Ben Gomes likes to connect his childhood experiences in Bengaluru with the mission he has at Google. Gomes, a Goan born in Tanzania, moved to Bengaluru when his father retired from his job at Tanzania. It was 1970. Bengaluru was still a garden city. Good weather, good schools. Gomes was two then.
He went to St Joseph’s Boys School. The teachers, he says, had a huge impact on the students. “The chemistry teacher, Mr Chatterjee, influenced a generation of students, imbued in them a love of science,” Gomes told TOI on a visit here last week. The previous day he had visited the school and interacted with students — he noted with regret that most of his teachers had passed away. Gomes was initially passionate about chemistry, but by the end of high school, he had also started learning to program. In high school, the British Council Library was his favourite haunt. Though he had to take two buses to get there.
“It was this education, this access to information, that enabled us to go to places we couldn’t have dreamt of going,” he says.
Gomes went to Case Western University, US, after school. And then did MS and PhD in computer science at the University of California in Berkeley. After a brief stint with Sun Microsystems, where he worked on the Java programming language, he joined Google in 1999, barely a few months after the company was founded. He was closely involved in the early search algorithms and, in subsequent years, was part of every fundamental change that happened in search. Last year, he was elevated to senior vice-president, with overall responsibility for search — the business that remains Google’s flagship.
With search, Gomes wants to give the same access to information to everyone that he enjoyed in Bengaluru. Even better. Gomes recollects that during his high school days, there were no books on electronics. “There was one graduate-level book, but I couldn’t understand it because I was a high school student,” he says.
But now, with the web and search, he hopes anyone can find information on any topic they are interested in and pursue that interest. Gomes is taking search into new frontiers. In its early days, it was about finding documents that contained the same words that users keyed in, though with a ranking based on what the Google algorithm figured were the most useful. But very soon, Google realised there was a deeper problem in search, which was the problem of language. For instance, many users misspelt words because they did not know how to spell them, and so did not get the results they were looking for. So Google found a way to help them correct their queries.
Ordinary users also used languages different from experts. While an expert would “replace” a tyre and “convert” a currency, a lay person typically would “change” the tyre and currency. So, Google found a way to get the right synonym based on the context — the other words in the query — to get the best results. It required understanding language, and it took years to build the system, though at the time they thought they would solve it in a week.
This was, however, still largely one word mapping to another. Google still couldn’t answer questions like ‘Who is the Prime Minister of India’ or ‘What are the ingredients in Mojito’. “We realised that if we have to begin to answer questions, we needed to have a model of the world,” says Gomes. So they set about building systems to connect just about everything in the world — people, places, things — to create what they called the knowledge graph, which became a big step-up for search.
Now, Gomes is using machine learning, AI and neural networks to make the systems understand language even more. “These are enabling us to take bigger leaps,” he says.
They are being widely used in speech recognition and, as Google goes more and more into vernacular voice search, they are becoming invaluable. They are being used in vision processing. Google is beginning to tell you that a segment of a video — from this minute to that minute — may be relevant to your query.
More than anything else, they are being used to understand what a user means, instead of going literally by the words used. When there’s a burning issue, and you search for historical information related to it, Google often fails to understand your query, and lists the “news” links because that’s what most people are looking for.
“Twenty years into the problem, there are still so many interesting problems to solve,” says Gomes. “And that’s what makes the work so exciting.”
Source: Economic Times