"It is important to define essential features that contribute to the technical effect of the invention at a sufficiently detailed level, even though these features, considered separately, are algorithmic in nature." – Jean-Marc Deltorn
speak to three researchers at the European Patent Office (EPO) about their advice, their pets, and approaches to researching computer-aided innovations, particularly in the area of artificial intelligence (AI), and how the EPO compares with the US patent analysis system. It was a broad and in-depth discussion with three professionals who, of course, are accustomed to this area, and who have been ready to present enthusiastic views on how applicants can and who should improve their technical info so as to maximize the likelihood of acquiring a patent.
These EPO researchers are in the order by which they spoke for the first time in the following interview:
Jean-Marc Deltorn (JMD) is a Ph.D. Deltorn has been a patent professional for EPO in the area of ICT since 2003. She works in many various areas from image analysis to sample recognition and machine studying, and at present focuses on speech recognition. He’s additionally a member of the EPO Working Group on Pc Assisted Design (CII), and has written course materials on "patentability of artificial intelligence".
Abderrahim Moumen (AM) is a grasp's degree and Ph.D. diploma in telecommunications and radar. He joined the European Patent Office in 2000 as a researcher in the ICT area. In 2017, he was appointed group leader, main a workforce of patent software researchers in the area of antennas and microwaves. He focuses on antennas (sensible antennas and antenna optimization), telecommunications and the Web of Issues (IoT).
Anton Versluis (AV) studied Applied Physics with a specialization in Artificial Intelligence. He labored on intelligent robots for individuals with disabilities at TNO in the Netherlands and joined the EPO in 2006 as a researcher on Model Identification (G06K9), a technical area of main importance to AI, comparable to self-management.
Without additional ado, right here's an interview with researchers Deltorn, Versluis and Moumen
GQ: We don't have daily to get an official ebook with researchers, but I know applicants all the time marvel what we will do to help streamline the course of, to make better purposes and make your life simpler, as a result of I know if we make your life simpler and offer you what you need, it should lead to higher and hopefully quicker responses to us. What advice do you give to the candidates who’re most necessary with the EPO?
JMD: The first thing I want to say here is that AI-related innovations are patentable in the EPO. I would advocate studying the instructions. The newest model, released in November 2018, offers an in depth overview of EPO practices and up to date elements of mathematical methods and a new part of artificial intelligence and machine studying. Based mostly on apply, in phrases of software design, the contextual presentation of the underlying ideas of the invention is all the time very helpful; for instance, by means of the prior artwork. Defining the essential features can also be essential.
Reveal, Reveal, Reveal
AM: There are two things I would really like to give advice on. The first is to clarify in the description the technical drawback to be solved by the invention and how it is solved by studying AI or machine, and then displaying how this answer differs from the prior art. This is very useful for researchers in dealing with AI patent purposes: it factors us in the very early levels of what the claimed enhancements actually consist of and allows us to assess whether there’s any sudden technical impact that would not have been foreseen prematurely from the info out there earlier than the software deadline. Additionally it is good that there’s a back-up system in the description, for instance by mentioning some concrete instances. This will increase the potential of the applicant – especially when it could show permissible modifications in the prior artwork when the software comes to substantive analysis.
AV: I would really like to build on these comments and say that it’s really essential to us that the invention be disclosed in a clear and enough manner by an professional to carry it out. This is applicable to both the essential invention and the contingency preparations. You’ve got to take into account that descriptions can’t later be changed in a means that provides to the subject. I'm positive it's the similar in the United States. That's the point we frequently come throughout in the EPO. Additionally it is proper to embrace formulation or pseudo-code [a formal but non-programming language description of what a computer program or algorithm does] to explain the invention. We now have seen that patent purposes fail because the plus and minus sign has been canceled and it’s a pity. Subsequently, it is crucial to make sure that both the disclosure and the description of the technical advantages of the feedback positions are perfectly clear
GQ: This is actually nice stuff. Let's take a remaining point because once you speak about pseudocode, do you all agree that it's a good idea to embrace it in your software?
JMD: Completely. The pseudocode is especially useful for describing AI and machine learning algorithms. Patent researchers working in AI-related fields have been educated in studying pseudo-code, and this makes understanding our algorithm traits easier than making an attempt to categorical the similar ideas in pure language.
GQ: Abderrahim, do you agree?  AM: The extra detailed the description is, the better is naturally the patent inspector. I perceive that there is a trade-off between applicants for what’s stored secret and what’s communicated. They’ve to find the proper stability there. This can be a challenge as a result of my colleagues have already mentioned that it isn’t legally potential to add a subject later. If the claims don’t seem to be new or creative at an early stage, the investigator will want to determine whether or not there’s anything left in the remaining description. If there's not a lot else there might be an end.
