The latest news in machine learning is all about hardware. New chip designs and additional GPUs are optimized for resource-intensive machine learning algorithms. Of course software hasn’t disappeared, but it seems almost secondary to specialized devices built exclusively for machine learning. The problem is this trend focuses on a select number of uses that aren’t solving industry’s most pressing problems.
Is Hardware Eating Machine Learning?
For years now software has been consuming physical devices. Open source software, commodity hardware, and cloud computing have made it increasingly cheaper and easier to analyze data. But now that many of the largest companies have coalesced around a number of machine learning techniques, researchers seem to be going back to hardware. They are trying to squeeze everything they can out of those techniques by increasing their computing power.
Where the trend has been hardware running more specialized software, companies like Nvidia and Intel are pushing back. These manufacturers are investing heavily in devices designed for machine learning. IBM has even gone so far as to redesign the silicon chip itself. They’ve built one from the ground up around existing neural networks software. Now Google (Alphabet) has just revealed the hardware component optimized for their own machine learning software.
Function over Fashion
New machine learning innovations are exciting, but they focuses on the needs of a small number of companies. Machine learning on a chip may be ideally suited for self-driving cars and image classifiers inside data centers, but most companies aren’t looking for highly specialized solutions. The average data worker isn’t trying to “process all the text stored in (Google) StreetView images…in just five days.” Instead they want to know what’s in the data they already have. Or how that data will impact their business. And what they can do about it.
In spite of all the attention on a few very specialized use cases, it’s helpful to remember the simple things like usability, accessibility, and ROI.