Managing software and AI inventions as IP

Tamara El-Shibib: So, good afternoon everyone. My name is Tamara El-Shibib. I am a senior patent and tech transfer consultant at Gowling WLG based in Dubai.

Thank you for joining the fourth session of our webinar series on tech transfer and innovation. Before we start with panel introductions, I wanted to introduce the webinar series for those who are new to it. So, the purpose of this webinar series is to discuss different topics within the umbrella of IP, tech transfer and innovation. There are a lot of information gaps in this region when it comes to these topics, so the focus is to really share knowledge and best practices in these areas but also to discuss the challenges and what we can learn from them.

For today's discussion, we are focusing on IP strategies for managing software-implemented innovations. We are going to focus on AI because there is a massive surge in AI-based innovations. And frankly, there is a lot to learn, especially if you do not come from an AI background like me. So, having the right IP strategy is key for getting your technology to market and giving it the best chance of successfully getting to market as well.

Before we dive into the topics, I wanted to say there is an opportunity for questions at the end of the session. So, please type in into the chat, and we will get to them in the last five or ten minutes of the session. You will also be prompted to complete a short survey at the end of the session. We would appreciate any feedback you can give us on the session. If there are any particular topics you would like us to cover in future, please also let us know.

With that, I am going to turn to the panel and ask them to introduce themselves. One of our panellists is actually having a bit of a tech issue joining the session so that is Matt Hervey, and we will get him in as soon as we can. For now, left on my screen is Vivian. Vivian, could you give us a little introduction to yourself?

Vivian Wei Cheng: Hi good morning, good afternoon and good evening everybody. My name is Vivian. I am a Singapore patent attorney and head the patent and industrial design group at JurisAsia LLC, which is the Singapore office and South East Asia arm of the global law firm, Gowling WLG. I have a life science and clinical medicine research background, and my daily job involves drafting and prosecuting patent applications domestically and in foreign jurisdictions as well as providing strategic advice on IP portfolio and risk management as well as litigation support and it is an honour for me to present here today as a panellist.

Tamara: Thank you, Vivian. It is good to have you. Next on my screen is Sean. Sean, can you give us an intro?

Sean Flanagan: Thank you. Good afternoon, I am Sean Flanagan. I am the director of technology and commercialisation at King Abdullah University of Science and Technology or KAUST here in Saudi Arabia. Prior to that, I was in the same role in the National University of Singapore, where I also served the role of as director of commercialisation and IP for AI Singapore, a national level AI research initiative. I have been in the University tech transfer space for over 20 years, and I am pleased to be sharing whatever insights I have today.

Tamara: Thank you Sean. We are pleased to have you. Fawaz, could you give us an introduction?

Fawaz Al Qahtani: Good morning, good afternoon, good evening everyone. My name is Fawaz Al Qahtani. I am acting director of Tech transfer of research at Hamad Bin Khalifa University, Qatar Foundation. I have been in the tech transfer industry since 2017. And, prior to this experience, I was a research scientist at Texas A&M for more than eight years. I hold a PhD in electrical and computer engineering. Basically, my daily activities that I do is just to oversee the IP activities from IP capture until the commercialisation, so that is basically what we do. Thank you.

Tamara: Thank you Fawaz. So today's discussion is going to focus on these topics. We have got about an hour. So, we will try our best to cover all of them. This is sort of an informal discussion, so we will sort of follow the flow of the discussion. But basically, we are going to discuss trends in R&D and IP for merging tech with a focus on AI. Processes for identifying and evaluating software and AI inventions. We are going to discuss IP protection considerations for software and AI innovations and then patenting considerations for software in AI because these are treated differently to traditional tech. We will also cover data assets so, these can also be really valuable when you are dealing with AI. How can you protect your data assets? And then, we will briefly discuss the challenges and commercialising software and AI-related inventions. There is a lot to talk about when it comes to the challenges, so I think we will need a separate session on that, but let's just dive straight in.

When it comes to emerging high-tech areas so they are broadly categorised as these five, and there is a lot of IP development in these areas particularly because a lot of these are sort of platform technologies. And you can apply them across various industries. I think the fastest growing in terms of these top five is definitely AI and machine learning. And there is a lot of stats sort of backing that up, which we will cover in a second, but before we do that, I want to turn to the panel.

Sorry, hi Matt, are you there?

Matt Hervey: My camera is showing me upside down, let me just see.

