MUM: the new Google algorithm to (still) improve searches

If you think that Google Search can somehow reach a stalemate as soon as possible, because it is already optimal as it is, you are very wrong. During the annual conference, Google demonstrated that it can further improve what it had already been done with BERT.

The new model based on the Bidirectional EncoderRepresentation from Transformers to manage increasingly complex searches it takes the name of MUM, Short for Multitasking Unified Model.

If BERT was already incredible, according to what Nayak reported on the official blog, MUM is a thousand times better, i.e. it contains a thousand times the number of nodes, i.e. the decision points of a neural network that make the response outcome of the research. MUM is trained for remove low-quality data and focus on what really matters: the crunchy, solid, robust and reliable content.

What's so special about the MUM algorithm?

We can summarize MUM's new skills in one word: extensive experience. And yet, it would still be completely reductive in terms of his concrete abilities. MUM in fact, it does not limit itself to understanding human language, but generates it to go and guess the holes in our understanding of a topic – in order to help you fill them.

Sounds amazing, but it's even better than that. MUM makes Google searches sort of conversation with an expert. Pandu Nayak gives the example of all those searches that are carried out every day by each of us to satisfy a specific need. It's never just one: very often we look for five, six, ten similar keywords to get as many results and as much information as possible. In this way we get a clear and 360° overview of the researched topic.

A practical example of MUM

Is the answer still very technical and impractical? Let's give you the same example that Nayak did on his blog. “Imagine that you have climbed Mount Adams and have a trek to Mount Fuji planned for next year. What needs to be done differently to prepare?”

It's not a question you can ask a modern search engine, as you can easily imagine. Yet, MUM aims to give an articulated and complete answer to such questions, answering you as if it were a real authority on the subject.

At the moment, the results obtained thanks to MUM are far beyond incredible, and the Google team can't wait to exploit its potential by putting it on the road as soon as possible.

“Every time we take a leap with artificial intelligence to make the world's information more accessible, we do it responsibly. We put each Google Search improvement through a rigorous evaluation process to ensure we're delivering relevant and useful results." It continues: “We will be bringing MUM-based features and improvements to our products over the coming months and years. We are at the beginning of this journey of exploration, but this is really an important step towards the future of Google, where the search engine will be able to understand all the different ways people communicate or want to receive information.

Information in 75 different languages

Google MUM will not only be able to find as much information as possible on a given topic, but will be able to do so using even 75 different languages. In other words, if said Mt. Fuji information is only available in Japanese, you may never read it. With the arrival of MUM, however, it will be possible transfer them and use them to create relevant relationships to your search key. And so you never stop learning, even going beyond any language barrier.