Google sharpens focus on AI for search
Investments in machine learning and artificial intelligence will result in more relevant results, says Google Search chief John Giannandrea
Mumbai: Typing a query in an online search box is straightforward for users. It’s not so for search engines that have to crawl trillions of pages, track links on them, sort them by content, then index the pages and also have their algorithms understand what the queries mean before dishing out the answers—all in less than a second.
More so, for a company like Google, which processes billions of searches daily—making search “core” to the company’s mission of organizing “the world’s information” and making it “universally accessible and useful”.
When Google was founded in September 1998, it was serving around 10,000 search queries per day. The company now processes more than 40,000 every second on average, which translates to over 3.5 billion searches per day and 1.2 trillion per year worldwide, according to internetstatslive.com.
More searches also translate into more money for Google, which dominates the world search ad market. Google will remain the dominant player in worldwide search ad spending. The company is estimated to capture $47.57 billion in search ad revenues in 2016, or 55.2% of the search ad market worldwide, according to a 26 July report by eMarketer.
To make its searches more effective and relevant for individuals and companies, Google is now sharpening its focus on machine learning algorithms and artificial intelligence (AI), according to John Giannandrea, who took over as senior vice-president of Search this February after Amit Singhal retired.
Giannandrea, who has been with Google since 2010 after it acquired Metaweb Technologies Inc., where he was co-founder and chief technology officer, now also leads Google’s Computer Science Research and Machine Intelligence groups.
In a phone interview, Giannandrea said that Google’s investments into machine learning and AI research “have made big advances in recent years and tied directly to progress in areas like voice and image recognition, natural language processing and translation”.
AI technologies, which include machine learning and deep learning, is core to many teams at Google—from the self-driving car to the search results page. While machine learning is broadly defined as the ability of a machine to teach itself from mountains of data without the need for programming, deep learning—which uses artificial neural networks that are loosely modelled on the human brain—can be said to be a subset of machine learning.
“Machine learning improves the products we offer everyone, and makes possible what was impossible just a few years ago. You can talk to the Google app on your smartphone. Thanks to deep learning, you can improve speech recognition—approximately 25%, and natural language processing (NLP) helps understand what you mean. With neural networks, Gmail now blocks over 99.9% of spam, including some never-before-seen spam,” Giannandrea pointed out.
Google Search has undoubtedly evolved over the past 18 years of its existence. It has come a long way from PageRank—a system for ranking web pages developed by Google founders Larry Page and Sergey Brin at Stanford University to sophisticated algorithms like Panda, Penguin and Hummingbird (2013). Moreover, Giannandrea’s own role at Metaweb Technologies is said to be the basis of Google’s Knowledge Graph, which was launched in May 2012. If you search for ‘Narendra Modi’, Google search will throw up a box of highlights about the prime minister to the right of the links—the box is powered by Google’s Knowledge Graph.
Search has become “a good bit smarter, no longer relying on just matching keywords”, acknowledged Giannandrea. “Now that you can use voice search to ask questions of the Google app in natural language, it’s much easier to get what you need at any given moment. That wasn’t possible before Google made some big breakthroughs in machine learning over the past few years—both in recognizing speech more accurately in more languages, and in understanding the meaning of words.”
Google Search today works with 159 languages, and Voice Search is now operating in 58 of those. In 2015, Google rolled out RankBrain—a machine-learning algorithm. RankBrain, according to Giannandrea, has “helped Google understand long or complex questions better. And the Knowledge Graph ties it to real-world people, places, and things”.
Google uses the RankBrain algorithm—a ranking signal that uses deep learning to improve results. Google Photos lets you search for anything from “hugs” to “dogs” because the system uses Google’s latest image recognition system. And Google Translate app can help users communicate in 103 languages through machine translation.
RankBrain, according to Giannandrea, is one of over 200 signals that Google uses to improve ranking in Search. “It uses an approach called word embedding where a deep neural net learns how words relate in a very high-dimensional space, and then uses that to predict what search results might be best for a given query,” he said, adding that the process is “particularly effective for long-tail queries—searches that are long, or rare, or haven’t been seen before”.
Commenting on whether RankBrain will impact search engine optimization (SEO), Eric Enge, CEO and founder of Stone Temple Consulting, wrote in a 9 March report that “Truth be told, at the moment, there is not much impact at all. RankBrain will simply do a better job of matching user queries with your web pages, so you’d arguably be less dependent on having all the words from the user query on your page.” RankBrain, according to Enge, results in “...an increase in overall search quality”, and “an increase in Google’s confidence that they can use machine-learning algorithms within the core search algo, which has already likely led to more such projects being launched”.
To be sure, other search engines including Microsoft’s Bing, China’s Baidu, Russia’s Yandex and Yahoo have similar knowledge graphs and also make heavy use of algorithms to make searches relevant. In a March 2013 official blogpost, Richard Qian, development lead for the Bing Index and Knowledge team, wrote that to make Bing “more than a collection of blue links pointing to pages around the web, ...we introduced a feature called Snapshot, which enables answers at a glance in the center column of the search results page”. The Bing team christened the underlying technology for Snapshot ‘Satori’, which means understanding in Japanese.
Baidu, also known as “China’s Google” because it dominates web search in that country, has opened a new artificial-intelligence research lab in Silicon Valley that will be overseen by Andrew Ng, a Stanford professor who played a key role at Google in deep learning.
Yet, the fact is that Google dominates both desktop and mobile search worldwide. According to Netmarketshare.com, Google leads the desktop search engine market with a 71.11% market share, followed by Bing (10.65%) and Baidu (8.73%). On mobiles (including tablets), Google has a 95% market share.
This May, during its annual developer (I/O) conference, the firm introduced Google Assistant—an intelligent personal assistant that is considered an extension of Google Now (which has morphed into ‘Now on tap’). Google Assistant can engage in two-way conversations. It is integrated in the Allo app—launched this month to rival WhatsApp, Facebook’s Messenger and Apple’s iMessage.
“As we develop the Google Assistant, we are using all our experience building Search, and extending it to all kinds of other things you need to get done in your daily life. We’re still early on in this journey,” Giannandrea said.
Editor's Picks »
- Same-store sales growth trips at Future Retail
- Cipla Q4 FY18 results no reason to reverse stock underperformance
- Dr Reddy’s Q4: It’s a wait and watch, share price spike notwithstanding
- What SBI Q4 results say about the Indian economy and the bank
- Patanjali’s slowing growth does not mean that Colgate’s is accelerating