From self-driving cars to personal assistants, Google has found the use of machine learning in everything. They have now implemented a new machine learning-based system that will further improve Gmail’s spam classification capabilities.
Google has leveraged the power of its in-house machine learning framework, Tensorflow, to train new spam filters. With these new filters, Google has been able to block an extra 100 million spam emails. The figure might sound huge, but given that Gmail has over a billion users, the number looks believable.
Google was already blocking 99.9 percent of the spam emails. When considering how many spam emails are generated every day, and the number of users Gmail has, the last 0.01 percent translates to a substantial amount.
Talking to The Verge, Neil Kumaran, product manager of Counter Abuse Technology at Google said, “At the scale, we’re operating at, an additional 100 million is not easy to come by.”
It must be noted that it gets increasingly difficult to track spam messages as Google nears 100 percent of blocking spam emails. This is due to the fact that there are spam messages that have features that might not be detected by filters.
“Getting the last bit of incremental spam is increasingly hard, [but] TensorFlow has been great for closing that gap,” said Kumaran.
The traditional way of classifying an email involves rule-based filters. Here a set of rules are applied on messages which give a message score. The score is used to determine if the message is spam or not. Machine learning models try to detect more complex patterns where rule-based systems might fail. The idea is to train the network of a huge data that has messages classified as spam and not spam. Google uses Tensorflow to realize its machine learning algorithms.
AI is expected to replace the majority of the human-centered jobs in 2030 and we are already seeing the impact of AI and Machine Learning in systems around us. Unlike humans, computer programs neither need an off time nor do they get tired, making it difficult for humans to compete against them. A better way to approach the scenario would be looking at AI as complementing humans skills.