Google’s long-time SVP of search, Amit Singhal, is retiring after 15 years with the company.
As it turns out even genius computer nerds can write eloquently. ("From a little boy growing up in the Himalayas dreaming of the Star Trek computer, to an immigrant who came to the United States with two suitcases and not much else, to the person responsible for Search at Google, every turn has enriched me and made me a better person.")
As his post explains, Singhal plans to spend this next phase of his life on charitable endeavors. As for Google, its choice for Singhal's replacement is quite telling: it has tapped John Giannandrea, director of engineering and the head of Google's artificial intelligence and machine learning efforts.
It is telling because Google would normally have been expected to select someone from Singhal's team to step into his shoes. By putting Giannandrea in charge of the company's core business, its bread and butter, its whole raison d’etra at one point, that speaks volumes not only about Google's view of the future of search but also the future of Google.
Indeed, not only has Giannandrea been put in charge of search, but Google is also merging the two divisions.
The Mysterious RankBrain
The most noteworthy achievement on Giannandrea's Google CV is RankBrain, a machine-learning artificial intelligence system that processes Google’s search results. Google tipped its hand about RankBrain in October 2015, when it gave the story to Bloomberg. In it, Greg Corrado, a senior research scientist with the company, explained to the publication that for the past few months a "very large fraction" of Google's search queries had been interpreted by RankBrain.
Specifically, RankBrain is a neural network — a highly-advanced form of AI in which the programs and data structures more or less mimic the human brain. Neural networks are able to apply what is known about relationships between data or objects to new data or objects. A very simple, basic example is a child able to recognize that animal is a dog even though she has never seen that particular breed before.
RankBrain is not the only signal that Google uses for its vaunted search engine algorithm — it has said in the past that is has about 200 major ranking signals and a whopping 10,000 subsignals. But RankBrain is among the most important. According to Corrado, in October it was the third-most important signal.
And Then There Was Knowledge Graph
What fewer people may remember is that Giannandrea also helped spearhead the development of Google's Knowledge Graph when he arrived at Google, via its acquisition of Metaweb in 2010.
Despite this, per the Twitterati, there is talk that Giannandrea may be emphasizing Knowledge Graph less.
.@srchvrs actually, amit was a partisan of the knowledge graph, while john giannandrea is shifting search toward ML/DL. it's a trend.— deeplearning4j (@deeplearning4j) February 3, 2016
Another clue about Giannandrea's current mindset comes from an interview he gave to Fortune last October. He said that the "holy grail" of machine learning is not self-driving cars and their pedestrian-detection abilities but rather is language understanding and summary.
But that cannot be achieved until a better computer that truly is able to mimic the human mind is developed, he said. Apparently RankBrain doesn't make the cut, at least not yet.