During its second annual search event entitled Search On, Google unveiled a number of developments to search results, specifically a new artificial intelligence (AI) technology called Multitask Unified Model (MUM). MUM is 1,000 times more powerful compared to the BERT model that is currently powering Google Search. By gathering information from images and texts, MUM can answer specific questions such as "What is needed for a hike on Mt. Fuji?"
"Take the question about hiking Mt. Fuji: MUM could understand you're comparing two mountains, so elevation and trail information may be relevant. It could also understand that, in the context of hiking, to 'prepare' could include things like fitness training as well as finding the right gear," explained Pandu Nayak, Google Fellow and Vice President of Search, in a blog post.
According to Nayak, MUM can understand and generate language. It is trained across 75 different languages and can handle various tasks simultaneously, letting it produce a more extensive understanding of data and world knowledge versus previous models. Since MUM is multimodal, it can comprehend information across text and images. This capability can be extended to other modalities such as video and audio in the future.
To further explain how MUM works, Nayak referred to the Mt. Fuji example again. If a user wants to get information on what they need for the climb, then they're probably looking for nuanced data. This is where MUM can help as it can reveal insights based on its deep knowledge of the world, showing specific results suggesting that you bring a waterproof jacket if you want to trek Mt. Fuji in the fall, which is the rainy season in that area. Other subtopics will also be shared with you such as high-performing gear and training exercises that will let you prepare your body for the climb.
Considering that there is content on Mt. Fuji published in a language you might not understand, the language barrier can affect your access to information. MUM is capable of addressing this concern as it can transfer knowledge across languages, taking in information from sources that aren't written in the language you wrote your query in. By doing so, MUM can provide you with that information.
As MUM is multimodal, it has many possibilities. Google said that someday, you might be able to snap a picture of your hiking boots and ask the search engine if the footwear is suitable for a Mt. Fuji trek. This is because MUM is capable of understanding the image, your question and is able to use that data to answer your query.
Written by Sophia Lopez
#Google #GoogleMUM #GoogleMultitaskUnifiedModel #GoogleMUMMultimodal #GoogleMUMTrainedAcross75Languages #GoogleMUMCanComprehend75Languages #GoogleMUMGetInfoFromImages #GoogleMUMMultimodalPossiblities #GoogleMUM1000TimesMorePowerfulThanBERT #GoogleBERT