Speaker 1  0:06  
Young, hello and welcome to today's autm webinar, advantages and risk in using generative AI tools. My name is donville Young. I'm a member of Autumn's educate team and today's staff host. All lines have been muted to ensure high quality audio, and today's session is being recorded. If you have a question for our presenter, we encourage you to use the zoom to the Q and A feature on your zoom toolbar. Should you need closed captioning during today's session, the live transcript feature is turned on and available on your toolbar. Before we begin, I would like to acknowledge and thank Autumn's online professional development sponsor, Marshall Gerstein, we appreciate your ongoing support. I would now like to introduce today's presenter, Lawrence husen,

Speaker 2  0:56  
good. Good day to you all. My name is Lawrence Husson. I'm legal counsel at erosion MC and Erasmus MC holding. And I'm specialized in corporate law, actually, but I'm also having a background in valorization and it. And today I'm working as a legal affairs of Erasmus MC after I changed departments after five and a half years working for the Technology Transfer Office in the first place, I've been asked to give you a presentation about the advantages and risks of using generative AI tools for our own profession. And how they could benefit you, and I will give this presentation from an independent point of view. This means I'm not going to sell you anything today, and it's up to you what you're going to do with this information. I want to start off this webinar with a discussing Dutch court case. This court case hit the Dutch headlines because a judge in the verdict admitted that it was using a judge GPT to collect information while deciding on a case. The case was about two neighbors, and one of the neighbors had solar panels on the rooftop while the other neighbor was extending his house, resulting in the fact that the other neighbor had a reduced yield of his solar panels. Of course, they get into the fight, and our so called solar neighbor was kicking off isolation material off the rooftop of the other both were claiming damages. And during the case, the Dutch the Dutch judge well used jet GDP to estimate the lifespan of the solar panels, but also to determine whether the damage damaged isolation material was still usable. And this case was creating a lot of fuss in the legal world of the Netherlands last year, and I want to, well, I want you to have this case in the back of your mind during this webinar. I'm not going to do to dive into the scientific part of the artificial intelligence today, but I want to keep this as practical as possible. I will start off with some introduction to AI and some history and some general facts and some, yeah, general applications. And then I will provide you some advantages of generative AI tools, but also some examples of tools and prompts, and I will finish today with some awareness sessions. To start off properly, I need to discuss the question, what is artificial intelligence? Artificial Intelligence is a technique that imitates the human brain as much as possible. You can think of functions like knowledge, interpretations, judgments, reasoning and decision making, but all experts are agreeing on one thing, and that that is that an AI model does not have a free will, and that's very important when it comes to intellectual property, for example, which we'll see later on. Interesting of this slide is also that this image of on this slide is also produced by a generative AI tool, which is one of the first advantages I can mention for us professionals, when you want to give a presentation, you. Have to see or have a huge database with stock images, you are able to ask generative AI to generate your own pictures for your slides. Of course, there's no guarantee that you won't infringe other third parties intellectual property rights, but I think you will limit the risk, at least here you can see a little chart mentioning the history of the development of AI. Although AI, at the moment, is very hot and happening, the development of AI is already started at 1956 by the scientist John McCarthy, not to be confused with a member of The Beatles. Of course, you can see that it's deviated in three waves, and each wave has peaks, which are called summers, and the valleys are called winters, and the main reason to enter into a winter are having too much of too high expectations, frustrations, which also lead to lack of interest and also lack of funding. At the moment, we are in the third wave, and it's a little bit dedicated to generative AI tools. And the big question is, for how long will this third wave, this summer will will last? And my personal opinion is that it will last a little longer. You know, this chart is ending at 2021 but at as today, 2025 I don't think this summer will end during this year, if we take into account the investments that the government of Trump, but also the developments within China. I think we're still not at the end of this summer,