AV: Another facet to keep in mind is that in this space we are dealing primarily with abstract ideas. It is straightforward to overlook to say how sure elements of the invention relate to each other. If I say that the artificial neural network is educated, and I overlook to state the place it will get its info, and the data of the info, the invention has not been properly described. This additionally applies to different interconnections: they are necessary particulars.
GQ: Yeah, I couldn't accept more. One good thing with a code, pseudo code is certain, however the code explicitly, is whitening time. When coping with software particularly, I find that folks transfer extremely quick when they are in the archive, and typically they actually haven’t any invention, or typically they’ve it built, and they have to get one thing into the archive shortly for any number of reasons. I'm a fan to say, if in case you have something inbuilt, just remember to give enough so at the very least that's what you could have, is built. And leaving some important codes could be a great way to be sure to have that help in the software for every thing you need. Do you agree with that?
JMD: It all is determined by the place the invention is. It will be important to define the important features that contribute to the technical effect of the invention at a sufficiently detailed degree, regardless that these individually thought-about options are algorithmic in nature. These algorithmic options ought to then be disclosed in adequate element, for example by means of a pseudo code, to be put into apply by a talented individual.
GQ: Proper, and it makes us back to the undeniable fact that many occasions, individuals don't want to do it because they want to hold it secret and just win the function of getting the whole patent. So, it's one thing they’ve to assume prematurely, I assume, and not simply drop their toe in the water. One other factor I see and would really like to have all of your ideas on this is that when I take a look at software patents, no matter what the subject could also be, whether it's unbiased operating or machine learning or AI, there's just too little flowcharts and too little schematic drawings. I don't assume there are enough pictures in these purposes. Do you agree that you would agree or disagree and why?
JMD: I assume the graphical or written shows of the course of might be as readable to a talented individual. Stream charts might be very useful in understanding the logic of the process.
Distinguishing between CII and Software Inventions
AV: I want to touch upon the wording of your query. In the EPO, the time period "computer-implemented inventions" is most popular to "software patents" as a result of our regulation clearly states that software program as such can’t be patented.
GQ: It's really insightful and I hope we did extra of it in the US as a result of the software program setting drawback seems to be a telephone software that helps you monitor the weight you are attempting to lose, is software and has nothing really vital or progressive. And Watson, which IBM designs, can win in a human Jeopardy program, and additionally it is a software program and utterly totally different. And sure, it’s fascinating that you simply really make a difference – you simply refer to it as software in the EPO?
AV: That's right. Our tips apply to CII or computer-implemented inventions.
AM: Our coverage does not give attention to the software itself, but on what happens when it works on a machine, system or network, and a technical drawback that solves the challenge. When taking a look at AI and machine studying, algorithms are typically not used in the claimed system: In self-driving automobiles, for instance, algorithms can run in the cloud or on a remote system and talk with automobiles by way of 5G networks. Which means the claims made by the applicant should make sure that the invention is definitely included in the claims. This affects each the class of claims and their extent.
JMD: The reference to "computer-implemented invention" helps to make clear that we refer to a technical unit, i.e. one that essentially incorporates some technical element. With this specific wording, it’s easier to understand that we are referring to a technical object in the sense of the EPC
GQ: This makes lots of sense. For instance, the man who acquired the first US "software patent," Marty Getz, has written to us a number of occasions on IPWatchdog, and he keeps saying, "I stopped using the word" software "because that's when the problem actually arises. "And I have to admit that I know that it's true, but I'm unsure that I really obtained it until I heard that you simply do the difference. A light-weight bulb went on my head and I'm virtually embarrassed to admit it, but now I get what Marty is making an attempt to get me to perceive many years.
I was talking when you have been making an attempt to speak about drawings and diagrams and diagrams. Do you’ve another thoughts about it, Anton?
Don't combine and match
AV: We've already discovered lots of charts in US patent purposes. The one advice I give is that in case you are going to make backup arrangements – more phases or aspect phases or alternate phases – that you simply make it very clear, because our policies don’t permit for mixing and matching of embodiments. Typically we encounter candidates who attempt to mix the options of totally different embodiments, and sadly we can’t permit it. Concerning software program patents and computer-implemented innovations: In 1973, when the legal framework was created, the European Patent Settlement (EPC) was not so refined. Subsequently, we now have seen the improvement of the case regulation then adapt to more refined computer-based programming varieties till the first patent was granted in 1997 for what we might now describe as a computer-implemented invention. But in 1973, pc packages principally meant adding numbers and retrieving info, which was thought-about an administrative matter and not essentially revolutionary.
AV: Fields I Work With – Pattern Recognition (G06K9 in CPC) – There are literally so much of classes of purposes and tons of courses for using know-how as a result of know-how all the time goes hand in hand with the software. One other area we now have in EPO, Machine Learning, is more summary in nature, which is harder to patent without precise implementation. In Part II, we dive deeper into EPO's strategy to AI purposes.