Where it all began

In 2012, a neural network called AlexNet won an image recognition competition by a margin no one expected. It wasn't just an incremental improvement — it was a leap. The kind of leap that makes people look up and ask: what just happened?

What happened was that three ingredients converged at the same time: massive data (ImageNet had millions of hand-labeled images), computing power (gaming GPUs turned out to be perfect for multiplying matrices) and an algorithm that had been waiting for its moment for decades (deep neural networks).

"We've been working on this for 30 years. The difference is that now we have the data and the hardware to make it work." — Geoffrey Hinton, 2012

But the real beginning wasn't in 2012. To understand the current race you have to go further back, to the 1950s, when a group of researchers gathered at Dartmouth with an ambitious idea: to create a machine that could think.

The Dartmouth proposal, written by John McCarthy, Marvin Minsky, Nathaniel Rochester and Claude Shannon, is one of the most optimistic documents in the history of science. They estimated that a summer of work would be enough to make significant progress on the problem of artificial intelligence. It took decades.

What followed were two "AI winters" — periods where funding dried up, results didn't appear and the community lost credibility. The first occurred in the 1970s, when initial promises went unfulfilled. The second in the 1990s, when expert systems — the big bet of the 1980s — showed their limitations.

Each winter left a lesson. The first: intelligence cannot be programmed with hand-written rules. The second: data matters as much as algorithms. The third, the most important: patience is a scarce resource in technology, and the right ideas often arrive decades before the infrastructure to execute them.