Google is rolling out some changes to its search algorithms intended to help people in times of crisis, according to a Mashable article.
Their search engine already flags searches explicitly related to suicide or domestic violence and attempts to provide users with assistance. However, more implicit searches would often go undetected.
For example, a user considering suicide may search for a particular location where they might harm themself. Previously, Google employees needed to manually flag locations where suicides tend to occur. However, Google’s new AI systems can mark unflagged locations autonomously.
“Not all crisis language is obvious, particularly across languages and cultures,” Anne Merritt, a Google product manager who worked on the harm reduction project, said.
Similarly, long, complex searches about relationships can contain information suggesting potential abuse. Google’s new Multitask Unified Model, or MUM, is better at detecting these signs than their previous system, Bidirectional Encoder Representations from Transformers, also known as BERT.
Google claims MUM is 1,000x more powerful than BERT, which could already analyze context when determining a word’s meaning.
So, MUM really does know best.
