What started in the 1970’s with ‘expert system’ and continued in the 1990’s with IBM’s ‘Deep Blue’ defeating chess world champion Garry Kasparov, is turning into a race between industry giants to dominate artificial intelligence technology.
Based on recent announcements, it seems open source will play a critical role in this race as all major players are releasing their software to the masses. Making these projects available to individuals and organizations that wouldn’t otherwise have the resources to develop it, will inevitably stir up innovation as this technology permeate to all products and services.
Over the past decade, there’s been an enormous improvement in artificial intelligence, and it’s been spearheaded by Internet giants like Google and Facebook. Progress in fields like computer vision, natural language processing, and automated translation has been so quick that we now take for granted technologies that were simply not available only 10 or 15 years ago.
The foundational techniques that have made this possible, such as artificial neural networks, have been around for decades. However, gradual improvements in pattern-recognition algorithms, backed with the computing infrastructures and the enormous data sets that are available to companies like Google and Facebook, have given rise to the recent improvements in AI that have become known as ‘deep learning’. And now, the biggest players in the software development space are open sourcing a large part of their deep learning stacks.
In early November, Google announced that it is open sourcing TensorFlow, its artificial intelligence engine, under the Apache 2.0 license. The underlying technology is currently used in over 600 projects inside Google, and while Google has released some of its software under open source licenses before, this is the first time that it is open sourcing software that currently runs so many of its core services. According to Jeff Dean, one of Google’s key engineers, this is a “pretty big shift” for Google.
Then, on December 5th, Facebook announced that it will be open sourcing Big Sur, its hardware design used to train deep learning networks on GPU clusters. This announcement complements Google’s TensorFlow release, as well as Facebook’s Torch, its own deep learning library, earlier in the year.
Along with Google and Facebook, Microsoft and IBM have both recently open sourced their own machine learning and artificial intelligence toolkits. In fact, anybody looking for a world-class artificial intelligence library suddenly has a mass of high-quality choices. And yet, as huge as all these announcements are, they might not be the biggest or most relevant open-source AI news that’s come out in the last few weeks.
Why are all these companies going open source? “This is a way of telling manufacturers, ‘here’s what we use, here’s what we need,’” explains director of Facebook’s AI Research team Yann LeCun. Other companies have also expressed their interest in getting ‘support’ from thousands of contributions, and obviously using it to recruit the best developers.
The extraordinary $1B commitment
On December 11th, a non-profit artificial intelligence research organization called OpenAI was announced. OpenAI is headed by two tech heavyweights, Elon Musk of Tesla/SpaceX and Y Combinator’s Sam Altman. According to the announcement, OpenAI has already secured over a billion dollars in commitments from a group of tech luminaries (including Musk and Altman) as well as from Amazon Web Services.
The stated mission of OpenAI is to create AI in the best interests of all humanity and to prevent governments and large corporations from developing an AI that will ‘rule them all’. For example, one key concern is the ability to make autonomous weapons that could attack targets without human intervention – the whole “Terminator” Skynet scenario.
OpenAI will be a multi-decade project, and will release its research and any patented technologies under open-source licenses. Unlike Google and Facebook, OpenAI hasn’t developed anything yet. It might be months or years before they make their first release. However, they’ve assembled a team of top research engineers and scientists, and given the names that are involved in the project; any upcoming releases should be immensely exciting and valuable.
Large enterprises, such as banks and financial companies, will probably be the first to take advantage of the recently announced open-source AI projects. University research labs will also profit. But there are enormous opportunities for smaller organizations or even for individuals (take Mark Zuckerberg new year’s resolution of building an artificial intelligence system that can control his home as an example. He clearly states he will build upon existing technology ).
Whatever it is that you are developing, chances are that it could be made better if you slipped in a world-class open-source AI library to recognize and predict complex patterns. These AI toolkits will also make it possible to completely automate some old products and services (think of the self-driving car or smart home appliances like the Nest thermostat that use AI under the hood).
And if you are looking to get hired or even bought out by a tech giant, then learning and using these AI libraries would be a good bet. Of course, Google and Facebook open sourced their software to speed up the development of these tools. However, competition for deep learning talent is fierce, and the word is that these companies are also open sourcing their AI projects as a recruiting tool.
One thing is certain. The recent announcements will open up a huge frontier in the tech industry as more and more services begin to take advantage of open-source AI libraries. And with each new application, the underlying technology will likely improve, leading to an AI revolution that might finally fulfill the dreams of sci-fi authors.