A new blog post from Pandu Nayak, Google Fellow and Vice President of Search, explains how Google Search uses MUM and BERT to deliver safer search results.
- Google uses MUM to better detect when a query indicates a searcher is in crisis and will start rolling out these improvements within weeks.
- Google uses BERT to better understand when a searcher is searching for explicit content.
- Using BERT in this way has reduced researchers’ “unexpected shock hits” by 30% in the past year, according to Google.
Google uses MUM to better serve researchers in personal crises
“…People in personal crises search in all kinds of ways, and it’s not always clear to us that they need. And if we can’t accurately recognize that, we can’t program our systems to show you the most useful search results,” Nayak wrote.
Using machine learning to improve its understanding of language helps Google accurately detect when search results should include the phone numbers of relevant crisis lines, for example.
“MUM can better understand the intent of people’s questions to detect when a person is in need,” Nayak explained, adding that this helps Google “more reliably show reliable, timely, actionable information.”
Google plans to roll out these improvements in the coming weeks.
Google has reduced shocking search results by 30% this year
Unexpected search results are rarely a good experience – and sometimes they can be harmful and cause you anxiety.
This is why it is imperative that Google can better read each searcher’s intent so that the results they provide can match their expectations.
SafeSearch mode allows searchers to filter out explicit results. However, there are occasions when this is exactly what a person might be looking for.
“BERT has improved our understanding of whether searches are really looking for explicit content, which helps us greatly reduce your chances of encountering surprising search results,” Nayak wrote.
He revealed that over the past year, using BERT in this way has reduced “unexpected and shocking” outcomes by 30%.
According to Nayak, BERT has been “particularly effective in reducing the explicit content of searches related to race, sexual orientation, and gender, which can disproportionately affect women and especially women of color.”
Google will use MUM to extend anti-spam in multiple languages
Google uses artificial intelligence to reduce spam and useless results on different sites.
And in the coming months, MUM will expand these safety measures even when it has very little training data.
This is possible because, as Nayak explained, “When we train the MUM model to perform a task—such as classifying the nature of a query—it learns to do so in all the languages it knows.”
Google assured the researchers that these latest changes have been and will continue to be rigorously tested, including being evaluated by manual search evaluators.
Google promoted the post via the @SearchLiaison account:
Learn more about how AI systems like BERT and MUM will help Google Search better provide people with personalized crisis support information, as well as reduce the chances of unexpected explicit content or spam in our results: https://t.co/TMOIgKvx9e
– Google SearchLiaison (searchliaison) March 30, 2022
Image source: Shutterstock / metamorworks