Gordon Harris: Good day and welcome to the latest in our series of interviews with leading and influential people in the world of IP.
My name is Gordon Harris, from the Gowling WLG International IP leadership team. I have been involved in the world of IP for 35 years now and it is fascinating for me to look both back and forward at the trends which have emerged over the last decades and all the developments which are already influencing the future of intellectual property globally.
Today, we have our eyes very much on the future. There is nothing likely to be more influential on the evolution of IP protection in the coming years than the rapid development of artificial intelligence. Whether it is the impact of the technology itself or the philosophical issues regarding authorship and protection.
A.I is already a hot topic and the subject of a great deal of discussion and thought. Now the World Intellectual Property Organisation has established a division of artificial intelligence policy to monitor, research and develop policies designed to ensure that WIPO maintains its role in the promotion of innovation and the evolution of the global intellectual property system and my guest today is the first director of that division, Ulrike Till.
So, welcome Ulrike – great to see you.
Ulrike Till: Great to see you. Thank you for having me Gordon.
Gordon: Ulrike, you have had a long and varied career in IP leading to this point. Perhaps it is best if I let you describe the path by which you have reached this particular destination.
Ulrike: Sure. Thanks Gordon. I mean we have known each other for quite a few years and I believe I have a recently meandering path through IP. Maybe a jack of many trades and master of none. I will let people be the judge of that.
I count myself incredibly fortunate to have more than one life in IP. I am clearly German by accent, although I did a lot of my training in the U.K and that's where we met for the first time.
I have a background in chemistry and bio-chemistry with a PhD from Oxford and I settled to do law with a very specific aim of working in the IP field and IP litigation and at some point I can tell you there is another side-line story of Christopher Huge being instrumental in getting me on my path.
I started off as a trainee at Bristow's doing patent litigation. I was lucky enough to have the opportunity to requalify as a lawyer in Germany and to work in Dusseldorf, so had many years of a life as an IP litigator, patent litigator in private practice although I had partnership before jumping into an in-house role in the generic industry.
In my last role I was heading the global IP team of one of the largest companies worldwide of 80 people and about five different continents before jumping into a world intellectual property organisation and what I am doing right now.
What I like to think is, in very many ways although my meandering career path has been – has a little twist and turns in it – everything that I have done so far in my career is a huge benefit on the breadth and scope and the questions we are looking at right now. So it is one of those fortunate things where I think everything that I have done has fallen to place to really help with my current role.
Gordon: Yes. That is really interesting because as you said, you have had a huge variety of things. As you said, our paths have crossed a few times over the years. Not least when you doing some litigation – we were doing some litigation together on fairly important medical devices.
Do you miss that? Do you miss the cut and thrust? I mean litigation is a pretty all-encompassing thing is it not when you are involved it is very absorbing. Do you miss that?
Ulrike: Well I think if you ask any litigator the question, the reason why we do it is because we think it is the most exciting job in the world. I think any litigator will also say there is a real wish to win that drives all of that – ambition.
But there is also, I think for me, when I look back why I loved litigation as much as I did. I mean there is the question of really complex problems in litigation and the context problems are not always about winning. They are, what is the best outcome that you are trying to achieve, and how you are going to get there?
There is some real out of the box thinking that you need to do in getting there because it is not just all about winning although that is really nice and I probably am one of the people that whenever a case settled, it is always slightly disappointing because you really wanted to know what the answer was.
Gordon: I think we are all a bit like that.
Ulrike: So litigation for me, there is such an element of problem-solving. There is such an element of how many times you look at it, each case is different and you find another question that you do not have the answer to.
And for me personally, one of the joys has always been to work with the experts on the expert report and really understanding the technology and seeing how that translates to cases.
And when I think about it this way, what I do right now ticks all those boxes, so it has complex problem-solving and there is understanding the questions and really it has got the added benefit of when you are a patent litigator you very often look at a technology that is coming to the end of a lifetime when a patent is expiring, so you are 20 years out of date, what are you getting right now is the future – it is cutting edge, it is evolving.
