How does Google Search deal with fake news and information fragmentation?

Google has a wide range of services, from services such as Google Maps, Gmail, and Google Workspace to devices such as Chromebooks and Pixel series, and Google Search is the starting point.

Google Search, which is indispensable for using the Internet, is steadily evolving, including support for mobile environments and the adoption of AI. What was the process like? To commemorate the 20th anniversary of the Google Japanese corporation in September 2021, we interviewed Mr. Ryuichi Imaizumi, who is in charge of the Japan office in the Google search engineer team. We talked about the past, current status, and future of Google search.

It has evolved in three directions: universal search / mobileization / AI

――First of all, I would like to look back on the history of Google search. Can you tell us what was particularly impressive of the many updates you made?

Mr. Imaizumi (henceforth, title is omitted) The starting point of Google search was a service that "when you enter a search word, the information on the web page is returned". It's been 20 years since we started the service in Japan, and Google Search is evolving in three directions.

One is "universal search," which returns various information other than websites as search results. For example, if you search for "pansy", you will also see an image of a flower. Universal Search has greatly improved the convenience of searching by incorporating various search results such as images, videos, maps, and news.

Next is the "mobile" wave. In the last 10 years or so, the use of search on mobile terminals such as smartphones has progressed worldwide, and the way search is used has changed significantly. Google keeps updating to suit your usage and strives to provide a user experience that suits your new environment.

The third is the evolution of "AI," which has been moving significantly over the last few years. A technology called AI is evolving search engines into something that is "more understandable." Today's search engines return search results when you enter a few keywords, but even with natural spoken words such as "Why is the sky blue?", It is becoming possible to read the user's intention and return the answer. , Is evolving day by day.

――When you think of changes due to mobile, I imagine that the appearance of smartphones was a big trigger, but when did Google search start to support mobile and how did it change?

Imaizumi Technology development for mobile began shortly before smartphones became the leading role. At that time, I was involved in the development of a mobile search function for feature phones, so-called garages, and the search world at that time was a "big Web" for personal computers and a "small" for feature phones such as DoCoMo's i-mode. The Web "existed respectively.

Then smartphones came out, but what was shocking was the easy access to the big web on a small screen. When I touched the early iPhones and Androids, I was surprised that "I have a desktop computer that fits in the palm of my hand." At that time, I was wondering, "If smartphones become mainstream, I wonder if mobile web searches will be unnecessary."

But it wasn't that simple. I noticed that the “behavior” of a website is different when accessed on mobile and when accessed on a desktop or laptop.

Smartphone support is a project from Tokyo

――What does it mean that “behavior” is different?

Imaizumi For example, on the website of a pizza shop, a menu page was displayed in the search results. However, when this page was opened on a smartphone, it was set to automatically transition to the top page of the mobile version, such as "m. ○○ Pizza.com". Users are taken to a home page that has nothing to do with the search results they expected.

If the same website as the computer is not displayed on the smartphone, Google search must be created separately for the computer version and the smartphone version. Therefore, around 2010-2011, the mobile search team created a "Google bot" that collects smartphone versions of web pages, and started by understanding smartphone sites.

The creation of this mobile version of Googlebot was a project led by the Google search team in Tokyo at the time.

――Do you mean that the smartphone support was a project from Tokyo?

Imaizumi Yes. There is a reason for that, partly because Japanese mobile phones were at the forefront of the world at that time. In the era of feature phones, Japan had mobile internet such as i-mode and EZweb, and it was a country with many users in the world. As a result, the mobile web market was large and there were many engineers who were familiar with it, so the Google search team in Tokyo was able to specialize in mobile support.

The second reason is that there was data that Japanese users still use smartphones at a high rate while shifting to the smartphone era. The Japanese team has taken the lead in mobile search, partly because it is easier to try new services on smartphones in such a market.

――If you develop a search engine specialized for smartphones, it seems that completely different search results will be displayed on smartphones and PCs, but what about the actual situation?

Imaizumi In the end, that didn't happen. In the past, we considered trying to separate the search results for smartphones and PCs, but now we have settled on the direction of using the same index as the appropriate method for smartphones and PCs.

The reason is that indexes of robotic search engines like Google Search require a huge amount of data. Creating and maintaining indexes for smartphones and PCs is not realistic either.

