Does open-source point the way for artificial intelligence evolution?

By Warren Miller, contributing writer

Sony’s recent announcement that it’s made its artificial intelligence (AI) core libraries from its deep-learning tools available as open-source could be a major factor in AI evolution. AI applications have found new homes as newly developed neural network algorithms and have been used in conjunction with deep-learning programs to address previously intractable real-world problems, such as image and voice recognition, robotics, and automated translation. One of the last roadblocks to faster improvements in AI applications has been the lack of a wider and more active AI development ecosystem.

Perhaps an open-source development environment like the one Sony is launching could create a much more active ecosystem, one in which mutation and natural selection could help guide AI evolution to faster breakthroughs and broadening applicability. If an AI development ecosystem can support the speedy development of new capabilities and applications and then provide a way for them to fight it out in a competitive environment, natural selection (maybe “artificial selection” would be a better term) and evolution would assist developers in identifying the most useful AI applications. Evolution has a habit of finding approaches that traditional thinking overlooks and further accelerating possible breakthroughs.

In order for Sony’s open-source effort to support this type of ecosystem, several pieces need to fit into place. To begin with, the open-source system needs to be easy for developers to use. Sony seems to have had this in mind being that the core libraries are written in C++11 and are compatible with a variety of operating systems and hardware platforms, including Windows, Linux, and Nvidia GPU — just to name a few. Additionally, the choice of the Python programming language as the interface — one of the most popular and widely used research and do-it-yourself (DIY) languages — opens up a broad range of potential developers.

Image source: Pixabay.

Sony’s success in AI-related fields adds some credibility to their offering — an important element in a successful ecosystem. Their inroads into incorporating AI into consumer robotics, for example, shows that their AI core libraries can be successfully applied to very complex problems. In addition to traditional robotics applications, the Sony library has been used in deep-learning products such as AR Effect, SmartAR (augmented reality) for the Xperia smartphone, and the price Estimator Engine that provides accurate price estimates for real estate.

Another key piece is establishing a very broad and active development environment. Sony doesn’t seem to be there yet, but it is still early in the launch. Sony will need a much larger percentage of the developer population than the .1% of the size of the Google TensorFlow platform, as recently reported. This is an increasingly crowded field with not only Google, but IBM, Facebook, Apple, Amazon, and Salesforce, among Sony’s many competitors.

Once a broad and active development ecosystem is in place, some form of natural selection needs to be put into place. The environment needs to have some imposed pressure to evolve or be displaced. Perhaps a process something like the popular “Survivor” reality TV show would work: AI functions would face different “challenges” that test them against other contestants. The winners would advance and the losers would be held back. Only after winning a variety of challenges would a particular contestant be allowed into the next competition. Gradually, natural selection would identify the better AI implementation, which could then be added to the next core library. Maybe the ecosystem would be popular enough to create an AI reality TV series that shows the battles between rival AI functions — kind of like “Battle Bots” meets “Survivor” — and maybe the loser could even be “gonged” off the stage.