Machine learning is the nerdiest side of artificial intelligence. It’s the part you don’t see as much in the science fiction movies or read about as much in the tech press.
However, machine learning is critical to the actual development of artificial intelligence, and it has been increasingly used in large scale projects in recent years. Most notably, machine learning is used in products like Google Photos, where its implementation centers largely around the identification of objects.
But now Google is making machine learning much easier to grasp for the wider geek community who may be interested. Google’s new offering, TensorFlow, can scale from a smartphone up to a
massive data center cluster. However, the current publicly available release actually can only run on one machine, but the design is there for major scaling. TensorFlow is open source, using the Apache license, which means other major companies will be able to fork their own versions. Smaller outfits can also make use of this ability, of course, shedding parts that may be unnecessary and building in new modules that may be required.
The TensorFlow website features two introductions to the classic machine learning problem, a MNIST data set. The beginner level introduction calls MNIST, “MNIST is a simple computer vision dataset. It consists of images of handwritten digits.”
Machine learning is still very much a nascent technology. Many projects who employ it still have to use human labor to verify the results. This can be witnessed most any day of the month on Amazon’s Mechanical Turk website, which is a market place for people to do small tasks for small rewards. Firms routinely use the labor via Mechanical Turk to verify machine learning data results, or sometimes in place of machine learning, even.
TensorFlow is written in both Python and C++. Likely enough, one of the earliest forks of the project will be into Google’s own Go language, which is built on similar principals as Python. However, using Python makes at least a good portion of the program widely accessible to the development community, among which Python enjoys serious popularity. Python is considered a “high level” language, meaning that it takes less work for the human coder to communicate to the computer what he wants done, whereas a “closer to the metal” language means the coder has to be more instructive to the code in order to achieve similar results. Both types of languages are important in computer programming, and always will be. Without underlying “close to the metal” libraries for Python to interact with, it wouldn’t be nearly as efficient. The point is that by using Python, Google is making the software more attractive to more developers around the globe.
Google may have a new business opportunity on its hands, at any rate. If it could make machine learning and its usage as easy to understand and manipulate as it has with AdSense and similar business-oriented software, it could see a new revenue stream by handling the machine learning of other outfits, delivering more bang for the buck and less overhead for smaller fees.
Images from Shutterstock and Google.