Greg Corrado, a senior research scientist at Google, informed Bloomberg of the workings of the system this week. It follows several large investments in the development of machine learning and AI, indicating the company is taking the potential of the emerging technologies seriously and is actively deploying it across its core businesses.
Corrado told Bloomberg that RankBrain has assisted with the processing of a “very large fraction” of the millions of queries submitted to Google in the past few months. The AI primarily helps with interpreting the ambiguous searches that Google has never seen before. Questions like “what’s the title of the consumer at the highest level of a food chain?” account for 15 percent of the total queries made each day, a surprisingly large proportion that Google has traditionally struggled to handle correctly.
RankBrain has only been enabled on the public site for a few months but it has already become the third-most important system used to provide the results for queries. Corrado said its success was a surprise, telling Bloomberg “I would describe this as having gone better than we would have expected.”
The AI links the written language that users send to Google with computer-friendly mathematical expressions and models that the search engine’s servers can actually understand. By embedding the words into mathematical vectors, RankBrain can look for words and phrases it has never seen before. From there, it is able to determine the phrases that most closely resemble the unknowns and filter the search results accordingly. The process makes the results for first-time searches more accurate than would otherwise be possible if relying on traditional information retrieval alone.
RankBrain still sits alongside hundreds of other systems that all influence the final algorithm responsible for displaying the listings on Google’s search results pages. However, its success since launch has demonstrated that machine learning may be the best way forward in search, something that Google CEO Sundar Pichai has already expressed.
During the company’s earnings call last week, Pichai described AI as “transformative” and something capable of “rethinking everything we are doing.” The technology is also used by rivals including Microsoft whose Bing search results are influenced by intelligent systems powered by the company’s Azure cloud servers.
Corrado stressed that machine learning isn’t an easy way to perfect results though, noting that it still has to be tailored to each individual use case. He said: “Machine learning isn’t just a magic syrup that you pour onto a problem and it makes it better. It took a lot of thought and care in order to build something that we really thought was worth doing.”
Google is continuing to monitor RankBrain as it expands but the system has already won over its engineers and its users. In an experiment, the company found disabling the AI “would be as damaging to users as forgetting to serve half the pages on Wikipedia” after establishing that RankBrain could guess which pages Google would return in search results with an 80 percent success rate. The humans only managed 70 percent.
