Discussing Infringing on AI Patents with IBM, SINTEF and Simplifai
How Do You Know if Your Machine Learning is Illegal?
A few weeks ago I participated in a panel debate on artificial intelligence in Oslo at a coworking space called MESH at Media Monday Oslo. The main discussion was pertaining to AI regulation. I have covered this topic previously so I thought I would pose a question during the debate. Luckily the question was caught on video and I was able to transcribe part of the panel debate. I will give some key points as background information
Summary — numbers from 2013–2016 (WIPO, 2019)
- Patents in AI are growing rapidly, quicker than any other category within technology
- Machine learning is the dominant AI-technique. Deep learning and neural networks are the fastest growing techniques
- The field with the highest amount of patents is telecommunications
- The most frequent combination in filing is deep learning and computer vision
- Companies are dominating AI patenting. The company with the most AI-related patents is IBM.
- China is growing rapidly in the amount of AI-related patents.
I covered this topic more in-depth before if you want to read up on the status of patenting AI through the World Intellectual Property Organization report from 2019:
The Panel Participants
Alex Moltzau (me) from KPMG | AI Social Research
Anders Bryhni, SINTEF Digital, a commercial research Actor run as a nonprofit organisation
Lars Erlend Leganger, PhD Director | Head of AI/ML | PwC Norway
Loek Vredenberg, Technical Leader at IBM Norway
Miriam Øyna Product Manager at Simplifai
An Excerpt from the Debate
Nick Sheriff (Media Mondays): I think that was varied insight into looking at things from a different point of view. I think that was an interesting view, it is at the end of a day a tool. We are interested in the planet. We had the first event that combined sustainability and tech in Norway. Coming in between both of you guys I think we have that. We have powerful AI to look into all these algorithms, go into the data of organisations and see how we can make our companies more green and sustainable, so on and so forth. Two interesting perspective.
Anders Bryhni (SINTEF): Just a comment, because I agree with you Alex. It is a tool, but we have to consider the amount that goes into these models. We all are digital ecosystem. You mentioned social media. I don’t know what is the footprint of my phone every day. The whole ecosystem, but there is quite a bit of research about to happen on getting model training more efficient working and working on energy consumption.
Nick Sheriff (Media Mondays): so one of the questions sent to Miriam, and I think it goes to everyon in the panel as well. How does simplifai start infringing on existing patents.
Miriam Øyna (Simplifai): so we built the technology from scratch and of course we talk to others about what they do. We have colleagues that are searching for the latest technology to keep an overview of the market. I think perhaps we can make it even more structured form. I don’t know if you have anything to add? Do anyone else have any thoughts? The best way to ensure this.
Alex Moltzau (author): I can say a quick word, although I want to give you the microphone instead. It was a question from me actually… Sorry! But the patents within machine learning is the highest growing category within technology. There is a massive increase in the patenting. IBM owns a lot of patents, they are one of the companies that own the most patents. I don’t know if you have a solution? It is nice to have you next to each other.
Loek Vredenberg (IBM): again a couple of answers. Because we have so many patents we are number one in number of patents in US six years in a row, last year we had 9100 patents in one year, next in line was 3500 or something like that. So yes we have AI technology and how this can be used in production. We open-source these things because we want to collaborate with universities and academia on developing this as a society, because IBM alone cannot do that. We do the patenting to reward people who have good ideas. If you don’t do that people stop working on good ideas and you will not have the progress. But at the same time we as a company decide that we want to open up open-source it and share it with the rest of the world. When we develop our idea both from a basic research perspective, but also a technology usability perspective we are very much aware of the patents of other companies.
Alex Moltzau (author): just a quick one, it would be awesome to see some LawTech AI because that’s probably what is going to happen. I see a lot of new LawTech startups in London and there are discussions in the LawTech environment in Oslo about this as well. So I think that might be a reaction against it, using text based machine learning to find out if you are infringing on machine learning.
Nick Sheriff (Media Mondays): btw we have an exclusive with four lawyers in 2020 to talk about the latest technology disrupting the law. This is already mainstream in the United States where you have lawyers fighting for people’s parking fines. Once again, one of the questions from the audience, did anyone have anything more? Talking about trusting AI, what drives AI is the data so should we not be talking about the data?
Loek Vredenberg (IBM): Data curation is one thing. We are working with the best cancer hospital in the world. They train the AI, we develop the technology, but the training is done by doctors. In doing so there is a fine line between supervised learning and unsupervised. Once a system has been taught thing to recognise patterns, it has been taught to reason over these patterns. There are doctors, two-three-four at one time taken out of rotation, so you need to calibrate all the time. We cannot allow for too much autonomous learning, so there is the algorithm to get to a decision. Predicting fairnesss, another is getting the samples, so predicting the information is that much better for all parts of the population, if not you will have a biased system. We do a lot of work and research to avoid this, but we will not succeed a 100%.
Discussion after the Debate
Clearly I think this is an interesting situation. Especially when you have one of the largest actors in AI and a small company in the same debate. A few question that arises:
- What is fair patenting and policing of AI? Fairness should be considered equally in this area.
- If there is a patent or use that is important for humanity or to help save the planet from a climate crisis, can we make exception? Consider for example the Internet and what would have happened if that was not made freely available, and instead commercialised. If this situation comes up: how do we know?
- When is a small company liable if they have no way to check the large existing frameworks of growing AI/ML patents?
- Large companies such as IBM can afford to patent, will there be an easy process for smaller companies to do so? If that is the case should we even make patents easier.
I have found a few people writing on this topic that you may want to check out.
Patenting AI: Let’s start with a history lesson
When I start talking about artificial intelligence (AI) I go into full geek mode. This stems from a fascination with…
A worldwide overview of patents by the British Intellectual Property Office:
The paper I mentioned by WIPO:
AI in a patenting strategy:
Managing Intellectual Property
Artificial intelligence (AI) and machine learning is allowing life sciences firms to better analyse data, produce…
World Economic Forum on AI and Patent Law from 2018:
Hope this was helpful! There is currently no great immediate solution (if there ever was to IP management), however it is an important discussion that we must have if we are to implement both fair and legal solutions in the field of artificial intelligence.
That’s a nice algorithm you got there– is it an illegal one?
This is #500daysofAI and you are reading article 185. I write one new article about or related to artificial intelligence every day for 500 days.