Return to Timeline of the History of Computers
2015
Google Releases TensorFlow
Makoto Koike (dates unavailable)
“Cucumbers are a big culinary deal in Japan. The amount of work that goes into growing them can be repetitive and laborious, such as the task of hand-sorting them for quality based on size, shape, color, and prickles. An embedded-systems designer who happens to be the son of a cucumber farmer (and future inheritor of the cucumber farm) had the novel idea of automating his mother’s nine-category sorting process with a sorting robot (that he designed) and some fancy machine learning (ML) algorithms. With Google’s release of its open source machine learning library, TensorFlow®, Makoto Koike was able to do just that.
TensorFlow, a deep learning neural network, evolved from Google’s DistBelief, a proprietary machine learning system that the company used for a variety of its applications. (Machine learning allows computers to find relationships and perform classifications without being explicitly programmed regarding the details.) While TensorFlow was not the first open source library for machine learning, its release was important for a few reasons. First, the code was easier to read and implement than most of the other platforms out there. Second, it used Python, an easy-to-use computer language widely taught in schools, yet powerful enough for many scientific computing and machine learning tasks. TensorFlow also had great support, documentation, and a dynamic visualization tool, and it was as practical to use for research as it was for production. It ran on a variety of hardware, from high-powered supercomputers to mobile phones. And it certainly didn’t hurt that it was a product of one of the world’s behemoth tech companies whose most valuable asset is the gasoline that fuels ML and AI—data.
These factors helped to drive TensorFlow’s popularity. The greater the number of people using it, the faster it improved, and the more areas in which it was applied. This was a good thing for the entire AI industry. Allowing code to be open source and sharing knowledge and data from disparate domains and industries is what the field needed (and still needs) to move forward. TensorFlow’s reach and usability helped democratize experimentation and deployment of AI and ML applications. Rather than being exclusive to companies and research institutions, AI and ML capabilities were now in reach of individual consumers — such as cucumber farmers.”
SEE ALSO: GNU Manifesto (1985), Computer Beats Master at Go (2016), Artificial General Intelligence (AGI) (~2050)
TensorFlow’s hallucinogenic images show the kinds of mathematical structures that neural networks construct in order to recognize and classify images.
Knight, Will. “Here’s What Developers Are Doing with Google’s AI Brain.” MIT Technology Review, December 8, 2015. https://www.technologyreview.com/s/544356/heres-what-developers-are-doing-with-googles-ai-brain.
Metz, Cade. “Google Just Open Sources TensorFlow, Its Artificial Intelligence Engine.” Wired online, November 9, 2015. https://www.wired.com/2015/11/google-open-sources-its-artificial-intelligence-engine.