everything you need to know about this technology
Since this year we are one MUM richer. MUM is Google’s Multitask Unified Model, which it proudly presented in June. As a user you don’t do anything directly with MUM, but you do a lot indirectly. This is roughly what that model entails.
In this article we explain what we know about the Multitask Unified Model. Of course, not everything is known about the technical operation of the model, as it belongs to the algorithm of one of the world’s largest companies. Obviously, information about the technical operation of the model is scarce.
Photos, videos and text
On the other hand, it is also more interesting to know exactly what it means to you. MUM is intended to provide search results that transcend language and format. This means that it can translate like no other, but also that it can process videos, photos and text. The number of languages that MUM can currently handle is 75.
MUM is a savior, but also a killer. She puts an end to BERT. BERT stands for Bidirectional Encoder Representations from Transformers. No Bumblebee is involved here, but a system that combines the logic behind words and the links that come with it. BERT was already an asset to Google, but MUM completely drowns him out.
The artificial intelligence has been significantly improved and as a result online searches are processed even better. It almost gives the search engine its own form of intuition: from a few words it can make links with what you actually mean. Squares will also appear in which Google (but actually MUM) will give you several options. So if you are looking for earplugs, for example, MUM may indicate in-ears, headphones, wired, and so on.
A little more context
It already searches for the context for you and therefore the depth, so that you no longer have to start multiple searches yourself. Hugely valuable. It also does this not only with text, but also with images, videos and even podcasts. Something that was not possible before.
There are still caveats, because where artificial intelligence is used, ethical issues always follow. How will Google ensure that MUM remains as objective as possible? This question is not an affectation: Amazon’s recruitment bot was only inviting men at one point, and Twitter’s crop software always showed the white person instead of the black person in the photo.
Google is aware of this and has implemented several factors to counteract this. For example, human feedback is used. Users can measure the information along the Search Quality Rater Guidelines, a kind of yardstick. It is also the intention that MUM will play a major role in research into how CO2 emissions can be reduced. Large neural networks, such as Google Search, contribute enormously to those emissions and MUM must continuously look for ways to do something about it.
How does MUM work?
The question is, of course, how exactly MUM works. First, it transcends the silo thinking that Google once started. In addition, language barriers are removed. Everything that users key in is viewed from all sides, according to Google. In this way, it hopes to know better and better what people are looking for and comes up with the desired search result faster.
The new model will probably also affect SEO: Search Engine Optimization. That is good to know, because many companies use this to ensure that their website stands above that of the competition. There are rules against which you can measure whether something is SEO-proof, but those rules could just change with MUM. There is nothing to say about that at the moment.
In any case, MUM is shrouded in mystery and probably always will be. In any case, it is positive that more context is given to a search. Especially for people who are digitized. When we learn more about MUM, we’ll update this article.