Tamara: No worries. It is also showing you with my name as well, but I'm just going to turn back to introduce you. Hi everyone, just to get Matt up to speed. Matt Hervey is a partner with Gowling and the head of our AI team based in the UK, and Matt, we are just going through the topics for today, and then I will loop you back in.

Matt: Fantastic.

Tamara: So turning to the panellists, I want to start first by asking Sean and Fawaz in terms of R&D, what are the trends that you are seeing when it comes to these areas? Sean, do you want to take that first?

Sean: Sure, thanks very much. I think definitely here in KAUST we are seeing a tremendous interest in approaching almost all levels of research, whether it is desert agriculture or biology; any of our traditional areas of research are seeing an ever-increasing interest in developing AI or ML solutions to go alongside of those technologies. We are seeing an increased utilisation of the data that has been collected through the physical manifestation of research often sensors or devices for collecting information, and how we can use AI and ML in order to process those mountains of data in order to determine what the patterns are. So, what we are seeing is there is definitely an uptake in the amount of software-enabled technologies or software-enabled hardware. We are definitely seeing a lot of standalone software, especially in the AI institute. And we are seeing an ever-increasing interest in adapting new solutions in AI and ML for historical approaches to, whether it's drug discovery or any other analysis that might have otherwise taken a lot of significant repetitive research.

Tamara: And Fawaz, would you say that it's the same for QF as well?

Fawaz: Yeah, it is, exactly. I think right now, for example, last year, we had more than 150 inventions disclosures, so by looking like into the whole spectrum of IPs, we identified more than 60 to 65%25 of the whole activities, they are in the domain of AI. Some of the AI, like in the health, lots of even material science right now, I see lots of activities they are using AI, which is I mean this is for the first time I see it like this year, and I see lots of involvement and a lot of engagement. And I see even in the industry, there are like they want to explore more about getting AI into the domain of material science. And you know even if you looked at the AI last year, it is completely different from this year. I mean, we have been discussing within the tech transfer and within the research institute for the last three weeks about some of the areas that we have been doing research for the last seven or eight years, after the introduction of large language models like chat GPT and the multimodal GPT4. So, right now, we are going through the different workshops, talking to people, to the scientists, and we are trying to explore all these. Shall we do exactly the same as we do before or things are changing? So, I think the industry is moving so fast, but definitely, AI, it is getting into all of the disciplines, it is getting into all the portfolios. We have Bayou, and we have health, we have Bayou, we have energy and environment, and we have computing. And I can see lots of the AI machine learning they are getting into these different disciplines, and I believe things will be changing in the future in terms of AI involvement in these areas.

Tamara: And I think the numbers here completely correlated with what you are saying. So, there has also been a bit of an AI patent boom over the last ten years, and a lot of that is sort of focused on machine learning as well. So, here is just to give the audience a little bit of background into what is going on with IP with AI in the IP world. There has been sort of over 340,000 applications according to WIPO, that relate to AI, and half of those have been filed in the last ten years. When it comes to patent applicants, the majority of them are based in Japan, US, China and the top five AI patent finders as you could guess it is sort of the big tech giants. In terms of university research, a lot of that is coming from China. So, they are leading the way in terms of scientific publications and leading the way in terms of university applicants as well when it comes to AI.

Generally speaking, and before I turn to Vivian and Matt, my understanding is when you are talking about AI innovations, these can be grouped into three areas. Either techniques used in AI, like machine learning, or its functional applications of AI, so natural language processing or computer vision and then there is the application of AI in different areas. So, like I said, a third of patent applications that are filed relate to machine learning. I am going to turn to you, Matt. Is that one of the trends you are seeing as well when it comes to AI, is it machine learning that is taking over in terms of IP activity?

Matt: Yeah, 100%25. So, AI has been a search area since the 1940s and for the majority of that time, people would presume what is now known as an expert system. Whereas, humans trying to code a machine to replicate a human task. But really since about 2010 the artificial neural network as a form of machine learning really re-established the concept that machines can teach themselves using training data, how to automate certain tasks. And among those the two you have pulled out the natural language processing computer vision have been incredibly significant commercially because it allows you to automate a huge range of human tasks, everything from autonomous vehicles to voice assistance. And it is really that technology that has made AI appear on the industrial strategy of almost every major economy. And the only thing I would say about IP trends, in general, is you are absolutely right, vast increases in IP filings. It has certainly doubled in the US in about ten years. I read this morning it is 6%25 of all filings in India relate to AI at the moment. I think the other big trend is trade secrets, huge increase there. So litigation in the States is up 14%25 annually over the last decade, so very significant. And then the other area that has become superhot now is generative AI, that is AI to create images and texts and code and just about anything you can think of, and that is not only becoming significant so open AI garnered 100 million users and claimed to be the fastest growing technology in terms of adoption, ever. But, it has also sparked litigation. So, in the US and the UK, we are seeing the first litigation about generative AI in terms of copyright infringement.

