The other day I crashed in on a seminar on the topic of Strategies and tactics for protecting digital IP. It’s was very interesting. Here are a few of my take-aways:
European exception to the rule: European intellectual property law differs from most other parts in the world in that you can’t patent methods of medical treatment. Instead, you have to go for protecting equipment. It’s somewhat less complicated when you go for diagnostics, but even there you’d be wise to scope your patent application in such a way that what you’re trying to protect relates to the pre-processing before a human in the loop makes the actual decisions / diagnosis.
The most common hurdle in AI patenting: In order to be granted a patent, you have to be able to explain exactly how your invention work; to the point that someone else can replicate your results. In the case of machine learning, this is often not possible due to the black-box problem, meaning that the only way to declare how your invention works, would be to include all the data that your model was trained on. That’s why you don’t see a lot of ‘Using ML to solve problem X’ type of patents granted.
Having said that… You can protect the way you train your model. Also, interestingly, adding noise to an AI model is patentable. (Adding noise to models is sometimes done for the purpose of performance benchmarking, and is also the same method that would be used for the type of AI-watermarking that many now propose).
AI a trending topic in patenting: It’s often said that patents don’t make sense if you’re in software because the field is moving so fast. In fact however, artificial intelligence is one of the fields where we currently see the highest number of patents filed (one of the few fields that is hotter is Medtech).
Sweden punching above its weight: In 2021, WIPO placed Sweden on 12th place when ranking the most productive nations in terms of numbers of patents filed. Ericsson is doping that number, but if you filter Ericsson out and focus narrowly on the field of life science we’re still on 15th place. The same procedure for AI renders us on 20th place. That’s pretty good for a nation of a measly ten million inhabitants.
Yes, but… What are the number of patents—filed or granted—really saying? In and of itself, it’s not necessarily indicative of innovative supremacy. It *is* however a fair proxy of how adept a certain community (or nation) is at turning intellectual property (which in it self isn’t necessarily valuable) into business value. (which explains why WIPO ranks Sweden as second only to Switzerland in their Global Innovation Index).
Patents as social lubricant: When corporate players interact with startups, they often prefer if the entrepreneur have a patent portfolio; the absence of which creates a scary gray zone. Having a patent is the equivalence of clearly brandishing your weapons, or waving your flag, if you like.
Database protection is… complicated: The person or entity that collects and curates data owns it and the ownership is protected by the legal mechanism database protection. If someone else were to massage that dataset (eg. adding to it, or tweaking the metadata), then the results of that process—from a legal point of view—is a new dataset, with a new owner. (Andrew Ng recently wrote in his newsletter about a version of this problem, which arise when company A uses the output of company B’s LLM as input to train its own models.)
A patent is no seal of approval: Whether a piece of technology works or not is important from a business point of view, but not necessarily relevant in patenting. Ending with a funny anecdote: A patent engineer describes how she once received a filing from a company that wanted to protect a hair dryer which had a certain type of healing crystals embedded in it. The claim was that it’d boost your chakra while also fixing your hair. For evident reasons, she was hesitant to grant this particular patent, so asked the applicant for scientific proof. Which she promptly received, in the form of one hundred plus pages of “scientific evidence”, all of it in Russian. Consequently, she was left with no choice but to grant the patent. This is a great story, but also informative. Just because something doesn’t work, doesn’t mean it’s not patentable
If you’re interested, the previous posts At the Crossroads of Intellectual Property and Software Engineering and Just Because You Wrote It Doesn’t Mean You Own It sort of touches on the same topic.