So in some ways it is patent litigation, all the benefits plus!
Am I missing it? Yes! Does it make me less excited about what I do right now? Absolutely not!
Gordon: Yes. Interesting. I mean I think we both agree that patent litigation is a great and exciting thing to do, but you did say to me the other day that you thought that you now had the coolest job in IP.
Gordon: It must be exciting to be involved in the very early stages of – not just of the law and the technology but also creating a new division. So what are your first priorities as you get established in this new role, and what would you say you have done already, maybe?
Ulrike: As you say, we are brand new. The process at Whitehall started, or with IP all started in September 2019. The division itself started in January this year. So we are eight months into establishing that.
I think, personally for me it is a challenge although I think every skill that I have acquired in my past lives, really helps in what I do right now? There is an element of understanding how that fits into global policy system and into a truly international organisation and system and how do you translate that.
It is understanding how you do policy on that level? I mean, also for me personally there is a real managerial skill and question involved in building an entirely new team and it is a fairly small team at the moment, but it is a very new team that also arrived very recently at WIPO.
At the global COVID-19 pandemic, where you are trying to build a tightly knit team and a small task force to do that and setting up going forward, then suddenly in lockdown when everyone is working from home, so personally that is a formidable challenge. There is a challenge of course, how does it fit in with WIPO because clearly the sectors have massive experience – whether you look at patents or copyright and how do you draw from that experience and pull it together and I guess, we will talk at some point about the overall policy strategy big picture because I think it needs all the players in order to do that. It is building both relationships. It is building the relationships with the member states in order to take that forward.
I think probably the biggest challenge I find is learning. I come home every day and my partner will tell you, usually my sentences start with "you wouldn't believe what I learned today". "Did you know that?"
There is an element of actually getting the basics in place and very much when I talk about personally and for the team, also counts for the entire process that we are running whilst really having a foundation of understanding what the questions are. Because I know that there is huge enthusiasm to talk about some of the answers to the questions and I am a great believer that you need to understand some of the technology and you really need to understand the overall set of questions before you end up on the track discussing what the potential answers might be.
Gordon: Well, that sounds like you have laid the foundation anyway. You talk there about all the number of issues and questions that arise and there are many, are there not? The whole business of A.I raised is moral and ethical considerations as well as legal and economic ones.
One line of discussion is whether or not the grant of IP rights, whether it is patents, designs or copyrights, is that – does that reflect the reward for human endeavour? The contribution of the individual, or is it really the reward for economic investment. So whether they then arise from A.I or an individual would be irrelevant! So, very much appreciating here and I want to emphasise this that you speak for yourself and not WIPO on this and the other topics.
I would be really interested in your views on this point and whether they have changed at all as a result of the work you have done so far in this division.
Ulrike: I think when we go back to basics really what are the IP systems designed to do. The IP systems overall are designed to incentivise human invention and creation and that is really the crux of A.I/why A.I raises such big questions because, until very recently, creation and innovation was solely with a human remit and that very much defines us as a human species. It is the human creators, the human artist. It is the inventor in the picture, in the garage and really, the fundamental question for me is, how you balance the value of human invention and creation over A.I invention and creation.
There is a question whether A.I needs incentivising at all, because when you think about the speed with which A.I can develop, where it might take a creator a long time to come up with a single song of A.I comes up with a million songs within a space of a very short time. Again, what does that mean for the reflection of the value in it? I think at some level there is the discussion about, does A.I need to have incentives. And that is understanding the technology and what is putting into it.
I also think one of my biggest takeaways from what I have done so far, and why what I am thinking has shifted, is for me it is really important to see the bigger picture. Because A.I raises questions, not just for IP. A.I raises questions of ethics, of security of competition law of labour law of education – I mean the list goes on and on and on and really what the IP system does – and IP policy needs to fit into the overall jigsaw of where do we want to go as humanity with A.I and what do we want to incentivise or not.
Then there is the small – well not small but smaller section of what does that mean for IP policy? And then when you look at IP policy and you come to the conclusion that you want to incentivise A.I. again it is a big question of saying, how do you do that?