Tamara: I am going to come back to the point about trade secrets because I think that is really interesting that there is a rise in trade secret litigation at the same time as we are having this AI boom as well. Before I do, Vivian, I'm going to come you. Are you seeing similar trends, I guess?

Vivian: Yes, definitely. Definitely. We are definitely seeing very rapid developments in this area, especially covering Metaverse, NFTs block chain technologies. And I would like to mention something interesting that, even in the trademark field we have seen a surge of metaverse and web 3 trademark filing in recent years as many companies would file applications for marks for use in the virtual environments and in relation to NFTs. So, in Singapore, there is an example concerning the company Nike. So back in the 1990s, Nike obtained a single trademark registration for that famous swoosh logo for footwear, clothing and headgear. But recently, they have filed and obtained acceptance for another single trademark application for the same logo, but it covers various virtual goods and services. So, I think this shows that they really expect more local users of the Metaverse, which is kind of a promising sign for other businesses to start mapping out the plans for Metaverse presence.

Then in the patent field, I would say that increasingly the life science companies are also exploring the potential of AI. So, to name a few applications of AI in life science, AI has been used to assist drug discovery by identifying potential drug candidates. It has also been used to diagnose disease using medical images, and it has also been used for AI-driven robotic surgeons, intelligent drug applicators or even intelligent logistics such as predicting demand peaks and adapting the supply chain accordingly. So, I would say that in Singapore generally, especially in life science area, and assuming other areas, AI is a hot topic, and everybody is trying to grasp the benefits from it.

Tamara: Yeah, when it comes to the UAE, it is quite similar as well. A lot of the work that we are seeing is also AI-related across different industries as well, health, ag-tech, basically across the industries that are relevant in the region as well. So, definitely mirroring that trend. When it comes to managing sort of AI inventions, for me personally, I see a lot of developers working on solutions using AI. But what is the first step in terms of evaluating whether you have got an AI solution versus an AI innovation? I am going to turn to Sean and ask you what are the processes that you have internally to evaluate when you have got sort of a new AI or software-related invention?

Sean: Well, the first thing we had to do was develop an entire software disclosure system. We saw that with the increasing numbers of non-patent related discoveries that were taking place, we were not really prepared for handling that kind of influx and while it collects a lot of the same data it critically asks some important questions about the origination of the concept and what tools were used in order to develop the solution. I agree, we are seeing applications more than entirely new algorithms. We are seeing approaches to using and understanding an adaptation, and that creates a lot of opportunity for contamination of the original work by outside source, whether it is a shared code or code that is taken from or data that is taken from a source that is not proprietary to the parties that are disposing it. So, our approach is to gather as much information as we possibly can about the software that is being disclosed. Understand what the pathway through to the anticipated market because again, because there are solutions driven, we are seeing that the anticipation is a very fast market. So, we do not have the traditional timelines that we can with patents, in order to evaluate which way we are going to go. Because often we are looking at, do we have the right material in order to move through to the venture funding to get through that first mover advantage and a patent often is not going to be the right pathway for these. So, I agree, we are taking trade secret a lot more seriously which is kind of the antithesis of the university, and we were in the business of not keeping secrets. We are about publication. But increasingly, as some of these tools, especially for student start-ups, are coming out, we are having to find ways in order to make it available. So, I mean, I think it is about collecting the most information you have about how the solution was derived and then taking a critical evaluation as to what they intend to do with it, so that it can survive whatever diligence is going to come thereafter.

Tamara: Yeah, I think what you mentioned as well about sort of figuring out the origins as well is really key. We have seen a few cases where for example, it is very common to use open-source software because it helps you save development time and cost, but sometimes some of this open-source software is under or has restrictions relating to patent ownership or commercially use, especially if it like a subscription-based software that you are using. So, I think it is really key that before you invest time in trying to patent or commercialise something, you need to understand where it has come from and what you can do with any derivative work. Fawaz, I am going to turn to you quickly. Do you have a similar process in place as well?

