10 Years Ago This Week: TensorFlow brings machine learning to the masses

To celebrate 10 years of Creator Weekly, I’m sharing tech highlights from 2015 that still resonate 10 years later. This update was for the week of November 14, 2015.

Ten years ago this week Google announced TensorFlow, an open source machine learning platform for everyone.

Watch the Short version.

Open Source Machine Learning

TensorFlow home page in November 2015

On November 9, 2015 Google released TensorFlow, their open source machine learning system.

It was faster, smarter, and more flexible than their previous systems. And it could run on a single smartphone or scale up to a whole data center.

At the time Google was already using TensorFlow for speech recognition in the Google app, search in Google Photos, Smart Reply in Inbox (which is gone, but there's Smart Reply in Gmail instead), and RankBrain in Search.

By making it Open Source, it was available to academics, engineers at other companies, and even hobbyists.

The view from 2025  

TensorFlow has been used in a number of real-world applications beyond Google. 

• Airbnb used it to classify images and detect objects in listing photos. 
• Coca-Cola used it for frictionless proof-of-purchase for its loyalty programs.
• GE Healthcare used it to help identify specific anatomy during MRIs.
• PayPal used it in their fraud detection systems.
• Spotify used it to improve their recommendation systems.
• Twitter used it to build a "Ranked Timeline" (rather than chronological) Home feed.

It is interesting to see the shift in terminology. It was all about "machine learning" in 2015, but now it's all about "AI" (which makes it sound more exciting, I guess?). 

Machine Learning on a Chip

Google also developed a Tensor Processing Unit, or TPU, a custom chip for neural network machine learning which is tailored for the TensorFlow framework. Google uses TPUs in its own data center infrastructure. 

The first TPU was announced in 2016. This year the 7th generation TPU, code name Ironwood, was released, designed for the "age of inference", which is "a move from responsive AI models that provide real-time information for people to interpret, to models that provide the proactive generation of insights and interpretation."

Use TensorFlow in your Own Machine Learning Projects

TensorFlow in 2025

Development of AI systems has exploded over the past few years, and it's not clear to me how much TensorFlow itself is currently used by Google. 

But it's still free for anyone to use for their own projects. New versions of the TensorFlow open source platform are still being released, and there are courses and a community. 

While there are open source alternatives like PyTorch (developed by Meta), Google was the first to bring machine learning tools to everyone.

References

TensorFlow.org (snapshot of the site in 2015)

Google: Our AI Journey

Google Blog, 9 November 2015, TensorFlow: Smarter machine learning, for everyone

Google Cloud Blog, 18 May 2016, Google supercharges machine learning tasks with TPU custom chip

Google for Developers, 15 February 2017, TensorFlow Version 1.0

TensorFlow Blog, 30 September 2019, TensorFlow 2.0 is now available!