Because I think a lot of the discussion we are seeing at the moment and because it is a fascinating one and people can see and feel it, is whether A.I can be an inventor or a creator. But when you take a step back, A.I really, for me, has three elements. You need the training data, you need input and it is really the availability of training data that has been driving the A.I revolution. You need the A.I algorithm in the middle and that generates the output.
I sometimes think of it as a cooking recipe. You need the ingredients, you need the recipe and out comes a dish at the end.
There is a question really of, if you do decide you want to incentivise A.I with IP, where do you put the incentives in that overall system?
And then at the end of the day, the question becomes, once you know where you want to incentivise, how are you going to do it? Does it fit within the existing legal frameworks? Do you just tweak those? Do you have to create new frameworks in order to make that happen?
The one thing I know is, I thought I had quite firm views when I joined WIPO. It is one of those feelings, the more you know the harder it gets. I think I have become much, much better at understanding the questions and much less clear about what the answers should be because I think when you take that step back and you understand really how complex the situation is, the answers become less easy.
And I know that is a bit of a fudge, because you wanted a definite answer from me.
Gordon: No, no, no. No. I mean this is an ongoing process. Everybody is learning as we go and things keep happening that make you think, well we do need to take a step back and look at this. As things stand, the machine cannot be named as an inventor. The EPO decision in the Dabas patents makes that pretty clear.
But nothing is ever as simple, as clear-cut as it should be is it? An A.I generated creation, whether it is acceptable to copyright or patent protection or both, is probably not going to be simply the result of switching on a machine. As you said, there is inevitably human input whether it is in the learning data at the beginning or polishing and perfecting at the end.
Is that the simple answer to this? Is it that you will find some human input and there may be you will find your inventor? That is in the nature of putting things into the existing boxes or maybe we need a new box to describe how this particular process works.
You mentioned a fascinating example earlier on about creating a song and you mentioned an example to me the other day, which seemed to be quite a good illustration of this. Would you like to talk about that a little bit?
Ulrike: It is absolutely right. The Dabas decisions and cases are fascinating but in some ways they create the question of, can A.I be an inventor or creator and it very much is, what does the legislation say and does that need to change.
In some ways, again you will hear me saying a lot, it is taking a step back and really understanding what is going on because whether or not the legislation at some point has changed and then A.I can be an inventor or a creator, I think there are much deeper questions, much more fascinating questions and really questions that make a lawyer's brain hurt, irrespective of what the outcome to that is.
It also goes to, one of my big hobby horses is needing to have an understanding of the technology and what it does, what it can and cannot do. Because this is what policy and law operates on.
I think when you take that step back, it becomes more complicated and it becomes more diverse. And it is an awful thing when a lawyer says, it depends on the facts of the case. But there is an awful lot of that.
The example I can give, and I have not told you about the A.I exhibition yet and I can do. I skipped that! But WIPO are putting together and exhibition on A.I and policy that will open in September. One of the sections and the purpose of the exhibition is really to raise an understanding of what the questions are without guiding any of the answers, but really having a level of everyone engaging with the right set of questions.
One of the sections of the exhibition is looking at A.I and music and it gives some examples which – I mean this is part of the things when I come home and say you would not believe what I learned today. But I give you two examples on the opposite end of the scales of A.I and music.
There are certainly a couple of companies out there that have created an A.I algorithm? That is being fed by millions of training data of music samples of existing music and you can go on the internet. You can log in, it is free. You press a button and within the space of a couple of seconds, you can select whether you want rock music or jazz and within a couple of seconds, it will create your own soundtrack and your own song.
And you will be able to hear some of that in the exhibition so watch this space but you will get some samples and you can make up your mind of whether you can even distinguish that from pure music.
That is one end of the scale and clearly there is a lot of training data and there is a lot that goes in the A.I algorithm. But you could almost imagine it is a press of a button and something comes out that is almost – if you want to go to the scale of autonomously created by A.I.
Although, I mean we can discuss whether that is really autonomously created. There is the other end of the scale and again, it is a fascinating story I did not hear until recently.