Since then, the spread of smartphones has accelerated, and the world of the Web has changed significantly. Most users who use Google Search are also accessing from mobile devices such as smartphones.

Users are always searching and now more often looking at this moment

――What kind of change has happened when the most used device has changed from a personal computer to a smartphone?

Imaizumi I think the most important feature of mobile is that it can be used at any time.

There is a "wave" in the usage pattern of any web service. For example, they are not used at night because they are all sleeping, or they are used more actively on weekdays than on weekends. In a desktop environment, the waves tend to be big.

On the other hand, mobile search is used outside and at home, Sunday and Monday. You can't use a computer unless you're sitting at your desk, but you can easily pick up your smartphone. This difference manifests itself in the way services are used.

Another thing to mention is that smartphones are often asking for information about "this moment". For example, searches that look up related information in the event of an earthquake will greatly increase the use of mobile search.

User voice and data are the starting point when thinking about new features

――In Google search, you can display information according to the search word, such as information on nearby shops and Wikipedia, in card format. Is this how to put out the information sent to how it is used on smartphones?

Imaizumi The answer is "yes". Not limited to Google search, there are two ideas that will be the starting point when Google thinks about new features. It is "user's voice" and "data".

User voice is the most important factor for Google to develop new services. Although we sometimes have discussions with members of the company, we often gain new knowledge through interviews with actual consumers and use it for service development.

Data, on the other hand, is important for investigating how new features are perceived on this service.

The actual development flow is to first create a prototype for verification, introduce it on a trial basis, and then look at the results and make trial and error.

There are many ideas that will be put into storage if you try to make it and the reaction is disappointing or it is not used more than expected. After repeated trial and error, the surviving rough diamonds will be widely deployed as new functions and services.

--Are there any new search features that are unique?

Imaizumi It's not for smartphones, but if it's been released in the last 1-2 years, the "function to find and propose the answer to the search content in the video" is unique and convenient. This feature was developed by the Google search team in Tokyo.

For example, do you want to know "how to put on the correct mask"? The new feature will display videos as search results for these questions. In addition, the proposed video will be played from the part of the tens of seconds that you really want to see, that is, the part with the mask on.

This function was realized for the first time due to the flood of huge videos on the Web and technological advances such as the ability to judge and understand the contents of the huge videos with AI. is.

Search results give as broad an opinion as possible to the question asked by the user

――While it has continued to evolve, I think there was a time when fake news became a hot topic in the last few years and information was becoming more fragmented. How do you deal with Google search?

Imaizumi I think this is a very important question. First of all, this problem should be considered in two parts.

One is simply incorrect information. Content that contains incorrect information or invites misleading. We believe that content that is false in fact should not be released.

At Google, we work with fact-checking organizations to verify what is clearly different from the facts and support activities that bring out the right information to the world.

The other type is a theme in which people have different opinions even if they see the same event. It is natural for people to have diverse opinions, and there are various factors behind them, such as political ideas and beliefs. Google does not unilaterally decide the "right person" for such opinions and backgrounds.

Google believes that we should strive to be as broad and diverse as possible in answering the questions asked by our users.

――Even if Google search gives various opinions as search results, I think that users may search for "only the information they want to know" with the words they know in the first place. It's also called a so-called filter bubble. Is it possible to take measures against such a situation?

Imaizumi To be honest, it's a very difficult problem. If a user searches only for specific questions from a particular point of view and returns an answer that he or she does not want at all, it is of no value to the user and may simply be ignored.

We believe that what Google can do is to provide an answer that includes a wider range of positions and opinions before we get caught up in a particular echo chamber. For example, an effort to interpret the intent of a question from a user's search word and propose a search word that gives a wider range of answers is a concrete example.

Or, with Google News, we try to provide topics from as broad a perspective as possible to the topics that users are interested in.

――You are trying to show that there are various options depending on the search results.

Imaizumi Well, that's what we are aiming for.

Displaying the granularity of information obtained from location information etc. is the most satisfying result.

――It's a little personal, but when using Google search, I sometimes feel that the search results are closer to the "current location". For example, even if you want to search for overseas English news, it may be difficult to reflect it in the search results, probably because you are in Japan.

From a disappointing point of view, this personalization seems to drive information fragmentation, but what do you think about region-based personalization?

