February 9, 2016
Google just indicated one of its future initiatives when Amit Singhal, who oversees the Google search engine, stepped down, and was replaced by John Giannandrea, who oversees Google’s work in artificial intelligence and, by association, what’s called “deep learning.” Google has already used deep learning to reinvent Search, via RankBrain, a deep learning system to generate responses to search inquiries. As of October of last year, RankBrain has grown to handle a “large fraction” of the queries to the search engine.
According to Wired, Singhal, who approved the introduction of RankBrian, grew to resist the use of machine learning inside Google Search, a concern echoed by others in the industry. That concern, says Wired, is because “it was more difficult to understand why neural nets behaved the way it did, and more difficult to tweak their behavior.”
But bringing on Giannandrea is a clear sign that Google believes artificial intelligence is the future of Search and other aspects of the company.
Although scientists don’t know the details of why AI works, it can, in some cases, “handle queries better than algorithmic rules hand-coded by human engineers.” Google ran a test that matched its search engineers against RankBrain, by looking at various Web pages and predicting which would rank highest on a Google search results page. RankBrain was right 80 percent of the time, and the engineers only 70 percent of the time.
Wired wonders what the impact will be in the future on, for example, the current case Google faces in a European antitrust investigation as to whether it unfairly demoted the pages of competitors. “What happens when it’s really the machines making these decisions, and their rationale is indecipherable?”
Although deep learning may “sacrifice some control,” notes Wired, “Google believes, the benefits outweigh that sacrifice.”
“By building learning systems, we don’t have to write these rules anymore,” said Giannandrea to reporters at Google headquarters this fall. “Increasingly, we’re discovering that if we can learn things rather than writing code, we can scale these things much better.”