Can Algorithms be patented?
Between the 1950s and 2016 approximately 340,000 patent applications were registered in the field of artificial intelligence (AI). Most of these related to machine learning. The majority of applications today relate to new developments in computer vision. The exciting news for industry is that between 2013 and 2016 the number of new applications for robotics and control procedures rose by around 55 per cent a year – equivalent to some 10,000 applications. But what can in actual fact be patented?
Share
Patent attorneys are busier than ever. "In the last 18 to 24 months, we have seen more interest from our clients," says patent attorney Thomas L. Lederer in Munich. He qualifies this by pointing out that most inquiries relate to so-called weak AI, such as pattern or image recognition. The number of applications in this area is growing. So why is this exciting? At the present time it is mainly big corporations and universities that are applying for patents in large numbers. "But applying for a patent is not the same as getting one," cautions Lederer. It takes another three to five years to learn whether an application has been successful or not. According to industry observers, the numbers are being driven up in part by the competitive pressures within the AI sector. At the same time, many SMEs don’t bother to register patents at all – "because they think AI applications are not worth protecting anyway," explains Lederer. Or because they have had bad experiences in the past. A patent has to be spelled out in sufficient detail for a technical expert to be able to replicate the design – which leaves the door wide open for competitors who deliberately ignore patent law. Many SMEs don’t want to take the risk, and prefer to host the applications on their own server – in the hope that no outsiders will take a look.
But there are national differences: "In the USA, businesses can patent so-called business methods, which is not possible here in Germany, or in Europe in general." One example would be an automated money transfer. In Europe, transferring money is not a technical problem. But it has to be solved by an invention in order for the invention itself to qualify for patent protection. And in Asia? There the picture is very mixed. "The German Patent and Trademark Office is in constant touch with the Korean, Chinese and Japanese patent offices," notes Lederer. Japanese patent law was established around the same time as its German counterpart. "There are certainly some basic similarities, but in practice the respective laws have diverged a little over time." As far as AI or computer-implemented applications are concerned, we have similar prerequisites in Japan and in Korea. But:
"In China things are more difficult."
No patents for the LSTM algorithm
But how do businesses go about protecting their AI projects? "An algorithm as such cannot be patented," explains the attorney. A real dilemma? In a way. If an algorithm is used in a system, or in an application, then it may be patentable. The qualifying criterion here is that a technical problem must be solved with a technical object. For example: if Alexa is used to play a track by Bob Dylan, then generally speaking we are not talking about solving a technical problem. If, on the other hand, somebody uses Alexa to operate a machine tool, then they have arguably solved a technical problem. "And that would make it patentable," explains Lederer. This is why there are no patents for the LSTM algorithm initially proposed by Schmidhuber and Hochreiter, while later developments, which solve a technical problem, are protected by patent.
So the technical application is patentable. But who owns the productivity gains delivered by AI? If a machine is made more productive by AI, and if a service provider is involved in the project, then he reaps the benefits. Sounds logical enough. But what happens if a machine develops a new industrial material with the aid of AI? Who owns this invention? The domain or subject matter expert (the materials scientist), or the systems expert, the data scientist? "The law on this has yet to be properly clarified," admits Lederer. That’s the theory. In practice, the patent is generally registered to the domain expert, because the actual invention is attributed to him.
In Germany, patent attorneys do not generally study law as such. In the past, aspiring attorneys studied mechanical engineering, electrical engineering, chemistry, physics or biology. But many other disciplines have been added to that list in the meantime, such as biochemistry, pharmacy, and of course, more recently, information technology. Thomas L. Lederer studied IT, and is one of around 50 patent attorneys specializing in this area in Germany. When future patent attorneys graduate, they are effectively apprenticed to an established patent attorney for three years, during which time they complete a special course of study in law for patent attorneys at the Hagen distance learning university. This gives them a basic legal foundation. "But this goes far beyond patent law as such," notes Lederer. "The course covers all those areas of law that regular attorneys have to study – apart from criminal law." There are approximately 4,000 patent attorneys now practising in Germany. "We are genuine specialists," declares Lederer with pride.
Interested in news about exhibitors, top offers and trends in the industry?
Browser Notice
Your web browser is outdated. Update your browser for more security, speed and optimal presentation of this page.
Update Browser