What is the Google MUM Update?

With users now searching for more complex queries, the Google MUM update will aim to provide more relevant answers than ever before.

In 2021, we are more reliant on Google than ever before. Users no longer are looking for a simple answer. Searches such as ‘How to..’ and ‘What is…’ are evolving into more complex queries. We want to make comparisons, ask multiple questions at once and make scenario-based searches. These can leave answers open to interpretation and Google has noticed… Introducing MUM.

What is MUM?

Described as an ‘AI milestone for understanding information’. MUM stands for ‘Multitask Unified Model’ and launched in June 2021. The Google MUM is set to transform how complex search queries are answered. By not only understanding 75 languages but generating them too.

Using world knowledge, and the information provided across these 75 languages, will develop a deep understanding of topics. Whilst also being multimodal, meaning it can understand information across text and images. MUM aims to answer questions that don’t have a straightforward answer. Simply by collating all the available information, and becoming an expert and translator.

Multimodal

MUM is multimodal. Meaning it can understand the information in various formats. Including text, images, webpages, and eventually even videos and audio. It could mean searches for ‘would this dress be appropriate for a work party?’ accompanied with an image of a dress, could provide relevant results. MUM would understand that the dress was considered ‘a great option’ and then provide blogs on ‘perfect dresses for your work party.

Language

Just like when having in-person conversations, language can be a significant barrier. Which can hinder you from accessing the correct or relevant information. This is exactly the same with Google. But with MUM, language doesn’t need to be a hindering factor, as it understands 75 languages. Meaning it can collect information from sources in multiple languages, to build up a greater understanding of topics and themes. ‘It learns from sources which aren’t written in the language you searched in and helps to bring that information to you’.

For example, You search for ‘best places in Italy for a Holiday’

There is probably lots of helpful information written in Italian, but you can’t read Italian. MUM will transfer this knowledge from sources and use this to find the most relevant results in your preferred language. This will enable users to access a whole range of information, previously inaccessible. Results could include the best quiet beaches in Italy, popular local restaurants serving homemade pasta, and even what cities are quietest in June.

MUM vs BERT

MUM uses the T5 text-to-text framework and is around 1000 times more powerful than BERT. It’s basically a turbo-charged BERT. Enabling Google to further understand context, user intent, and language around search queries. Then providing more accurate search results to these complex queries, and ultimately a better user experience.

If you’d like to discuss the Google MUM update with us, get in touch at [email protected]