Fawaz: I mean after capturing the new, we have the new invention disclosure, we try to look at the technology itself. Sometimes we gather more information, and for me and the office, we have these inventor meetings. We sit with them, we discuss with them. So what I realise there are two types of AI. When I discuss with the guys who come in from the AI background, it is different from the solution that is coming from a different, like from health they are using AI, and in health and environment and telecoms they are coming from AI, and I can see there are lots of activities they do is off the shelf type of techniques. And usually, we try to discuss with them, hey guys this is not patentable subject matter. But in the meanwhile, they are providing great solutions, so we tell them, hey guys since it has a great commercial value, then I think it is better off to have it as a trade secret, and then from there, we can further develop the technology to take it to the market. And then, for the AI solution, for example, that has come from the computing research institute, you know, machine learning is just only one block of the whole end-to-end solution. So you have the pre-processing, you have machine learning, and you have even going to the cloud coming back there are lots of ingredients. We can tell from there if this has the potential to be patented or no. And honestly, we are commercialisation driven, and since we are commercialisation driven, we need to make sure that anything that will be disclosed later on when we have a patent granted or patent application, you know this type of algorithm, it is very hard to police. That is when we miss the value. So as I said, the process is we meet, we gather lots of information, and then we decide what is the best scenario, based on the future activities, if this will be commercialised or not.

Tamara: That makes sense. OK, turning to Matt next. In terms of protecting AI inventions, what would you say are the elements that can be protected?

Matt: Well, potentially everything depending on which strategy you approach. So, obviously, the raw data or the training data or the extracted data. It might be in the form of a copyright work, so it could be your using images or academic articles, and they have copyright, and there the real issue is whether you are subject to text and data mining exemptions, which allow people to mine your copyright works without a licence, and that is definitely an issue which is going before the Courts in the US and the UK at the moment. And it is also an area where various Governments are looking whether they need to change the law in those respects. If it is not a copyright work, for data you are very much looking at trade secrets, and we can talk about in more detail later if you wish. Then for code, if it is human written it is protected by copyright. So if you have a human who has written the learning algorithm, you have a framework by which the machine is going to learn, and that will be copyright. But copyright is a limited protection, so it protects the expression how you have written the code, with choice of phrasing as well, but its code, not the underlying principle so, it does not create any monopoly to the methods or the ideas. And indeed in the EU and the UK, you are now allowed to prevent lawful uses of your software from reverse engineering, a term to that effect would be void. So where you are looking is to the extent you are not protecting, you want to copyright, you either need to fall under patent law protection. And obviously, patent law excludes computer programmes as such but does allow some software essentially to be patented where it improves the way an AI works or where it has some sort of real-world application outside of the AI. Then you have the stuff generated by the machine itself, either the model or the algorithm it uses to make its predictions or the outputs themselves. Generally speaking, in most jurisdictions, stuff generated by an AI isn't protectable as a form of intellectual property. There are some exceptions for the copyright. The UK is an exception there, for example, and also designs if they are outwards.

So, we are falling back there generally where IP does not reach on trade secrets and practical measures and contractual measures to keep your proprietary techniques and innovations to yourself.

Tamara: Are you seeing sort of clients in for trade secret protection more than patents when it comes to AI? Only because I know with AI, there is a lot of rapid advancements in the area, and there are ways to sort of invent around, and you need to be able to detect if somebody is infringing on your patent for it to be worth patenting.

Matt: It is hard to measure, obviously. The only public data is on applications. There is no public data on trade secret options, albeit that enforcement is clearly rising, particularly in the US, which suggests people are taking trade secrets more seriously. But, what I would say anecdotally is that clients are increasingly asking us to help with their procedures and the documentation and the like for trade secrets, particularly where their assets relate to AI.

Tamara: That makes sense.

Tamara: So, we will touch briefly on patents and patenting AI inventions. So I think the key message here is, yes, you can patent AI-related inventions. Usually, you have to frame the invention in terms of a problem and a solution and provided you have to look at what is the AI being used for. Is it a technical purpose or a non-technical purpose? If it is technical, is it an obvious use or not? So, it has the same criteria, I would say, as patenting any technology. It has to be something that is new. Something that is not obvious or inventive and something that is technical. When it comes to patenting issues around AI, what are the main issues that we are seeing? I am going to turn to you, Vivian, for this.