This year there was an A.I song contest – it is what I call Eurovision goes A.I. It was an international contest I think the idea came from a Dutch broadcaster where teams worldwide were used to create a Eurovision song using A.I.
I think it was 13 teams worldwide and the winning entry, I think they were announced in April is a little company in Australia called Uncanny Valley. The song is beautiful the world. And it is fascinating because when there is a story how they created that song and clearly they fed in training data or some of the training data was every Eurovision song that had ever been created.
There is an A.I algorithm but then, you see them sitting in their sound lab and you hear them listening to some of the samples and some of the results and selecting what they are going to use. They are producers, they are sound technicians, and there are some human additions. I think they put some Australian animal noises into it to make it authentic.
The lyrics were completely generated by A.I and that was put together. But I think that example at the other end of the scale, really is not just an A.I question, it is a question of in a digitalised world that we live in, the value creation chain is getting more complicated and we are really going from, and this is a stylised view, but from an inventor in a garage and musicians with a guitar, sitting in their living room, to a system where there are so many different players contributing to an invention or creation.
And when you look at the example of A.I it is someone creates the training data. Someone or a group of people write the algorithm. There is the person that asks the algorithm to do something on a certain basis of data. There is an output. Do you then have to find a human that could do something with it? Feed it into something.
And for me, one of the most fascinating questions is that value chain. Because when you go back to basics and you say, who is an inventor, who is a creator? Even without A.I in it, what makes/gives you enough of a contribution to be a co-inventor, to be a co-creator. Are these all similar? Where is that element that makes someone a creator or an inventor?
I will just leave you with a thought. If each of the elements are very, very small, even if you come out with an invention or a creation but you cannot really identify the big impact contribution to it because it is all marginal that it is too each other. What does that mean for – can it be patented? Can it be copyrighted?
So I think – this is where my brain hurts and I go round and round in circles. I think, the question whether A.I can invent or create what does legislation say is a very legitimate one, but even that aside, when the complex value chains come in, in the digital world whether that is A.I or not, there is some really unresolved questions in all of that.
Gordon: Fascinating. An example involving Eurovision songs brings back to mind something we used to say in the early days of computing, which was rubbish in rubbish out. But I mean, there is a more serious point there is there not. Because you talk about training data and the key role it plays in all this.
There is a fear that you can in-build prejudices into A.I through the training data and certainly the EU is onto that and is looking at ways to try and ensure that this does not lead to all manners of prejudice.
Is this something that you are having to concern yourself with?
Ulrike: Certainly from an IP policy side, there is a real question – data raises a universe of questions. I think from an IP policy side, one of the questions is, can you protect data? What protection do you have for data? Should there be different protection of the data? And that is very much recognising that – not big data and big data is fairly – it is fairly cheap and the value creation is a creation of training data.
When I say training data that is standardised labelled data that machine learning and A.I needs and for those that are not familiar with it, the example I always give is – image recognition. If you want to train an A.I algorithm to distinguish pictures of cats and dogs, the way you do that is you have a labelled set of training data of thousands of pictures of cats and dogs that you label beforehand as a cat or a dog? And as you feed it into the system, it will break down the pictures but it will realise that pixels in these corners, not in those corners – like the cats have the pointy ears, the dogs have hanging down ears.
Then you feed the A.I a set of pictures that are not labelled. And you look at the results of the outcome and a lot of pictures will be labelled the right way. Others will not be and you need to relabel those and feed them back into the system so the A.I gets better and you can imagine that that creation of the labelled training data can be an investment to get it done.
There is a question of that training data you use. What do you do with the investment? Should you protect it? Do you want reinforce that data? What are the rights involved in it?
The second bucket as you say, is the potential bias in the training data and where that comes in in the IP policy field. As you can imagine for example, there is a fair few A.I algorithms out there that create paintings and pictures in certain styles. So if you want to have your own Van Gogh, the training set of pictures of everything Van Gogh has ever created, there is an A.I algorithm behind it. You press a button or you can even give it a photograph that you have and then the A.I turns that photograph into a Van Gogh style picture.
