Google deploys new AI tool to COVID-19 vaccine searches

Google’s answer to combatting COVID vaccine misinformation has been to display resources from health authorities like the CDC in information boxes on search result pages.

But the coronavirus — and COVID-related scams and fake news— have blown through international borders and language barriers.

How is Google supposed to serve up that same quality of information when people are using search terms in foreign languages that the algorithm might not recognize?

Enter: MUM. No, not someone’s British mother. It’s Google’s new AI tool, called Multitask Unified Model, that the company says will help the search engine answer complicated queries by (among other things) pulling information from sources in “75+” languages.

First announced at the company’s I/O developer conference in May, Google shared Tuesday that it has put MUM into action for its first job: Surfacing information about the coronavirus vaccine.

MUM has some information to share.
Credit: Google

Google says its analysis shows that there are more than 800 variations of names for the coronavirus vaccine — like “Coronavaccin Pfizer” and “CoVaccine” — in different languages. Identifying the names, and assigning information boxes to them, is a process it says would have normally taken “weeks.” However, MUM was able to do it in “seconds.” Google validated that MUM’s analysis of search terms was accurate, and the technology is being used for searches now.

“This first application of MUM helped us get critical information to users around the world in a timely manner,” Google’s blog post on the topic reads.

Google displays information from the CDC or the World Health Organization in its boxes. Google says it will also display information from “local health authorities, depending on where you’re searching from.”

Disseminating vaccine information is a fairly small-scale example of some potentially big changes under the Google Search hood. MUM needs fewer data inputs to generate answers, so Google Search will theoretically be able to adapt to new trends and information more quickly.

On the user end, MUM will be able to take context from pages in multiple languages to suggest more relevant search results in the user’s language. Google gives the example of someone visiting Mt. Fuji. It might use information from Japanese websites to provide a traveler better results in English.

That’s not all MUM can do. Google explains it will eventually let people ask Google Search increasingly complicated questions. It can also process questions in multimedia formats; for example, it could answer a question, posed by voice, about the contents of an image. You can read a more detailed explanation of how MUM works here. And don’t be surprised if you see more from MUM, soon.