Google has just announced that it's open sourcing TensorFlow under the Apache 2 license. That awfully nerdy sentence means that part of the software that Google uses to power its machine learning systems — the stuff that can translate words on a sign with your camera or learn what a cat looks like just by looking at a ton of photos — will now be free for anybody to use or alter.
Google says that TensorFlow can be used both by research institutions or by companies, and that it can work on a big, powerful computer or a tiny mobile phone. But as Wired notes, it looks like Google isn't releasing the version that works on vast arrays of computers. Google says it's releasing TensorFlow to "accelerate research on machine learning," but also points out in the video above that it's built to make the move from research "directly into products without having to rewrite code."
Google would benefit as a company if the tool it uses internally for machine learning becomes a standard in artificial intelligence research, as it would increase the pool of people improving the software it uses (and, as Bloomberg notes, act as a way to identify potential hires).
As part of the release, which Google itself will maintain, a dedicated website complete with documentation and instructions for getting started with its machine learning software. If you're interested in learning more but don't know where to start, Google has posted a "gentle introduction to multiclass classification" to try out.