Famously, in 1977, the European Patent Convention was launched prohibiting the patenting of software. Equally famously, ever since then, the legal system has been trying to find loopholes in this and has indeed succeeded in ensuring that software is patentable to the extent it has a “technical effect”. This is an interesting indication of how the law can sometimes misjudge the way in which technology might develop. Legend has it that, in the 1970’s, “hardware was everything” and people did not expect that software would be an important development. That turns out not to have been the most accurate prediction, and the legal and business environment has been trying to catch up ever since. Now, with the ever increasing prevalence of machine learning and artificial intelligence, various authorities are starting to look at how patenting may develop in parallel, and not before time if we are to avoid the errors of the 70’s occurring all over again. To its credit, the European Patent Office has begun a scoping exercise to understand some of the issues surrounding cloud computing and, significantly, patenting AI related inventions. This is timely as, with the application of current law, there could otherwise be problems.
Understanding the users of AI
To get a feel for the issues it may be worth taking a simplistic look at how AI and technology interact as being driven by primary, secondary and tertiary users, or originators, implementers and adopters. The originators are the creators of the AI engine itself – companies who develop neural networks, deep learning algorithms and so forth. The secondary market could be considered to be the implementers, companies who take existing AI engines and tweak them such that they work for specific areas of technology (medical, transport, data management etc.) The adopter is the customer – the end user who effectively buys in the technology to streamline their existing processes.
Patents and the users of AI
Again slightly simplistically, the patenting considerations vary significantly between the three parties. The adopter may well find it difficult to patent mere automation of their existing systems by simply plugging a neural network in as it may be difficult to identify an inventive step. On the other hand the implementers, those who modify existing AI to operate in specific technologies could well find patentability easily available, albeit narrowly to the specific field, as the modification required could well be technical in nature and hence patentable. All of this would more or less in line with existing patent practice.
However, the existing law may not necessarily provide adequate protection for the originators, the businesses developing the AI itself. Yet this is where the biggest developments are needed, and appropriate reward for the user required. The difficulty is that neural networks are effectively algorithms and another aspect of existing European patent law is that “mathematical methods” cannot be protected. This wording is once again a throwback to the 1970’s, but a pedantic application of the principle could cause problems. A simple solution may be to apply the question “who was doing the inventing”. If it is a mathematician then perhaps there is a problem, but here surely the real inventor is a computer specialist and hence a technical entity such that we can navigate our way around the patentability exclusion. This is where the EPO’s efforts to understand the technology – and the legal problems – could be key. If technology areas such as improved inference engines, generative or probabilistic models and the other paraphernalia underlying fundamental machine learning techniques is simply recognised as primarily technical then patentability will be purely on the merits of how inventive the development is, which seems right, Given that parallel areas such as image processing or encryption are already conventionally patentable, perhaps it is time to stop beating up the poor mathematician, or at least changing her job title to computer scientist.
Driving innovation underground
Reconsideration of the application of the software patentability exclusion would seem to be a very important step if we are to encourage innovators in the field of developing better AI engines. The risk otherwise is that we drive the innovation underground, developers relying instead on secrecy. If that happened we would not only move into a less regulated world, but one where the public simply didn’t benefit from the fruits of other parties’ research, contrary to the underlying principle of the patent system.
Future thinking at the EPO
It’s great news that the EPO has started looking at this so early and avoiding the mistake they made before of not fully researching how technology might develop in the future. Let’s hope that work continues in this area and as the law develops we can learn a lesson from the problems we created for ourselves in the 1970’s.