Robot wars: how AI defends Gmail from spam
In its eternal fight against spam, Google has decided to use artificial intelligence. The search giant has begun using the TensorFlow machine learning platform as a tool to optimize the identification and blocking of such messages.
Google is using the TensorFlow machine learning framework as a tool to improve the recognition and blocking of spam in its Gmail service. With the new filters in place since last month, the company claims to be blocking 100 million more spam messages per day.
With over a billion users, the figure seems rather small. However, Google often boasts of blocking 99.9 percent of all incoming spam, so this improvement could bring the results closer to perfection. “At the scale we’re operating at, an additional 100 million is not easy to come by,” said Neil Kumaran, product manager of Counter Abuse Technology at Google. “Getting the last bit of incremental spam is increasingly hard, [but] TensorFlow has been great for closing that gap.”
Gmail has been using a combination of artificial intelligence and rule-based filters for years. However, while rule-based filters can block the most obvious spam, machine learning finds new patterns on why an email should not be opened. The algorithms trained in this way analyze a large number of factors, from the format of the email in question, to the day and time it was sent. According to Kumaran, TensorFlow facilitates the analysis of that vast amount of data, while its open-source nature makes it possible to integrate research conducted by the community more quickly.
TensorFlow was launched by Google in 2015, and has become a very important part of its AI sector. The platform allows developers to create their own artificial intelligence tools for a wide range of tasks. TensorFlow fans praise its flexibility and, of course, its smooth operation with the other AI services of the technology giant.
Google also assures that the integration of TensorFlow in Gmail will allow a better customization of anti-spam filters. According to Kumaran, the process would have been going on for years, observing which signs attract users’ attention when identifying spam. However, it is with the implementation of machine learning that “those signs will lead to better results”.
What do you think? Do you use Gmail? Are you starting to notice the improvements produced by TensorFlow? Let us know in the comments.