“If you say ‘Google’, people think of search,” Greg Corrado said. He’s a senior research scientist at Google working in artificial intelligence, computational neuroscience, and scalable machine learning. “Search is the cornerstone.” Corrado met with the press to highlight, for the first time, the important place of Artificial Intelligtence (AI) in search. At Google they call their AI system RankBrain.

Making use of AI, RankBrain embeds vast amounts of written language into mathematical entities — called vectors — that the computer can understand. When RankBrain sees words or phrases for the first time it guesses what words or phrases it already knows that might have a similar meaning, and filters the result accordingly. This makes RankBrain extremely effective at handling never-seen- before search queries, of which there are more than a few.

Google handles millions of search queries a second, 24/7/365. Let that sink in. Of that total number of daily queries, about 15% of them are completely new to Google.

The hundreds of signals that contribute to Google’s regular search algorithm all kick in for any query, to be sure. And RankBrain is only one signal out of all those hundreds determining what results appear on a Google search page and where they are ranked. But in the months it has been deployed, according to Corrado, RankBrain has become the third-most important signal contributing to the result of a search query.

“The other signals, they’re all based on discoveries and insights that people in information retrieval have had,” Corrado said, “but there’s no learning,”.

And that’s the thing. RankBrain isn’t just precise in matching sites to queries. It also learns and improves its performance over time. Corrado says Google has used RankBrain to handle a massive search load over the past few months and has discovered that RankBrain is better at predicting top search results than Google’s own search engineers.

The Google engineers who spend their days developing the algorithms that power Google Search were asked to review selected pages and guess which ones Google’s search engine technology would rank on top. They were correct a stunning 70% of the time. Rank Brain had an 80% success rate.

RankBrain rolled out across all of Google search starting in early 2015 with the green light from Amit Singhal, Google’s senior vice president of search. The yearlong effort started with a team of about five Google engineers, including search specialist Yonghui Wu, and deep-learning expert Thomas Strohmann, before expanding to dozens of people. It took a long time to be certain the system was ranking things correctly.

“It’s very carefully monitored,” Corrado explained. Google periodically updates the system by feeding it new data to help it better reason with new concepts.