What is MUM?
Earlier this year Google announced it’s latest answer to search processing– the Multitask Unified Model (MUM). According to Google, this AI is 1,000 times more powerful than BERT.
But MUM is touted to be far more than a simple upgrade to BERT. Whilst BERT is primarily a language processing technology, MUM will draw together multiple facets of search in order to satisfy increasingly complex search needs.
Nor is MUM intended to act as a more advanced form of the “question-answering system” employed by Google for zero-click searches such as Instant Answer. Instead, the capabilities of MUM will enable Google to collate a range of results for a user which satisfy the most complex questions. This will reduce the number of separate searches needed in order to find the answer you are looking for. The ultimate aim is for MUM to reach across multiple languages and modals – including images and video content - to push the boundaries of search.
The way in which users interact with search will be transformed, as they rely on the search engine to collate all the elements of their query, without the need to ask multiple questions.
Let’s look at an example…
Google’s compares MUM’s results with the kind of answer you might expect to receive from an expert in a given field, drawing together knowledge and appreciation of all the elements involved. So instead of receiving a simple answer linked to the keywords, the answer will take into consideration a wider range of factors that might be relevant.
The example Google gives is that of a hiker who has climbed Mount Adams in the US and now wants to climb Mount Fuji in Japan. Their query is therefore:
“What do I need to do differently to prepare for Mount Fuji?”
To find the answer on a traditional Google search, the hiker would need to conduct various separate searches to find out the altitude, the weather, what time of year is best and what equipment is recommended for each mountain. An expert, on the other hand, would be able to anticipate the information you need and give you a fuller picture from just one question, advising on what exercise you should do to prepare, and what equipment you will need.
Evolving search terms
Google has put MUM to the test over the last 12 months in the context of the pandemic. Being such a wide-spread new concept, Covid-19 has resulted in a vast range of new words and expressions used across the Globe to search for the same answers. For vaccines alone, MUM was able to identify in a matter of seconds over 800 different terms used in search across the globe, based on only a small sample of official names for the coronavirus vaccines Pfizer, Moderna that it was given. Without the AI, identifying and classifying such terms – which span over 50 different languages - would have required hundreds of man-hours to achieve.
Access to multilingual results
Access to search results from different languages is one of the other big advantages of MUM. For example, if you were to search today in English for advice on completing tax returns in Italy, the results you would currently receive would relate to your English keywords. The majority, and arguably the most useful resources on your search topic would be written in Italian. But since these would be more difficult to find in your English language search your results would be limited. MUM will function across 75 different languages, collating information from all of them simultaneously.
Learnings the AI gains in one language are automatically transferred across other languages, reducing the amount of input needed to keep the AI updated.
Beyond just being multilingual, MUM is also designed to read and interpret images and will eventually expand into video and audio files. As online content becomes increasingly multimedia focussed, this capability will be an invaluable way to access information from a range of different sources.
Over time, this may well dramatically alter the way we conduct search queries we make, progressing beyond keywords or even voice search to including video or pictures in our searches. With MUM, Google will be able to make the leap between the content of an image or video and the intent behind the query to produce relevant results without any need for keywords.
The future of search
The announcement of MUM comes at the very early stages of its development. The plans for MUM’s future are ambitious. Far from being merely a new feature of search, MUM serves as a platform on which new features can be progressively developed over time. Teams across Google will be able to incorporate MUM into their projects to increase the speed at which they can evolve.
So the future of this new AI is difficult to predict. Much depends on how Google chooses to employ this new technology over time. But ultimately, it will open a lot of doors to progress.