Andrew Heinzman
Google Search is one of the first steps in preparing for a camping trip or writing an essay. But more often than not, one Google Search will lead to another, and another … and another. Now, Google wants to use an algorithm called MUM to provide deep, expert information for complicated searches, saving you from searching multiple queries on a single subject.
Demonstrated at Google’s I/O 2021 event, MUM is inspired by real-world experts who can provide detailed, wide-ranging, and topical information on certain subjects. You only need to ask an expert on snowboarding a few questions to know what equipment you need, what time of day you should visit a slope, and important safety information. Why should Google be any different?
Google built MUM on its open-source Transformer architecture. Like the BERT model that launched in 2019, MUM is a contextual AI that can understand and generate language. It looks for keywords in searches, parsing the context of those words to determine what information you need (and how deep that information should go).
When its put into use, MUM could show up at the top of some Google Search results to answer questions like an expert. Google says that the MUM algorithm can even source images, crafting unique articles and explainers like a journalist.
Google is in the early days of testing MUM, and it may take awhile for the algorithm to really make a splash. Humans will oversee the software’s activities to make sure that it’s providing useful, accurate results for search queries, which may be a challenge given the short queries people tend to write into Google Search.
Source: Google I/O 2021, Google MUM Announcement
