In May 2021, during the Google I / O 2021 event, the vice-president for Google search, Pandu Nayak, announced the launch of “MUM” (Multitask Unified Model), a new powerful algorithm. filled with artificial intelligence.
Historically, Google Search teams have tried to solve countless challenges to make Google Search work as well as possible for users. The goal of the world’s largest search engine with this announcement is to share how Google search is viewed and approached. Google engineers know that most Internet users often have to type a large number of queries when searching the web, and sometimes spend a long time getting the answer they need.
To illustrate this problem, Pandu Nayak gives a concrete example. Imagine you hiked Mount Adams last year. Next fall, you want to hike Mount Fuji, and you want to know what makes the two places different to better prepare yourself. Today, as we speak, Google could help you, but that would require a lot of thoughtful research: indeed you will have to Finland WhatsApp Number List research the elevation of each mountain, information on temperatures in the fall, the level of difficulty of the trails. hiking, etc. It is only after a certain amount of research that you will be able to get the answer you need.
Whereas if you could talk to a hiking expert, you could certainly ask them directly the following question: “What else do I need to do to prepare myself?” “. And you’ll get a thoughtful answer that takes into account all the nuances of the job at hand and walks you through the many things to consider.
However, Google Search teams have found that people make an average of eight queries for complex tasks like this one.
Current search engine algorithms are not sophisticated enough to meet user expectations like an expert might. But with the new Google technology called “Multitask Unified Model”, or MUM, the American search engine should greatly improve its results in terms of complex queries.
MUM has the potential
MUM has the potential to transform the way Google helps you with complex tasks. Like BERT, the algorithm deployed by Google in 2019, MUM is built on what is called in computer jargon a “Transformer architecture”, but it is 1000 times more powerful! MUM not only understands language, but is able to generate it as well. Pandu Nayak clarified that internal pilots have already been launched with MUM and that he is excited about the potential for improvement of Google products.
This algorithm has been designed in 75 different languages and to handle several different tasks at the same time, allowing it to develop a more complete understanding of information and knowledge of the world than previous models. In addition, MUM is multimodal, which means that it understands information from texts and images and it is expected that it can extend its capabilities to other modalities like video and audio.
If we go back to the hiking example, MUM will basically understand that you are comparing two mountains, and give you relevant information about the elevation and the trail. He might also understand that, in the context of hiking, getting ready might include information about physical training and proper equipment.
Remove language barriers
Language can be a significant barrier to accessing information. But the new MUM algorithm has the potential to break these shackles by transferring data and knowledge between languages. It can learn from sources that are not written in the language of your research and still help you obtain that information.
For example, we can legitimately consider that there is a lot of very useful and relevant information about Mount Fuji written in the Japanese language, which you will not find if you do not do your research in Japanese. With Google MUM, it will be possible to transfer knowledge from sources emanating from several languages and use this information to find the most relevant results in your preferred language. So in the future, when you search for information about hiking Mount Fuji, you may get additional results, about the best views, onsens and restaurants available in the area, etc.
Apply advanced Artificial Intelligence (AI) to research responsibly
Like every time Google takes a leap forward with AI to make information search more accessible on the web, the American company tries to do so in a responsible manner. This is why Pandu Nayak also indicates in its press release that “MUM will be subjected to a rigorous evaluation process in order to guarantee the best possible results”. To do this, Google has assessment teams, who will need to follow a set of guidelines called the Search Quality Rater Guidelines, which will help the search engine understand how the results help people find results. relevant and useful information.
MUM will therefore undergo the same testing process as BERT. Nayak says it will be more specific to look for models that can identify anomalies that could occur with machine learning, to avoid introducing biased data into the systems of the algorithm. Google also wants to take into account the lessons of the latest research of its engineers on how to reduce the carbon footprint of running-in algorithms such as MUM to ensure that the Search system continues to operate as environmentally friendly as possible.