Google Introduces MUM To Ease Complex Queries

Google logo


  • Google’s Prabhakar Raghavan announces a new technology called Multitask Unified Model (MUM) at Google I/O.
  • MUM is similar to BERT and built on the same technology, but is said to be 1,000 times more powerful. 
  • MUM can multitask and is trained in 75 languages.
  • MUM is also multimodal, meaning it understands information across text and images and, in the future, can expand to more modalities like video and audio.
  • Google is testing MUM internally. No rollout dates have been announced yet. 
  • The effect on search is likely to be massive. 

Solving Complex Queries 

One of the biggest challenges Google search has always faced is lacking a satisfying technology that provides a more precise search solution in the shortest time.

According to Google, it takes an average of eight searches to complete complex tasks. Or approximately one hour to get the answer you need on a complex topic. 

MUM aims to address this. 

What is MUM?

MUM stands for Multitask Unified Model. It’s being built to assist with complex search queries. Although similar to BERT, which Google introduced in 2018, they’re both built on the same transformer architecture; MUM is said to be over 1,000 times stronger. 

The most significant difference is MUM’s ability to multitask and understand information across different formats and languages. This enables MUM to acquire knowledge, understand the language and interpret text, images and video all at the same time. 

MUM doesn’t just interpret language; it writes it. 

As MUM understands over 75 languages and reads content across different formats – text, images, video and audio – it can pull information from literally millions of sources. Combining this information into one clear answer. Irrespective of the format or language. 

An example of how MUM would work

machine learning and Google's ai

A Google search today isn’t designed to answer a question as an expert would. You have to collect multiple pieces of information from various sites to get what you want.

Here’s an example.

You have previously gone mountain biking in Sardinia, Italy and now want to try the landscapes of Vermont.

You’ll have to search for the best clothes and gear for Vermont, the tracks and the weather patterns to expect.

But if you ask a mountain bike expert familiar with Vermont, you’ll get a clear and concise answer in a 10-minute conversation.

MUM will take the position of an expert now. It can surface vital information based on its deep knowledge of Vermont and point you to valuable articles, images, and videos.

The Power of Multitasking

The most significant characteristic of MUM is handling different tasks simultaneously. 

It’s also trained across 75 different languages and is multimodal, meaning it can understand data in different formats like pictures and videos.

What this implies is that you could take a photo of a jacket and ask, “can I use this to climb Mt. Everest?”

MUM will comprehend the picture by connecting it to the question and give you an appropriate answer. It will then direct you to a website with more equipment to use when visiting Mt. Everest.

Google Lens Technology

In addition to MUM, the company has also perfected its Google lens technology introduced in 2017.

Google lens technology

Since it was launched, Google lens has recorded 3 billion searches a month.

The technology allows you to search whatever you see using your search bar, camera, or even photos. 

Google is continually working on new ways to improve its search service. MUM will help people perform searches they previously thought were too complicated. 

The steps it has already taken seem very promising and might significantly impact the search behaviour.

In the words of Sundar Pichai, the CEO of Google and Alphabet – ‘with MUM, you could use one day plan a road trip by asking Google to “find a route with beautiful mountain views.” This is one example of how we’re making progress towards more natural and intuitive ways of interacting with search.’ 

How will MUM affect search? 

We don’t know.

We do know that search is changing and changing quickly.

As Google progresses with machine learning and this technology is being used to meet searchers’ ever-changing intentions and expectations, the search landscape will continue to evolve. And evolve at a faster pace than ever before. As we saw in our Pinball pattern blog post, the days of ten blue links are way behind us. 

BERT was rolled out at the end of 2018, immediately affecting 10% of all search queries and more since the disruption MUM could cause (being 1000 times more powerful)….is, well, massive! 

Exciting times ahead! 

About John Kramer

My name is John Kramer, and I am a Malaga-based SEO fanatic. Marketing is my passion. With over 15 years of experience in digital, I am constantly energised by the ever-changing landscape of search. A seasoned marketer with a passion for helping small and mid-sized businesses succeed in the digital age. Proud father, husband, and outdoor enthusiast.