The training data is really important but obviously on the training data in itself can be copyrighted. So if you feed in copyrighted works, is that an infringement? Is that not an infringement? And then the question, if you want to avoid the question of whether that is infringement or not, you are only going to feed it pictures that are out of copyright.
If you want to have a creation of pictures of women, the outcome of it will be based on non-copyright or copyright expired pictures would give you an image that might be 70 years plus out of date. So that is one of the examples where IP actually feeds into bias because how does – you probably want as current a set of data going in for the output that the A.I is generating.
Obviously bias has many more questions of the quality of data. The completeness of data. The reflection of the data of current ethical and moral standards. So bias is much, much wider and data, than just IP but there is an element of IP policy that goes into bias.
Gordon: Yes. I can see that and also on the question of infringement. Back in the day when we used to protect industrial design by copyright and even then since, in relations to design right where copying is needed.
You could induce copying in the way you commissioned something. You would say, well I did not do that. But the brief you had given someone to do it, basically put limitations on what they did in such a way that they were bound to have effectively copied and I suppose that revisits again in the context of machine learning and training data. You can induce a machine to create copies.
Ulrike: Yes. There is a whole different set of that of – again in the exhibition which I think everyone is going to visit – we have one exhibit of African mask art where the data used is African masks to generate new ones, but the gentleman that does that work actually does a comparison of how close is the output to the input.
It really becomes a question, does the output start competing with the input. Are you simply copying or is it creating enough in its own right and very, very different.
You are starting to hit the points – you are becoming like me, when you start realising the universe of questions and the more questions you ask, the more questions there are. It makes finding the answers more confusing.
Gordon: Yes. Looking at what the EU is thinking of doing in terms of regulation. They seem to be wanted very much to introduce concepts of transparency. Human agency and human oversight and is that something you think will influence your preparation and policy through WIPO as well?
Ulrike: Well I think I look at it from an IP policy hat. And really from the IP policy hat where have questions – again without any answers, is you know that players will find ways of protecting their investment. That is part of the game of development, of innovation of creation out there.
If there is an uncertainty about what protection is available. So you take the example of data, of training data. If there is potentially a question of, it is valuable can you protect, should you be able to protect it?
One of the answers to that question is to start relying on trade secrets, on the law of confidence in order to do that. I think from a transparency point of view that might be slightly outside the IP arena, but within the IP arena where it comes into, is there is a whole lot of questions of uncertainty within the IP system? But then turning into a greater reliance of trade secrets.
A.I has the difficulty that people describe as the black box of an outcome comes out at one side, but you do not necessarily have access to the algorithm and you do not necessarily know the training data that goes into it on top of the technology even if you knew all of that because of the way how neuro-networks work and how they are dynamic.
Very often, once you set the system going, you might not be able to predict – that is the whole point – you cannot predict what comes out at the end.
That feeds into questions of trust in A.I and to questions of bias. And certainly increased reliance on trade secret. There is a question mark of whether that then fuels the system of the black box.
Gordon: Right and so, from a policy-making perspective, you presumably you do not want to encourage people down the route of using trade secrets. When you are thinking about whether or not you need new boxes as we discussed earlier, may be when you are creating laws or proposing the creation of laws, encouraging transparency, encouraging people to not rely on trade secrets might be quite important?
Ulrike: Again, it is a balance because one of the things I find fascinating is that very often confidence of trade secrets are seen as a hindrance to innovation or creation and I think – one of the things not to forget about is the law of confidence can actually be a huge encouragement because it actually allows things that you would not necessarily share confidential information – that trade secret with its data, whatever it might be. Gives you cause of action so you can have confidentiality agreements. It is what happens in companies, what happens when inventors go and see big companies.
Actually having those provisions very rightly play a role in incentivising of innovation and creation. The question for policy then becomes in the A.I field. Again you are trying to incentivise innovation and creation and where is the balance between the two.
Do trade secrets and confidence – does it sway to the fight that actually harms innovation or puts it just in one corner of the people that, for example, have the data concentration?
