MUM’s the word: What is Google MUM? We look at Google’s new tech

Picture of Google building - what is Google MUM

It’s often tempting to think about Google changes from the viewpoint of an SEO or PPC manager. But changes to Google Search impact everyone.

There are 70,000 Google Searches PER SECOND, which clocks in at around 5.4 billion searches per day. So any change to how search works, even subtle changes, can affect how a huge number of people use the internet and retrieve information.

Most people wouldn’t notice an algorithm change at all, even as it sends the SEO industry scurrying for cover. But what about a bigger change to how search actually works? And the types of information that it can retrieve?

Google, at it’s annual backslapping event Google I/O, might just have announced a change to search that 99% of people will actually notice.

So what is Google MUM?

What is Google MUM - tattoo style image
Photo by Jon Tyson on Unsplash

What is Google MUM?

MUM stands for Multitask Unified Model. This does not sound too exciting, but it could radically change how users interact with search engines.

Quite often when you search for something, particularly if it’s a bit complex, you may have to perform multiple searches to get an answer. On average Google states that it takes at least 8 queries to “complete” a complex task. 

What MUM will do, in theory, is to cut down the number of searches it takes to resolve these complex queries. 

How Google is cutting down the number of searches per complex query, is the exciting bit.

What is Google MUM - image showing an Android phone
Photo by Daniel Romero on Unsplash

How does MUM work? 

Previously Google has announced BERT (Bidirectional Encoder Representations from Transformers). BERT is a pre-training natural language processing (NLP) technique that, put simply, helped Google to better understand the whole context of queries.

MUM is built on a transformer architecture, like BERT. But according to Google it’s 1000 times more powerful.

It’s trained across 75 different languages and not only understands language, but generates it. This means MUM can develop a deeper understanding of information and knowledge than previous models Google has used.

MUM is also multimodal. At present MUM can understand information in text and images. But in the future it will expand to more modalities like audio and video.

What is Google MUM - Image showing a globe illustrating the global reach of Google MUM
Photo by Sigmund on Unsplash

MUM says “mind your language”

This increased knowledge and understanding of the world has big implications for search. It could make it much more effective.

For example, your complex query may have great answers in a language other than, say, English. MUM will be able to transfer sources across languages. So if you’re searching for specifics about semiconductor production in Japan, MUM will be able to give you information from Japanese language sources in English very quickly.

This depth of transferable knowledge could also give you great recommendations for purchases. So in your Japanese semiconductor searches you might get a recommendation for a specific product. This could have big implications for international e-commerce. In theory it could help international users find niche brands in different countries. And it can do all of these things simultaneously. Hence the “multitask” in Multitask Unified Model.

Robot image showing an "AI" at work
Photo by Possessed Photography on Unsplash

AI, Search and Bias

When big changes like this are made, Google has a responsibility to look at it’s impact, which Google says it’s going to do. They certainly have policies for personalized ads which address this. But their record is also…spotty in this area.

MUM will go through a fairly rigorous quality control process, like BERT before it. Search Quality Raters should, in theory, help to avoid potential biases in the system.

As AI becomes more complex and touches more areas of our lives, Silicon Valley (and Google in particular) has a responsibility to ensure that bias does not form part of their products. Clearly steps are being taken in this area and time will tell if enough work is being done.

What does it mean for businesses?

At the moment, it’s far too early to draw any conclusions. If you run Google Ads for your online business nothing is going to change in the short term. If you work in SEO, nothing is going to immediately change.

But it’s worth thinking about big changes like this, which often go under the radar for the general public. If your business sells very complex products, this is something you may have to learn about. We also don’t yet know how this might change user search behavior. But when behavior changes, business has to adapt.


Contact us via hello@maqe.com.

Talk to MAQE

Want some help adapting to the future of commerce? MAQE can help! We can run personalization campaigns to improve ROI and help your organization adapt to on and offline change. Talk to us via [email protected].