The Data Pipeline for Superintelligence Starts With Your Screen

There’s a question that’s been haunting AI research for decades: how do you build a system that truly understands the world?

Not one that predicts the next token. Not one that passes benchmarks. One that actually models reality — the way a human does, the way an ecosystem does, the way a city does.

I think I know where the answer starts. And it’s not where most people are looking.

The Densest Data Structure on Earth

The human brain stores approximately 2.5 petabytes of information. That’s 2.5 million gigabytes packed into 1.4 kilograms of tissue. A Salk Institute study found that each of the brain’s 125 trillion synapses can hold about 4.7 bits of information across 26 distinct levels — ten times more than scientists previously believed.

To put that in perspective: Yahoo’s entire data warehouse, processing 24 billion events per day, holds less data than a single human brain. The IRS database tracking 300 million Americans? About 150 terabytes. Your brain holds roughly 17 times that.

You almost don’t see this kind of information density anywhere else in nature. A rainforest is staggeringly complex, but the data per cubic centimeter doesn’t come close. A coral reef, a city, a desert — all contain extraordinary information. But none of them approach the compression ratio of the three-pound organ sitting between your ears.

This matters enormously for the future of AI.

If You Want to Model Everything, Start With the Brain

Here’s the thesis: the best data pipeline for superintelligence is, first, a data pipeline for the human brain.

Think about it. If we could create a true digital twin of human cognition — not a language model trained on text, but a living model of how a specific human thinks, decides, and acts — we’d have cracked the hardest data problem on the planet. The densest, most complex information structure in the known universe, digitized.

And once you’ve built the infrastructure to model that, the same technology extends outward. Digital twins of teams. Of organizations. Of supply chains and cities. Eventually, of ecosystems — jungles, oceans, deserts, entire countries. Each one is a system of interacting agents making decisions under uncertainty. The brain is just the one with the highest data density per unit of space.

So if you’re serious about building toward superintelligence — or even just toward AI systems that truly understand the world — you don’t start with more text data. You start with the human brain.

The Screen Is the Window Into the Brain

Now here’s the practical question: how do you actually observe a human brain in action?

You could try neuroimaging. fMRIs, EEGs, brain-computer interfaces. They’re promising but limited — expensive, invasive, low-resolution for everyday use.

Or you could look at what’s already sitting in front of the brain for most of its waking hours.

For the majority of humans on this planet, the answer is a screen. A computer screen or a phone screen. Eight, ten, twelve hours a day — the screen is the primary interface between the brain and the external world.

Every click, every scroll, every document opened, every email drafted, every tab switched — these are the outputs of cognition. They’re the exhaust trail of the brain doing its work. And unlike neuroimaging, this data is abundant, high-resolution, and available right now, on every computer on Earth.

This is the insight behind what we’re building at Geodo. We’re not trying to read the brain directly. We’re reading the thing the brain talks to all day: the screen. By capturing and understanding the patterns of how people interact with their computers, we’re building the first layer of a digital twin of human work — and eventually, of human cognition itself.

From Digital Twin to Digital Ecosystem

The implications go far beyond automating someone’s email workflow — though yes, we do that too.

When you model how one person works, you understand their decision patterns, their information flows, their bottlenecks. Scale that to a team, and you’re modeling organizational intelligence. Scale it further, and you’re modeling how entire industries think and operate.

The same fundamental technology — observe agents interacting with their environment, build a model, and use that model to predict and automate — applies at every level. The human sitting at their laptop is the atomic unit. Everything else is composition.

This is why I believe the path to truly intelligent systems doesn’t start in a research lab training on internet-scale text. It starts at the most information-dense point in the natural world — the human brain — and works outward from there.

And the most practical way to start? Watch the screen.

Nadav Shanun is the co-founder and CEO of Geodo, an AI automation company building digital twins to automate repetitive computer work. Before founding Geodo, he built the first foundational AI model at NASA’s Kennedy Space Center and worked in nuclear systems at the Pentagon and satellite intelligence at the CIA. His motto: have your computer do work for you, so you can spend more time in nature.