Vivian: Yes, so I would say that the first consideration is that, not all the inventions are eligible for patent protection. So, for instance, I think around almost all of the jurisdictions in the world, mathematical methods such as algorithms per se are not considered inventions. But if the patent application relates to application of, for example, a machine learning method to solve a specific problem in a manner that goes beyond that underlying mathematical method, then the application could be considered as invention. So, an example would be the use of a machinery method in controlling the navigation of an autonomous vehicle. And so, I would say that it is very important to highlight the important technical advantages and practical details when drafting patent and this can increase the likelihood of success during the patent examination and bearing in mind that an AI tool can also be embodied in a physical form. So, actually, I have looked at some examples set by the EPO examination guidelines and some EP case laws. So for example, a method of providing medical diagnosis by an automated system processing physiological measurements is considered a patentable subject matter and using the method to identify irregular heartbeats or to predict the binding affinity of one ligand molecule on the one target protein is also considered a patentable subject matter. So, I think particular care should be taken when drafting patents application directed to AI algorithm as the patent application will be published and become part of the prior art so if an inappropriately drafted patent application is rejected, others can use the AI algorithm described in the patent application.

Tamara: I think, when it comes to AI as well, we in the UAE usually look to what is happening with the EPO and in the US as well. We don't have any specific guidelines in the region when it comes to dealing with AI-related inventions. But we do know that here in the patent offices, they do look favourably towards applications that have successfully been prosecuted in one of the other big patent offices and it usually helps speed up examination regionally as well.

When it comes to inventor ship and ownership, Matt, what is the main issue when it comes to AI invention? I think we all know about DABUS the case.

Matt: Yeah and the case has just been to the Supreme Court on that very point. So, DABUS we wait to see in a couple of months from the Supreme Court to see if our jurisdictions will allow invention by AI. But certainly, all the substantive hearings elsewhere have to date failed. But, I think the big challenge is the exceptions which we talked about. So things like mathematical methods and the like. There is also lots of publicly searchable prior art in this field in AI you know very well archive and other sites like that contain a huge amount of prior art.

You have also mentioned the issues with enforcement. So a lot of the patents granted are quite narrow. And so, workarounds are easy. It is hard to detect infringement because AI, as legend has it is the black box, you cannot tell exactly what is going on inside it often. And it is often deployed in the cloud outside the jurisdiction. So, you have to take a lot of care about how it is drafted to make sure you can make a claim within a single jurisdiction. And then the other thing to consider is sufficiency. So is the skilled person able to work the invention, and often valuable AI is based on simple proprietary data. So, you may need to frame your invention in terms of publicly available data to show that the invention can be worked by others. But again, I would say because of all of these issues, do also consider trade secrets.

Tamara: Yeah, that is a topic that we are going to come on to very shortly. Before I do, I just wanted to quickly, sort of, discuss the patent filing strategies that you can take advantage of through dealing with an AI-related application because there are developments that are happening so quickly in this space instead of you might want to speed up examination and prosecution, and so I know that some of the patent offices have that as an option. I was going to turn to you, Vivian. Are there any particular benefits for applicants to file through the Singaporean patent office?

Vivian: Definitely. So, the intellectual property office of Singapore IPOS, it has a range of programmes that allow AI innovators to accelerate patent applications. So locally, we have the Singapore IP fast track programme, which allows patent applications in all fields of technology, including AI that are first filed in Singapore to be granted as fast as six months. So, the applicants can also use their Singapore patent application to expedite the prosecution in more than 30 other jurisdictions through IPOS network of work-sharing agreements. So, for example, AI inventions are eligible for something called ASPEC acceleration for industry 4.0 infrastructure and manufacturing initiative. so under this initiative, the patent officers in eight other Asian countries they are able to rely on the Singapore search and examination report, and they are committed to respond within six months, whether those applications could be accelerated there or not. And in addition, IPOS also established the bilateral and global patent prosecution highways. Actually, it covers more than 20 other jurisdictions, including key markets such as US, EP, China, Japan, Korea. So yeah, there are lots of accelerating routes in Singapore.

Tamara: It is nice to see the patent offices trying to keep up with the trend as well, and offer applicants, I guess, ways to help them stay on top and make sure that they are not delayed in the processes, as well. Because it is so important in this area. I am going to turn to data now because I think this is really important and probably not discussed enough. When it comes to the training data, we are sticking with machine learning here when it comes to AI. Typically, when you are a patenting, you do not necessarily have to disclose the training data itself, but you kind of need to describe the type of data and how it would be used. So, I think it is really important that before you go ahead and patent have a good understanding of where you are getting your data from; what do you plan to do with your data; are you OK with, do you prefer to keep it sort of secret; or are you OK this becoming published? And another thing to think about is the model generating data that is valuable itself, so data the IP that is being generated as well.

