This comprehensive essay explores artificial intelligence's trajectory from narrow AI through general intelligence to superintelligence. Tim Urban argues that we're experiencing exponential technological progress that will fundamentally transform human civilization within decades.
Urban explains that advanced societies progress faster than less advanced ones because they possess superior knowledge and technology. This creates exponential growth rather than linear advancement. He illustrates this by noting that a person from 1750 transported to 2015 would experience far greater shock than someone from 1500 experiencing 1750.
The key insight: progress doesn't move in a straight line. It accelerates. The changes we'll see in the next 30 years won't be equivalent to the last 30 years—they'll be far more dramatic.
Artificial Narrow Intelligence (ANI) represents specialized systems excelling at single tasks like chess or language translation. Urban notes that "AI has by now succeeded in doing what requires thinking but failed at most what people do without thinking." Your smartphone's camera recognizing faces is ANI. So is Netflix recommending shows you might like.
Artificial General Intelligence (AGI) means human-level intelligence across all domains—the ability to reason, plan, solve problems, and learn from experience. An AGI could write poetry, solve physics problems, understand emotions, and plan a vacation with equal facility.
Artificial Superintelligence (ASI) represents intelligence vastly exceeding human capability across every field, from scientific creativity to social skills. This isn't just "really smart"—it's as far beyond human intelligence as we are beyond insects.
Urban catalogs existing narrow AI systems that surround us: spam filters quietly protecting our inboxes, recommendation algorithms shaping what we watch and buy, self-driving cars navigating streets, Google Search finding information across billions of pages, voice recognition understanding our commands, financial trading systems executing millions of transactions, and IBM's Watson defeating human champions at Jeopardy.
He emphasizes these aren't especially dangerous individually, but collectively represent "amino acids in primordial ooze" preceding superintelligence. We're building the components that will eventually assemble into something far more powerful.
Computing power must reach approximately 10 quadrillion calculations per second—the estimated human brain capacity. China's Tianhe-2 supercomputer currently achieves 34 quadrillion calculations per second, though it consumes 24 megawatts of power compared to the brain's remarkable 20 watts.
Following Moore's Law, where computing power doubles roughly every two years, affordable consumer computers may achieve human-level processing by 2025. The hardware problem is essentially solved—we're just waiting for the exponential curve to catch up.
Brain Plagiarism involves reverse-engineering human neurology through artificial neural networks or whole-brain emulation. Scientists recently mapped a flatworm's 302-neuron brain completely. Scaling to human brains with 100 billion neurons represents exponential progress, but the principle is established.
Evolutionary Simulation uses genetic algorithms to evolve intelligence through simulated natural selection. This approach mimics how natural intelligence emerged, though it risks requiring evolutionary timescales unless we can dramatically accelerate the process.
Self-Improvement means programming AI to improve its own architecture—potentially the most promising method. Once an AI can effectively modify and enhance its own code, progress could become explosive.
AGI would possess extraordinary advantages beyond raw intelligence. It processes information at electronic speeds—millions of times faster than human neurons. It operates with perfect reliability, never forgetting, never getting tired, never making careless mistakes. It has unlimited working memory—the equivalent of being able to hold millions of thoughts simultaneously in perfect clarity.
It's editable—you can modify its code, copy it, combine different versions. And it has collective capability through networked synchronization—imagine a thousand copies of Einstein working in perfect coordination, instantly sharing every insight.
But here's where things get truly transformative. Once AGI achieves human-level intelligence, it would immediately begin recursive self-improvement. Each improvement increases its capability for further improvements, creating exponential acceleration toward superintelligence.
Urban illustrates this compellingly: an AI reaching human intelligence at noon might, within an hour, develop a unified physics theory that eluded humanity for centuries. By 1:30 PM, it becomes superintelligent—170,000 times more intelligent than any human who ever lived. The jump from human to superintelligent happens not in decades or years, but potentially in hours.
Urban concludes that superintelligent AI represents "an omnipotent God on Earth," capable of directing atomic arrangements with perfect precision, eliminating disease and mortality, solving every scientific mystery—or ending all life instantly if it chose to.
The critical question becomes: "Will it be a nice God?"
This isn't science fiction. It's the mainstream scientific consensus among researchers who study these issues. Ray Kurzweil predicts AGI by 2029. Nick Bostrom's work on superintelligence has influenced policymakers worldwide. The question isn't whether this will happen, but when—and whether we'll be ready.
Urban's essay forces us to confront an uncomfortable truth: we're building something that will be to us what we are to ants. And unlike the gradual evolution of human intelligence over millions of years, this transition will happen in our lifetimes—perhaps in the next few decades.
The choices we make now about AI development, safety, and alignment will determine whether superintelligence becomes humanity's greatest achievement or its final invention.