A new blog post by Pandu Nayak, Google Fellow and VP of Search, explains how Google Search uses MUM and BERT to provide safer search results.
emphasize:
- Google is using MUM to better detect when queries indicate that a searcher is in crisis, and will start rolling out these improvements in a few weeks.
- Google uses BERT to improve its understanding of when searchers are looking for explicit content.
- According to Google, using BERT in this way has resulted in a 30% reduction in “unexpectedly shocking results” for searchers over the past year.
Google uses MUM to better serve searchers in personal crisis
“…people in personal crisis search in a variety of ways, and it’s not always obvious to us that they have a need. If we can’t identify this accurately, we can’t code the system to show the most useful search results ,” Nayak wrote.
For example, using machine learning to improve language understanding helps Google more accurately detect when search results should include phone numbers for relevant crisis hotlines.
“MUM can better understand people’s intent to ask questions to detect when a person is in need,” Nayak explained, adding that this helps Google “more reliably present trustworthy and actionable information at the right time.” “.
Google plans to roll out these improvements in the coming weeks.
Google cut shocking search results by 30% this year
Unexpected search results are rarely a good experience – sometimes, they can be harmful and cause distress.
That’s why Google must be able to better interpret the intent of each searcher so that the results they present match their expectations.
SafeSearch Mode enables searchers to filter out explicit results. However, sometimes this is exactly what one might be looking for.
Nayak wrote: “BERT improves our understanding of whether searches are really looking for explicit content, helping us greatly reduce the chance of encountering surprising search results.”
Using BERT in this way has reduced “unexpected, shocking” results by 30 percent over the past year, he revealed.
According to Nayak, BERT “is particularly effective in reducing searches related to race, sexual orientation, and gender, which may disproportionately affect women, especially women of color.”
Google will use MUM to expand spam battle in multiple languages
Google uses artificial intelligence to reduce spam and unhelpful results across different locations.
Over the next few months, MUM will work to expand these safety measures, even with little training data.
This is possible because, as Nayak explains, “When we train a MUM model to perform a task — like classifying the nature of a query — it learns to do it in all the languages it knows. .”
Google assures searchers that these latest changes have been and will continue to be rigorously tested, including by human search raters.
Google promoted the post via its @SearchLiaison account:
Learn more about how AI systems like BERT and MUM can help and will help Google Search better provide people with personal crisis support information and further reduce the chance of unexpected explicit content or spam in our search results: https://t.co/TMOIgKvx9e
— Google Search Liaison (@searchliaison) March 30, 2022
Image credit: Shutterstock/metamorworks