When it comes to protecting data assets, Matt, I am going to turn to you. What are the different tools that you can use to do that?

Matt: So, I just say as a preface, I don't know that the answer on sufficiency is settled and whether you do need to disclose your training data. I think guidance so far from patent offices has been somewhat bland that the normal rules of sufficiency will apply. I've certainly had patent professionals suggesting that they might need to literally upload all of the data or supply it in a way that I think the new data and other have been used in the past patent applications. So, that is to be seen, and it could lead to swathes of invalidity if the court decides sufficiency is tougher than we thought. But in terms of protecting your data per se, as I said earlier, it may be copyright, and you may be able to do it that way subject to text and data mining exceptions. But certainly, information per se is excluded from IP on policy grounds, and that is in all international treaties. So, you really will need to protect it as a secret. And that means a mix of practical measures. So keeping firewalls and access logs and the like, contractual measures so NDAs and the like with collaborators and employees. And then, if necessary, seeking remedies under trade secrets law, which are well harmonised under trips, and we all largely have a common definition these days. So it is a commercial secret that is valuable because it is a secret. But all-important that you have taken reasonable steps to keep it secret. So the all-important step, I think the company needs, as well as keeping it actually secret and avoiding a problem, to show you have taken reasonable steps. So, I think a lot of that comes down to making sure you have logs of what policies apply to your core trade secrets, who is responsible for it, who clears publications to make sure it is not leaked accidentally and to collate that evidence of the measures under each policy that applies and we can quickly get an interim injunction in the case of some sort of breach.

Tamara: I think it is key that you sort of show that you are taking reasonable steps as well, just based on the recent litigation that I have heard about, as well. A lot of times the Courts are looking to what are the extra protective measures that you are putting in place.

Matt: It has to be appropriate to the nature of the secret as well. So there is no point in having a firewall if it can be reverse-engineered from a physical object, for example. A rather cheeky defence that was attempted in the States was because the defendant had been able to steal the secret, they cannot have taken reasonable measures that that one was not enough to strike out the case.

Tamara: I'm going to turn to you, Sean. Do you have any mechanisms in place, at the moment, for capturing data?

Sean: No, it is funny we have had again, we go back and innovate our processes in order to deal with the innovations that are taking place because, you know, it takes me back to the days before the AIA, and when I used to give lecturers to new faculty or students about keeping good lab notes and everything, now we thought we were past that. But now, because of data, we are having to, again, make sure that we have got controlled access. All of those good practices are coming back in the university setting in case we have to use a trade secret route in order to protect the data. I mean, primarily, our approach is to look at data as part of the licence that comes into a collaboration, whether it is part of the background IP but also as the confidential information within the agreement to make sure that we have tightened up our agreement so that when we are giving a licence to our background IP routes specifically giving licence to our data because increasingly because universities around the world have historically collected a lot of data that can be repurposed for commercial applications nowadays. And we are finding that there is a market for certain of the data that we have collected as a result of the basic research. So, our normal approach is through licence and contractual controls. We are limiting access by not necessarily publishing all of the data when possible as part of the academic pursuit but to the extent that the information has to be published as part of the academic publication, as opposed to a patent application, then we are kind of out the door after that. So, often there is more data that we can monetize, and there is more data to provide access to, but increasingly data is a larger proportion of the discussions when it used to be just included in the definition of intellectual property incorrectly.

Tamara: Do you see that sort of through your licencing transactions, that data is a core part of the transaction when you are licencing your technology?

Sean: Absolutely, especially if we are licencing a model, an AI model for any particular purpose, the training data is an expectation that the training data is going to be included with the delivery of the model and we are having to educate. So, you can have access to the model, but the price point changes if you want to have access to the training data while you are building your own repository of utilisation data. So, it is causing some I would suggest unique conversations.

Tamara: That's interesting, and Fawaz are you seeing the same thing as well when it comes to data value?

