Google outlines the future of its search engine


In the engine room that powers its dominant search service, Google recently opened a powerful new tool.

According to the search giant, this new technology-a large-scale artificial intelligence model called MUM-could one day turn Internet search into a more complex service, like a virtual research assistant, which is online Screen solutions to complex problems on the Internet.

But the company’s critics warn that this is accompanied by a clear risk: it will accelerate a transformation that has seen Google provide more direct answers to user queries, leading other sites to “internalize.” Search traffic and lock Internet users in the Google universe.

MUM-short for Multitasking Unified Model-is the latest in a series of behind-the-scenes upgrades to the Google search engine, which the company claims has brought about a major change in the quality of its results.

Including the “knowledge graph” introduced ten years ago, it defines the relationship between different concepts and brings a certain degree of semantic understanding to search. Recently, Google tried to improve search relevance by applying the latest deep learning techniques through a tool called RankBrain.

“We think we are at the next such important milestone,” said Pandu Nayak, a Google researcher in charge of MUM.

Google gives First glance The new technology was introduced at the annual developer conference in May, although it hardly explained how to use the system. In an interview now, Nayak said that MUM can one day deal with many of the “fuzzy information needs” that people have in daily life, but they have not yet formulated these needs as specific issues that they can study.

His example is when parents want to know how to find a school that suits their children, or when people feel the need to start a new fitness program for the first time. “They are trying to figure out, what is a good fitness habit-up to my level?” he said.

With today’s search engine, “you have to really turn it into a series of questions, you ask Google to get the information you want,” Nayak said. He suggested that in the future, the cognitive load will be borne by the machine, and the machine will bear what he calls “more complex and possibly more realistic user needs.”

He added that in the end, the application of MUM may be far beyond the search range. “We think of it as a platform,” he said.

Explanation of how MUM handles fuzzy queries © Google

MUM is the latest example of the ideas sweeping the field of natural language AI​​. It uses a technology called a converter, which enables the machine to view words in context, instead of treating them as isolated objects to be matched through a large amount of statistical analysis-this breakthrough has brought the machine “understand” leap.

The technology was first developed at Google in 2018, but its most dramatic demonstration came from last year’s GPT-3, A system developed by OpenAI, which can generate a large number of data blocks, shocked many people in the AI ​​world Coherent text.

According to Jordi Ribas, head of engineering and product for Microsoft’s Bing search engine, this has triggered “all high-tech companies racing to introduce larger models that can better represent the language.”

At the beginning of last year, when Microsoft announced its Turing language generation model, it claimed that it was the largest system in history. But GPT-3, which was launched a few months later, was ten times larger than GPT-3. Google has not announced the technical details of MUM, but said it is “1000 times more powerful” than its first experimental model, BERT, which uses a transformer.

However, even with this huge leap, Google also faces daunting challenges. Sridhar Ramaswamy, former head of Google’s advertising business and current CEO of search startup Neeva, said that for the past 15 years, search companies have been dreaming of answering complex questions, but found this It is a much more difficult problem than they expected.

“There are many changes in all the complicated things we do,” Ramaswami said. “Trying to make the software understand these changes and guide us, it turned out to be very difficult in practice.”

This first time using MUM involves behind-the-scenes search tasks, such as sorting results, categorizing information, or extracting answers from text.

The difficulty of objectively measuring the quality of search results makes it difficult to judge the impact of such efforts, and many experts question whether other new search technologies have achieved the effect of hype. Senior search analyst Greg Sterling (Greg Sterling) said that many search users will not notice how much improvement, especially product search is still very frustrating.

Search companies say that internal tests show that users prefer results from more advanced technologies. According to Ribas, the ability to extract answers from text has enabled Bing to answer 20% of queries directly.

For most people, only when technology brings more obvious changes can they feel the impact of transformers. For example, Google said that MUM’s ability to understand text and images—video and audio will be added later—may lead to new ways of searching across different types of media.

Daily newsletter

#techFT brings you news, comments and analysis on big companies, technologies and issues, which are made up of experts from all over the world. Click here Get #techFT in your inbox.

Processing the more “vague” queries that Nayak thought of would effectively cause Google to collect information from many different locations on the web to provide a more precise response to each highly specific query.

“This integrates all the activities of Google assets,” said Sara Watson, a senior analyst at market research firm Inside Intelligence. “Everything shown on the first page [of search results] It may be everything you want. “Watson added that such a system may cause strong opposition from online publishers.

Google, which has been scrutinized by regulators around the world, denies that it plans to use MUM to retain more network traffic for itself. “It will not be an answer to a question [system],” Nayak insisted. “The content on the Internet is rich enough, and it doesn’t make sense to give a short answer. “

He also denied that extracting the results of multiple searches into one result would reduce the traffic Google sends to other websites.

“The better you do at understanding user intent and showing users the information they really want, the more people will come back to search,” he said. The effect will be to “plant pie” for everyone.

Search advertising is the lifeblood of Google’s business, and similar problems may also be faced. Reducing the number of searches required to answer user questions may reduce the ad inventory Google can sell.But Watson said, “If the query can be more complex and targeted, so can advertising. This makes [ads] Higher value, and may change the pricing model. “

Google’s major search advancements over the years

© Reuters

General Search-2007

Google goes beyond showing “ten blue links” to return images and other results

Featured Fragments-2009

Short text results began to appear in the box at the top of the results page, angering some publishers

Voice Search-2011

User can talk to Google for the first time

Knowledge Graph-2012

Google has established a network of connections between different ideas, producing direct, factual answers to queries


The latest advances in artificial intelligence using neural networks to make search results more relevant


Have a deeper understanding of many search tasks and are expected to respond usefully to complex queries


Source link