Your Google Search Soon Combines Articles and Images

In May, Google supervisors unveiled new products artificial intelligence trained by the words and pictures they claim to create on the internet hunting clear. Wednesday, Google provided an overview of how the technology would change the way people search online.

Starting next year, the Multitask Unified Model, or MUM, will enable Google users to integrate text and photo search using Lens, a mobile app that is integrated into Google search and other features. For example, you can take a picture of a shirt with a lens, and then check “socks with the pattern.” Exploring the “how to adjust” on a photo of the bike section produces instructional videos or blog posts.

Google will incorporate MUM into the results of other search engines that readers can search for. For example, when you ask Google how to make a paint, MUM can explain in detail the instructions, details, or how to use the paint. Google is planning in the coming weeks to bring MUM to YouTube video search, where AI will show search ideas below videos based on video animations.

MUM is trained to make ideas for words and pictures. The inclusion of MUM in Google’s search results also represents a continuation of the use of language types that rely on a large number of text removed from the Internet by this type of language. neural networks architecture called Transformer. One of the first experiments came out in 2019, when Google introduced the BERT language into the results of modifying browser settings and summarizing the content below.

Google vice-president Pandu Nayak said BERT represents a major change in search over the past decade but that MUM takes the AI-enabled language used in Google search results to the next level.

For example, MUM uses data from 75 languages ​​instead of English only, and is taught on pictures and text instead of text alone. It is 1,000 times more than BERT when you measure the number of phases or connections between the production neurons in the deep learning system.

While Nayak calls MUM a major milestone in understanding the language, he also acknowledges that the major language groups come with the complexities and dangers that are known.

BERT and other variants from Transformer have been shown to suck bias found in the data he loved to teach them. In some cases, researchers have found that the older the language, the more likely it is that he will develop a biased and aggressive voice. People who work to identify and change racial, gender, and other language problems say that analyzing the vocabulary used to teach these races is important in reducing the risk and potential for filtering. In April, the Allen Institute for AI also said that the lists used in Google’s entertainment programs to teach T5 languages ​​could make all groups less accessible, as people known to be queer.

Last year, several AI researchers at Google, including former Ethical AI coleads Timnit Gebru and Margaret Mitchell, have said that they have faced opposition from officials in their work stating that large forms of language can harm people. Among Google’s staff, Gebru’s dismissal following a dispute over a financial report and a culture of major languages ​​led to allegations of racism, invitations to alliances, and the importance of Strong protection of whistleblower for behavioral researchers for AI.

In June, five U.S. cinemas reported a number of alphabetic changes and Gebru’s removal on suspicion of whether Google products such as search or Google’s workplace were safe for black people. Mu letter to the supervisors, the filmmakers wrote, “We are concerned that algorithms will adopt data that promotes erroneous assumptions or excludes people from viewing advertisements for housing, employment, credit, and education or simply indicating eating opportunities.”