For most of its life, Google thought of itself as a search company. You typed words into a box, and an algorithm ranked the web. That algorithm was clever — PageRank had, after all, launched an empire — but it was still rules written by engineers. Somewhere in the 2010s, the people running Google began to suspect that the future belonged to a different kind of software: one that learned for itself. This is the story of how a search company decided it was really an artificial-intelligence company.
A team that taught a computer to see cats
The pivot did not start with a grand announcement. It started with a small research group. In 2011, three researchers — Jeff Dean, Greg Corrado, and the Stanford professor Andrew Ng — began building a deep-learning team inside Google, the effort that became Google Brain. Their bet was unfashionable at the time: that old-fashioned neural networks, fed enough data and enough computing power, would suddenly start doing things nobody could program by hand.
In 2012 they proved the point with an experiment that became famous. They wired together a large neural network, pointed it at ten million still frames pulled from YouTube videos, and gave it no labels — no one told it what anything was. Left to find patterns on its own, the network invented, among other concepts, a neuron that lit up for cats. A machine had learned to recognise something without being taught the word for it. Inside Google, that was a lightning bolt: the same techniques could power speech recognition, translation, and image search.
Buying the future in London
If Google Brain was the pivot's engine, its imagination came from London. In January 2014, Google acquired DeepMind, a secretive research lab founded in 2010 by the neuroscientist and former chess prodigy Demis Hassabis, along with Shane Legg and Mustafa Suleyman. The reported price — around £400 million — was startling for a company with no product and no revenue. What DeepMind had instead was a mission statement that sounded like science fiction: “solve intelligence, and then use that to solve everything else.”
The purchase paid off in public and unforgettable fashion. In March 2016, DeepMind's AlphaGo sat down in a Seoul hotel against Lee Sedol, one of the greatest players in the history of Go — a game so vast and intuitive that experts had assumed a computer champion was still a decade away. AlphaGo won the five-game match 4–1. In the second game it played a move on the nineteenth line, “Move 37,” so alien that commentators assumed it was a mistake; it turned out to be brilliant. Two hundred million people watched a machine display something that looked unnervingly like creativity.
Giving the tools away
A quieter decision may have mattered even more. On November 9, 2015, the Google Brain team open-sourced TensorFlow, the software framework it used to build and train neural networks, releasing it free to the world under the Apache 2.0 licence. It was an odd move for a company famous for guarding its crown jewels. Why hand rivals your best machine-learning tools?
The logic was strategic. If TensorFlow became the language everyone used to build AI, then every student, startup, and university lab would be training a generation of engineers fluent in Google's platform — engineers who might one day work at Google, or build on Google's cloud. It was the same instinct that had made the company open early and often: shape the ecosystem, and you shape the future. TensorFlow quickly became one of the most widely used AI frameworks on earth.
“AI first”
All of this needed a leader who believed it. On October 2, 2015, as Google reorganised itself under the new holding company Alphabet, Sundar Pichai became chief executive of Google. Soft-spoken and product-obsessed, Pichai gave the pivot its slogan. The company, he said, was moving “from mobile first to AI first” — a world where machine learning would sit underneath every product, from Search to Photos to Gmail's uncanny ability to finish your sentences.
It was more than a marketing line. Behind it, Google was designing its own AI chips, weaving neural networks into translation and voice, and treating research labs as core infrastructure rather than expensive hobbies. The search box would remain — but it was now the front door to a company that had decided its real business was building intelligence itself.
The bet would soon collide with the messy realities of the physical world, of regulators, and of employees who did not all agree on where the line should sit. Next: the company's ambitions run into the questions of ethics, power, and trust.
Comments
Post a Comment