears: labor market displacement and concentration of economic power. He predicts that 50% of white-collar jobs will be displaced within 5 years. This is what Andrew Yang calls, "the great disemboweling of white-collar jobs."

But most people who talk about the problem stop short of the solution: Upskill the humans!

I’ve gotten to ask famed AI researchers, leading VCs and leadership from the most important AI companies about the future of work and they are clear: there will be jobs. There will be new jobs. When asked how you train people for those jobs, their answer is, “well, that’s the hard part.” But that’s the important part!

This blind spot runs even deeper when it comes to emerging markets. In Machines of Loving Grace, Dario's companion essay on AI's potential upside, the billions of people in the developing world appear as beneficiaries, receiving drought-resistant crops, malaria vaccines, technologies invented elsewhere. The possibility that billions of the world's poorest might one day be enabled by AI, becoming creators, building their own solutions to their own problems, barely registers.

If this thought experiment has even remnants of truth, that the upskilling of humans to thrive in a world with AI is an important variable, then we cannot wave away “the hard part” as someone else’s problem. We must tackle that problem.

So I go back to my question: why are we investing hundreds of billions into building AGI, and almost nothing into teaching people to use it? That is the question of our generation. Why are so few people asking that question?