Google MUM: understanding this new technology
Google is the leader in search engines thanks to its ability to constantly improve and facilitate the user experience on a daily basis. With Artificial Intelligence (AI), Google is able to respond better and better to even the most complex searches. To do this, technologies analyze a multitude of information, and Google has developed a Multitask Unified Model (MUM) that is very efficient in this area.
In this article we will explain what this new technology is and how it will make it easier for you to find information.
Google MUM: what is this new technology for?
The creation of this new technology starts from a simple observation of Google. An Internet user needs to do an average of 8 searches to perform so-called “complex” tasks such as organizing a trip or an excursion for example.
In some cases, the user wants to make a search with nuances based on his past experiences, but it is not possible to have a direct answer in these cases with the current model.
The MUM algorithm therefore aims to answer more subtle queries that require a cross-referencing of information. The goal of the search engine is to provide an answer as precise and relevant as that of an expert.
Beyond the precision of the information given for each query, Google MUM also intends to reduce the number of queries to be made for a given search.
Google MUM: how does the technology work?
To be able to respond to very subtle search intentions, the MUM algorithm is built according to a Transformer architecture. This is a learning model that allows for the processing of information. Unlike classical models, it can process data in any order.
Like its predecessor (the BERT algorithm), Google MUM works in Transformer, but it is 1000 times more powerful and multitasking. This allows it to identify and generate new information.
Google MUM can understand 75 different languages and understands many different tasks. It has a better and more complete knowledge and understanding of the data and thus the world in general. MUM can also analyze text through images, text and will be extended to videos and audios in the future, so it is multimodal. This will surely impact the SEO approach of companies.
Google MUM: what are its advantages?
There are 3 main advantages with Google MUM that have been communicated by Google:
Provide support for queries that do not give a simple answer
Thanks to this Unified Multitasking Model, the algorithm developed is much more efficient and therefore provides better solutions to the requests. Its knowledge of the world is more in-depth and it can also propose sub-topics related to the keywords searched.
Abolish language barriers
The content on Google is huge and comes from all over the world. Some information is only available in certain languages. Often English is a very important source, but in some cases, the most reliable and interesting data is only relevant in the local language.
Only MUM can transpose information across languages. It can learn from sources that are not available and/or written in the language of the query and formulate an answer from these sources. That is, it could find the sources in your language that are most relevant to the foreign language sources.
Understand different types of information
MUM can understand information from different formats such as images, videos or web pages. It will therefore be able to analyze images and queries to associate the right data between them.
Google MUM: an example of use
To understand the different improvements brought by Google MUM, nothing better than a concrete example provided by Google teams.
The starting situation is the following:
You have hiked a mountain in the United States and you plan to climb Mount Fuji next fall. You need to know what you should do differently to prepare for this new climb.
Analysis of queries by Google MUM:
When you search, MUM will scan content from around the world. With a traditional system, if there is any very interesting information, but only in Japanese, you would never know about it by searching in your language. With MUM, you could have access to sites that are in the language of your choice and that will give you more local information like the best views, souvenir stores etc.
Thanks to its multimodal feature, Google MUM could also allow you to cross-reference information and for example analyze a photo or a video of your walking shoes and answer a query like: are these shoes suitable for Mount Fuji?
It could tell you if they are or not and send you to a blog page of adapted equipment.
So the new Google technology promises many improvements to Google navigation. Beyond all these features, Google will be able to better and better understand how people communicate. There is also a will from Google to reduce its carbon footprint regarding its training systems like MUM.