Imaizumi: There are two types of personalization that you may have pointed out. One pointed out that "the search results may have changed significantly depending on the past search results and the content of the displayed website."

The other point is, "Are you personalizing with information such as your current location?"

In other words, the former is the optimization of search results based on the user's personal attributes and past usage trends, and the latter is the optimization based on more general information such as the current location.

The former can have a serious impact on the fragmentation of information. The latter personalization based on location information does mean that the information obtained differs depending on the region, but this difference also enhances the convenience of searching.

For example, if you want to visit a bank's website in the UK, even in the same English-speaking world, and if you're looking for it in the United States, the same search keyword "bank" will give you different useful search results. If you return with the idea of ​​"all the same search results", the results will be useless to anyone.

In short, balance is important, and we believe that returning the desired granularity of information from the country of the search location and the approximate location information will be the most satisfying result for the user.

However, the point that Japanese search users are not always able to return appropriate results when seeking local information in the United States says that there is room for improvement in the mechanism for returning search results as Google. thinking about. Currently, you can specify the search language, but we recognize that providing options that users who want search results can choose is always an issue to consider.

The order of search results hardly changes depending on individual behavior

――So, what about the point that "the search results may have changed significantly depending on the past search results and the content of the displayed website"?

Imaizumi Google's basic idea is to "apply only when it helps users." However, in reality, such cases are very rare.

We may create a mechanism to use the service conveniently based on the behavior of many people, not on the optimization of each individual. A typical one is "autocomplete" of Google search.

If you enter "corona" in the search bar, you will see candidates such as "number of people infected with corona" and "corona vaccine". These search candidates are selected from words that many users have searched for in the past. This is because the benefits of labor saving can be obtained by displaying the words that many people are searching for in advance.

――For example, does the order of the search results that appear at that time change depending on individual behavior?

Imaizumi Basically not. If you use the "secret mode" of your browser, you will get search results that exclude the influence of your personal behavior history, so it will be easier to understand if you compare them.

The search results of Google search may change depending on the time of day you are searching and the rough location information of the user that Google recognizes. In many cases, the impact of these changes will be greater than the changes caused by individual user behavior.

MUM is a search model that integrates all kinds of information such as languages ​​/ photos / videos.

――Finally, I would like to ask you about the new search model “MUM”. At the demonstration at the event "Google I / O" for developers, I was shocked to see that I was searching across language barriers and across content types such as images and videos. Could you introduce the features of this technology?

Imaizumi MUM is a technology that was first unveiled at the 2021 Google I / O demonstration. Abbreviation for Multi-task Unified Model, which is read as "mum".

The May 2021 Google I / O show was just an introduction to the concept, but in September Search On '21 released a more hands-on, moving demonstration.

In the Google I / O demo, I searched for the question "I have climbed Mt. Adams before and want to climb Mt. Fuji this year. Do I need to equip additional equipment?" Information such as the altitude of the mountain by the engine, images of climbers' clothes, posts on SNS, etc. are aggregated, and "Mt. Fuji in autumn is easy to rain, so it is better to prepare a rain jacket." It was the content to answer.

At Search On '21, in a demonstration with Google Lens, we showed a demonstration of copying a jacket with a flashy pattern and specifying "sneakers with a pattern similar to this jacket" to search.

The biggest feature of MUM is that it is "multimodal". Information in all languages ​​such as English and Japanese, as well as all kinds of information such as photos, videos, and maps are integrated into one huge search model. It is interesting that multimodal can display the result smoothly even if the user asks a question in text while displaying an image like a demo.

――When and how will the new functions provided by MUM be available?

In the presentation at Imaizumi Search On '21, you used the expression "I'll do my best for next year" while saying that it is still in the demonstration stage. Actually, I think that some functions will be available from that point.

――When will it be available in Japan?

Imaizumi As a matter of fact, I think that the timing of deployment will differ depending on each function and service area. The MUM search model itself is easy to make into a service regardless of the language used in each region.

On the other hand, in order to provide it as a reliable service, it is necessary to test how well it works in a practical environment. At that stage, it is often possible that for some reason it doesn't work in this language and doesn't seem useful. Of course, it may be offered immediately, but you may find that issues that were overlooked in English are being offered in Japan. So please look forward to it.

[Photo gallery] * Tap the image to view it. It cannot be viewed from some SNS.