Fawaz: Like, for us, when it has come to the genome data when it comes to the images, like medical images, this one is mostly confidential. We let the research centre to disclose it to us, so we keep it on our records and our database. But definitely, we do not, we keep them confidential, we don't share them with anyone else. So, when it comes to academic, so I don't like remember we had any commercial activities by sharing our data with what we do we do research collaboration with different institutions either regionally or internationally. What we do, for example, we share with them our data, they send us the model in return. So we give them the data they give us the model or sometimes we get data from different universities worldwide, and we give them the model, this type of things. And sometimes, researchers or faculties, they need to publish their own data to become one of the recognition to what they are doing. So depends on the data, so the data if it is just only normal gathering data that have, like food images or, for example, yes we let them just take it and publish it. At the same time, we have Qatar National Library, which all the locals and residents of Qatar, they have access to Qatar National Library. So all the data we have, which is, I mean, I am talking about the data, not confidential ones. So we share it with the QNL, and we put it in the database so anyone can access through the Qatar National Research Library to this data. This is, basically, what we do. This is our activity, and honestly, it is not too much. But the genome project, it is a huge data that has been for a while for the last nine or ten years. We gather lots of data, but still, it is kept confidential, so far.

Tamara: That is a huge project. I remember that from my time when I was in Qatar as well.

Fawaz: It has been utilised a lot, lots of great research outcomes out of it and we are collaborating with some of the hospitals in Saudi and some of the hospitals in Europe so it is coming along.

Tamara: And I think it is really interesting as well when you have sort of open data that different parties can use to build solutions off of. I think that helps accelerate innovation as well. So, that is a great strategy too. Vivian, I know you have worked with a university before, and I know that there are certain measures that they can take or put in place when it comes to protecting and managing trade secrets. Are there any examples of a thing that they have done?

Vivian: I would say that non-disclosure agreements are essential in the early stage of disclosure especially if you would like to see third-party collaborations. So, always make sure that these documents are in place. Then it also includes, very importantly, to train employees, so that they understand their obligations to safeguard the confidential information. Because in a lot of US trade secret lawsuits, the defendants always they will claim that they are not aware that those information are actually confidential information, and they have obligation. So, I think it is very important to train employees early on and try to limit access to the smallest number of people possible who need to see that information. And other measures could possibly include having a good cyber security policy, whether it is a university or other company. And it is also important to have a robust on-boarding and off-boarding procedures when an employee joins an organisation or leaves. Sometimes I see that when they exit that company, they need to sign a memorandum saying that they are aware that certain information that they had access to during their employment are actually trade secrets, and they are aware that they have a special obligation to keep it secret. Yeah.

Tamara: I think a lot of entities are sort of coming on board with creating their own confidential information policies or policies around trade secrets. But I think that is still an area that is picking up slowly and probably correlates to the rise in trade secret litigation and the AI boom as well when you are dealing with innovations that you wouldn't necessarily patent and you might just like to keep as trade secrets…

Matt: Can I add actually that one thing we have in common with reverse-engineering of models. So Sean was saying they would licence the model, but you wouldn't give them the training data. For some models, they are quite vulnerable for reverse engineering. You've got data, test it with data, get the output, and you essentially reverse engineer the model or even elements of training data. So you need to think about practical measures like APIs, putting it in the cloud, and limiting what people can put into it.

Tamara: That is an important point. Thank you.

Sean: On that, just as a suggestion, as people might want to think about it, as we have taken a subscription approach, adopting some approach from other forms of copyright media exchange and saying that subscriptions are a way you can have access to the data, as long as you maintain your subscription you can utilise the data with the model.

Tamara: That is a good way of sort of controlling the data as well.

Fawaz: And can be used through API not to share the data, but you can access the data through the APIs and this work, as well.

Tamara: It's a good point. When it comes to commercialising AI-related inventions, I am going to turn to you Fawaz, because I know you deal a lot with that. What are the main challenges that are coming up for you?

