Google rolls out MUM to simplify, refine Search for Covid-19 vaccines

Bengaluru: Alphabet Inc.’s Google has started implementing a new search technology to surface high-quality information about Covid-19 vaccines across the world.

The search giant, at the Google I/O developer conference last month, had showcased its latest artificial intelligence model called Multitask Unified Model (MUM) that aims to simplify how people use its search engine for complex queries and research-based tasks.

The company had claimed that this model was 1,000 times more powerful than the BERT (Bidirectional Encoder Representations from Transformers) model, which is currently used to parse almost all English queries on its search engine.


The model not only understands information across multiple formats including text, images and videos but also generates information in these languages and formats. It also has the ability to do many different tasks at once.

"We trained MUM on a high-quality subset of the web corpus, where we removed low-quality content and explicit adult content. Importantly, it is trained on 75 languages at the same time,” said Google's Vice President of Search Pandu Nayak.

In essence, the long-term goal with MUM is that users should be able to ask one sophisticated query to get a comprehensive answer pieced together from various sources instead of needing to use multiple queries on different aspects of the same task, thereby resembling speaking to a human subject matter expert.

"That said, there is no intent to provide direct unattributed answers for complex questions. It is important to point our users to sources on the web, which provides an in-depth discussion of the issue so that they can decide for themselves how it is that they want to resolve the questions they have,” Nayak said.

While the model is still in development, Google on Tuesday detailed its first public application in the area of Covid-19 vaccines.

Improving Covid-19 search experience

Earlier this year, Google had expanded its Covid-19 search experience to offer a variety of authoritative information around Covid vaccines.

"As we launched this, we needed to know what queries we should trigger this special vaccine experience on, since we wanted to show them on queries that refer to Covid vaccines. But, people refer to the Covid vaccine in several ways" said Nayak.

Broadly distributed vaccines such as those made by AstraZeneca, Moderna and Pfizer have several different names all over the world, due to which it becomes critical to correctly identify vaccine names in order to provide the latest trustworthy information about the vaccine, he noted. "But identifying the different ways people refer to the vaccines all over the world is hugely time-intensive, taking hundreds of human hours.”

Last year, Google had faced a similar problem at the start of the pandemic, since the company had to figure out all the ways people refer to Covid-19 to trigger the special search experience. "That time, we had spent hundreds of hours carefully combing through all our data to identify the myriad ways people refer to Covid," Nayak said.

This year, he claimed that they were able to identify over 800 variations of vaccine names across more than 50 languages in a matter of seconds, using MUM, with an initial small sample of official vaccine names. After validating these findings, the company was able to roll out improvements to Google Search across the world in a faster manner.

"Over the next few months, we're going to see a whole series of such launches where MUM is used to improve different aspects of Google Search," Nayak said.

Last month, Nayak had noted in an official blog post that Google will carefully test these applications, specifically looking for patterns that may introduce bias into their systems.

"We’ll also apply learnings from our latest research on how to reduce the carbon footprint of training systems like MUM, to make sure Search keeps running as efficiently as possible,” he had said.