How 'Stupid Patents' Are Dragging Down Artificial Intelligence and Machine Learning

Some of them are just poorly thought out. Some are not so innocently designed to corner the market in essential machine learning and AI techniques. JL

Joe Mullin reports in ars technica:

Just as the US Patent Office problematically gave out patents for computers doing simple things like counting votes or calories, the office seems prepared to give out patents on "using machine learning in obvious and expected ways." Companies like Google and Microsoft are seeking to acquire patents on "fundamental machine-learning techniques. "It's worth pointing out because of the serious problems it could create for innovation in machine learning."
Each month, the patent lawyers at the Electronic Frontier Foundation shine a spotlight on one particular patent they believe is a drag on innovation. This month, they're looking at one of the fastest-growing sectors of technology: machine learning and artificial intelligence.
EFF lawyer Daniel Nazer has picked out an artificial intelligence patent belonging to Hampton Creek, a San Francisco food-tech company that markets products under the brand name "just." US Patent No. 9,760,834 describes what the company calls its "machine-learning enabled discovery platform" and ways of discovering new ingredients.
The patent claim is on the long side, so there's a whole variety of specific things one would have to do to infringe it. But EFF's Daniel Nazer says the patent "reflects a worrying trend" because the lengthy Claim 1 amounts to doing machine learning on a particular type of application. During the prosecution process, Hampton Creek argued that its patent should be allowed, in part, because earlier techniques applied machine learning to "assay data" rather than protein fragments.Other claims borrow from well-known, pre-existing machine-learning algorithms.
"Indeed, in our opinion the patent reads like the table of contents of an Intro to AI textbook," Nazer writes. He continues:

It covers using just about every standard machine-learning technique you'd expect to learn in an Intro to AI class—including linear and nonlinear regression, k-nearest neighbor, clustering, support vector machines, principal component analysis, feature selection using lasso or elastic net, Gaussian processes, and even decision trees—but applied to the specific example of proteins and data you can measure about them. Certainly, applying these techniques to proteins may be a worthwhile and time-consuming enterprise. But that does not mean it deserves a patent.

Nazer acknowledges that Hampton Creek's patent isn't as bad as some of the other ones highlighted in the EFF Stupid Patent series, but it's worth pointing out because of the serious problems it could create for innovation in machine learning.
Just as the US Patent Office problematically gave out patents in the past for computers doing simple things like counting votes or counting calories, the office seems prepared to give out patents on "using machine learning in obvious and expected ways." Companies like Google and Microsoft are seeking to acquire, and in some cases have acquired, patents on "fundamental machine-learning techniques," Nazer writes.A Hampton Creek spokesperson declined to comment on the EFF post. A company press release published earlier this month, just after the patent issued, says the patent covers the company's "one-of-a-kind robotics, proprietary plant databases, artificial intelligence, and predictive modeling," put together in a system called Blackbird.