Fawaz: OK, I mean, we have been very active for the last few years, and right now, I think we have up to nine start-ups that are coming out. I look at the commercialisation, it is end-end from the early capture until you go to the market. So sometimes we had issues, I mean, commercialisation bottleneck, like in the middle when we wanted to de-risk the technology sometimes you need partner to take it, for example, to work with them closely until we have the right product or the right solution to go to the market and sometimes this one doesn't work. Sometimes we find the right one, but you know the agreement, the terms that we discuss, we did not come to an agreement, so we had sometimes to stop. But, one of the scenarios, for example, we have a very well-developed solution, the solution goes to the market, and then you have the issues when it comes to the team and how do they manage and how do they have more customers, and some of them they fail. I look at this one as a, we have been experiencing this one, so any related type of start-up that is related to AI. Building a solution, it can be, you know, it is possible, but I think that we need to be careful just when we got to the market for solutions that is related to AI it is very challenging. You have, I mean, the founder or the CEO of the start-up you need to be more engaged, he needs to aggressively getting into the market having getting more customers because, for example, we have this technology that we developed in 2019. It was the state of the art, but we had an issue with the founder. He had been going back and forth. He did not have the right strategy to grasp the market share, and then I can see many technologies doing exactly what we do, and right now, they are getting more share than what we do. So when it comes to the commercialisation, there are lots of factors that you need to be careful of. So you need to look at how to further develop the technologies. Sometimes, when we increase TRL 4 or 5 to find more funding to de-risk the technology. When we succeed in this portion, to TRL 9 and we take the solution to the market, then we have stuff that, it's like any business. So what I see, the solution, I look at the solution, just only it is 30 to 40%25 of the whole end-to-end. But definitely, 60%25 is in the market and how to lead the product and how to get more customers. This is basically, this is the way, more than the technology itself.

Tamara: Yeah, so the solution is just basically one part of it but there are so many other steps and things that have to happen before you can successfully take it to market.

Fawaz: The solution is not, it is not the big part of it. I would say like 40%25 the solution it is worth but definitely, the business side this is very crucial.

Tamara: Yeah the investment, the resource that it takes to take it forward. I agree I think the solution is only like the first part and it can years to take something to market but in this area you have to be quick which means you need to have resources at your disposal. Sean, is there anything you want to add on that?

Sean: No, it is just fast and getting to that first customer revenue and validating the assumptions. I mean it kind of puts the lean start-up model on its head because you need to get out there as quickly as possible. But sometimes that means you will be finding out what your product is while you are speaking to customers with what you have. So, no, I agree completely with Fawaz. You have got to be fast, and it is not about the tech until it comes to raising money, and often that is one of the fall-backs, and somebody will say OK what's your IP provision and they say well we have got a trade secret and that creates a pause in the conversation around raising money which is I think just a matter of educating the investors and making sure that they understand that there are different ways of protecting the opportunity.

Tamara: I think when it comes to commercialisation there is a lot more than we could cover, but this webinar is actually coming to a close in two minutes, and I want to give some time quickly to a couple of questions that we have received.

So the first question is, for the invention disclosure form, would you find your minimum requirement on the invention details to be sufficient favourably past the IDF evaluation stage? So basically, when it comes to sufficiency of the invention disclosure form, what is the minimum that you require from inventors to move forward? I mean, if I were to answer that I would say it is very tricky to do a proper evaluation or analysis on an invention without fully understanding what it is that you are dealing with. But I am going to turn to you, Sean, as well, is there a particular process that you follow for this?

Sean: I mean, we use the disclosure as a starting point, and then we have a dialogue, and it is through those dialogues what we tease out is, do you know exactly everybody who was involved in the development of the solution that you have? For those people that were involved, have you confirmed what tools that they used, and have you articulated within the disclosure what available resources that might have been used? And then we drilled down to find out what the truly novel components are. And I think we err on the side of caution or generosity in so far as accepting what people come through with, but they have to understand that I mean the questions that we ask are going to be a cakewalk compared to what is going to be asked of them if they try to raise money against the software.

Tamara: And actually, the next question is somewhat related. So it is basically how much development work do the investors have to do before you consider something ready enough for you to take forward?

Sean: I mean, I guess it depends on where it starts. I mean, if they come to us with a concept and they have not done any of the coding, then we have got a lot of work to do. If they come to us with a mere commercial-grade piece of software, but they just need to work on the UI, then I mean, I guess, we are almost at the market. It depends on the starting point, it depends on where they want to get to. Often their resources are limited, and what we can do is help by quickly bringing money into the equation so that they can overcome all those challenges.

Tamara: I mean, I think the more development work there is, the more it comes back to what Fawaz was saying about de-risking, right? So, like you know it works, you've tested it out, and you can get more, I guess, investment, more resources assigned or allocated to that particular invention because it is further along.

So it is actually past 5 o'clock now, so I am going to have to say goodbye and end this webinar. But I wanted to thank you all for participating in this. I think this was an excellent discussion, and I think we might need to do a follow-up session at some point as well. And to the audience, thank you for joining us today. If you have any follow-up questions, please feel free to email those to us directly, and I can make sure it gets to the panellists and yes, I just want to end there and say hope everyone has a good afternoon or evening and thanks again.