GOOGLE hasn't stopped tinkering with its search engine, looking to innovations and in particular an algorithm to transform searches for answers to complex questions. This new technology, which is at the moment still at the project stage, could one day spark a revolution in search.


"Google it." "Ask Google." For decades, people have been asking Google just about everything. Tomorrow's weather, cooking recipes, tips for cleaning certain surfaces, information on skin problems, existential questions and much, much more. And the tech giant can answer all these questions pretty well thanks to its huge range of data and articles. But in order to continue answering questions that are ever more complex, the Mountain View firm aims to bring innovation to its search engine so that it can answer questions that are... "fuzzy."


These vague requests for information correspond to daily necessities that have not yet been formulated into specific questions that could be entered into the search engine. Google takes the example of someone who has hiked Mount Adam and now wants to hike Mount Fuji and wants to know what they should do differently to prepare for it. Currently, answers for such an inquiry can only be obtained through a series of questions in the search engine. On average, it takes about eight searches to find a complete answer to a complex need. With MUM, short for Multitask Unified Model, Google wants to simplify the life of its users.


Enormous potential

Thanks to artificial intelligence and a technique called "transformer," the algorithm can consider words in a precise context rather than as isolated "objects" and then analyze them through its gargantuan database. The technique was first developed by Google back in 2018, however it is thanks to GPT-3, an OpenAI language model that last year was demonstrated as capable of generating blocks of coherent text, that the process has greatly advanced. For the algorithm, this amounts to creating a real virtual assistant capable of understanding the subtleties of a sentence.


Since MUM defines itself as multimodal, this new algorithm can search and interpret information present in text as well as in images. In a few years, it could even be able to understand data from video and audio content. Here again, the potential of this technology is enormous. For the time being, no date has been revealed by the American firm, the algorithm is still in the testing phase to detect potential anomalies.