Google has been using AI along with rule-based filters to detect spam for years.
There is no one definition for 'spam,' and that what looks like a spammy message to one user might be a much-awaited mail for another one, and Google is fully aware of the fact. TensorFlow is also used by companies such as Intel, Qualcomm and Airbnb.
With Gmail apparently being used by 1.5 billion users every month, the service has a lot of emails to deal with.
In a blog post today, Neil Kumaran, a product manager for Google's Counter Abuse Technology department, describes how the company's current efforts already can block more than 99.99 percent of all spam, phishing attacks and malware from reaching user's Gmail inboxes. As Google claims that Gmail already blocks 99.99% of spam, so working out that last percentage is hard.
This does not dismiss the achievement of TensorFlow, though, as the blocking of the additional nuisance emails suggests that Google's spam-blocking functionality has been enhanced through machine learning.
Google launched TensorFlow back in 2015 and it has very quickly become an incredibly important part of its AI business. When you think about it, emails have thousands of potential signals and just because a certain email meets characteristics that could categorise it as "spammy", does not mean it is actually an unwanted email. This process has been taking place for years, says Kumaran, with Gmail looking for certain signals from users about what they judge to be spam, but TensorFlow is "turning those signals into better results".
Google isn't saying whether TensorFlow will help with the accuracy of spam detection when it comes to flagging non-spam email as spam, but the personalised spam detection should help.
Google continued: "Where did we find these 100 million extra spam messages?"