Or does that actually foster innovation because something that might – there is a question where they can be protected actually allows people and players to share it in the market.
I do not think it is quite as simple as saying, trade secrets are necessarily bad but again it is the jigsaw of the balance of where do we put that.
Gordon: Yes. It is another example is it not. You ask a question and the answer just opens up many more questions.
Gordon: I am not going to let you off as a former patent lawyer without asking a very patent specific question here.
Over recent years we have had the concept of plausibility has sort of found its way and in particular, I know you were involved in cases where sufficiency was a big issue. In that context I remember Robin Jacob once said the best place to hide a leaf is in a forest and the patent law has begun to set itself against the notion that you should identify thousands of options and somewhere in the middle of that, is the inventive step.
Now, obviously that was in the context of, people in white coats in labs could not be expected to test thousands of options. But maybe, a well-trained machine using A.I could test thousands of options in a matter of seconds?
So are we going to need to differentiate between sectors or are we going to need to change principles altogether to take account of what A.I can do in a context like that?
Ulrike: I think and this is very much personal opinion. I know that there are some arguments but this is very personal opinion, but there is arguments of different sectors have different requirements. We know that has been the case since I can remember even in patent law.
Does sufficiency mean the same in the lab and chemistry as it does in telecoms. I think those arguments are not necessarily new. I think you can always find reasons why differentiated rules are the way to go.
Again, having spent so much time in the U.K and the U.K courts, I have always found it fascinating that you can have a single set of rules and that the judges manage to navigate them quite well across industries being open to those questions of course and the question marks.
I am also from a background of a litigator, I am acutely aware of uncertainties creating legal uncertainties or loopholes creating uncertainties in litigation and that is for operating and innovation and creation, not necessarily a good thing because it opens scope for having disputes in the system.
I probably lean more towards having general rules that you can then adapt to the circumstances rather than breaking down the rules because you can then imagine the arguments – if you have different rules in different sectors, and so much of innovation or creation might overlap sectors and when you look at bio-technology, is that medical science? Is it engineering? Is it mechanics? Is it computer science and what rules apply?
Again it is the digitalised world where so many more fields come together.
Gordon: Everything is going to be tech soon is it not. I mean, we have FinTech, we have MedTech. We are wearing our medical technology already.
Time has overcome us I am afraid but before you go. When you look back on this role in the future, when either you stop work or you move onto something else. What do you want to have achieved? What would you like people to remember from your time as director of the A.I division at WIPO?
Ulrike: That is a really difficult question. I mean I have had some conversations with colleagues. Certainly in the dispute resolution centre at WIPO and it is fascinating to talk to them because they clearly remember when that was very small and only just started and it is almost unimaginable when you hear their stories, when it all started to what is has become right now.
Certainly in the utopia, if we could in some ways, replicate that that would be fantastic. I think on a much more – not achievable but on a level where I think we are going, is establishing WIPO as a forum to have the discussions because as we said, there are so many questions and while they are confusing and there is a myriad of questions I strongly believe that understanding the questions and having these discussions across a wide range of stakeholders will actually, in the long term, help finding the answers.
If we can establish that as the go to forum where you can have these open discussions and people come together? That would be a fantastic result.
On a personal level? I think life is very short and I think we should have fun with what we are doing. If people will look back at some point and say, that Ulrike person, it was a crazy time, we were all learning but we had a tremendous amount of fun doing it. Then personally I would be extremely happy and satisfied.
Gordon: That is a great note to end on and thank you very much indeed for being my guest today.
I wish you every success in this really exciting new role. You are right at the heart of the most disruptive and potentially game-changing period in the long history of IP, certainly for a very long time.
So thank you very much indeed, we really appreciate it and you have been, you have given me some really interesting insights. Thank you Ulrike.
Ulrike: Thank you for having me. Thank you for listening to my rambles and my answers. And as I said, it is the coolest job in IP out there and I hope I made the case for it. All the very best to you! And thank you again for having me.
Gordon: Thanks very much.
Ulrike: Thanks. Bye.