Speaker 2  7:16  
artificial intelligence is already everywhere in our daily life, even a situation that we don't even know it. Artificial Intelligence is is capable of managing trips from booking to final destinations, but also to assist pilots how to fly their plane, but even to purchase flight tickets and monitoring flights to their final destination. But it's also able to decide whether people are eligible for a mortgage, for example, or suitable for a vacancy for a job. Another example is, of course, the application of self driving cars. Artificial Intelligence allows a car to learn on different types of roads, to drive on different type of roads, and when you enter a roundabout, it won't drive straight ahead. A more relevant application of artificial intelligence is, of course, contract management. It's a little bit legal, legal wise, but it's also applicable for tech transfer in general. I guess below in the corner you will see it a little chart of the whole process of negotiating contracts and AI is capable of handling all elements of that process. It's capable of drafting contracts, comparing incoming contracts based on your own templates, but also identify risk and recommend suggestion based on the same templates that you already have on your hard drive. It's also capable of managing contracts even after execution, to make sure that a contract is expired or will be expiring, and even to judge whether milestones within a contract are met or not. I also have put some examples in the corner of this slide. I don't have any link with any of these branches, but these are providers that could provide you a system, contract management system using AI. Here you can see a chart of companies, of the number of companies using contract management involving AI, and you can see that. That it was peaking in 2000 22,017 and 2018 and this all is within the summer of the third wave. But you can also see that around 2021 we already entered into a winter. And I assume that this the reason before, because we ended in the into this winter, could be that when we want to have a contract management system involving AI, we intend to have a closed environment. But the downside of having a closed environment is that the development of the model will stagnate big time, with which will lead to, well, to high expectation for the next customer. So I'm not sure whether we already entered into next summer. For example, this chart is ending at 2021 and it's four years ago, and I have to admit that Erasmus MC did not implement any system of contract management involving AI, and we're still investigating a tool like that. What are the main advantages of artificial intelligence. Well, AI is capable of searching through giant databases in minutes. Well, if you are capable of searching giant databases in minutes, even when they are in foraging languages or in a non standardized format, you automatically will reduce time, and therefore also having less cost if you have, if you have less time, if you have A lesser I mean, if things will cost less time. You can also include more data, and when you can include more data, then it usually will lead to more accurate results, depending, of course, on the quality of the data that the model is trained with. Another example of AI is reducing a human cost error, which means that AI will not quickly count records of the database twice, while a human being is capable of more than capable of doing that. So now I want to give you some examples of the advantages. This is an article of the magazine amazing rasmuham, see. And that's a magazine that we use to publish innovations and innovative activities. And this is a example of an AI model that predicts the risk of skin cancer, and it really outperforms the current methods of doctors. Here I got an example of the Brazilian government that really wants to hire open AI to cut costs and to let AI perform some analyze cases to save money and time. Another example of outperforming human analysis is an article from the University of Chicago, and they have shown that large language models like chat GTP could perform a financial statement analysis better than the financial professional. And this article I bumped into last week, actually, there's an article of Dan kingsma, and it's titled like, hey, JDP, please write my plea. AI is a rival is arriving in Dutch courts. Well, it's not all about Dutch courts, because if you look into the paragraph that I put below, and you'll see that multiple judges from different countries are using AI chat GTP to speed up their processes. So you can see a Colombian judge are used is using chat GTP to write complete verdicts. A UK judge is using chatgpt for summarizing an area of law, and it's also being used in China, I see. Now I want to provide you some examples of generative AI tools. Here you can see eight examples of applications that. You can use, of course, this is this is not limited to these examples, and there are many more examples of applications you will see later on. Here I'm providing you some examples of tools. Again, there are many more tools that you can see within these applications. So, yeah, feel free to Google any other applications. Well, in the first window, you see the large language models and the famous one, GDP cloud from France, and also Microsoft co pilot and Dall E in the second window is also very famous. I guess. What you see here is not, are not really complete. AI models that are built from scratch. There are multiple applications here that you can see that are using other models. For example, chat GP we take copilot. For example, copilot is using chat GTP as a large language model. When you ask copilot anything to write, if you ask co pilot to generate an image, then it will use del E, for example. And so there's more to it than only, yeah, it's lending its own models. It's hard to explain, but it's more a shield that will help you prompting to get the best result.

Speaker 2  16:49  
I also want to draw your attention to automation of generative AI, automation is, yeah, with automation, you can speed up and automate daily processes by combining conventional tools and generate AI tools so you will be able to automate to create automated flows, examples of providers that enables you to automate or make Cassidy and perhaps also open AI assistant, but you might need some coding knowledge for the last one. Also here accounts the same. There are many more providers than that would enable you to automate processes. In the back of this slide, you see an example of make.com and I have to admit, if you look in are you, when you look carefully, you will not see an AI tool, but you will get the picture that you can combine these AI tools as well. You can see that there's a flow filling in PDF form. It's also connecting to one drive. It's using a Word document. It's able to send an email and, etc, etc. And by automation, you can create your own agent that's able to carry on, carry on inquiries on its own.

Speaker 2  18:29  
Now I want you to provide you some example prompts that give you an expression impression on how to use generative AI tools for your own profession, and I'm using different type of tools, and I will provide you the tool that I used in the back in the corner of The slide. The first example of a prompt is actually a wrong one. I'm asking chatgpt to draft me an assignment class for a license agreement. Of course, this is not a result that I wanted. What I got was a non assignment clause that could be used for any type of contract. But of course, I was looking for an assignment class to assign intellectual property to a licensee, which is also a startup, to enable that startup to to attract investors. And an investor usually wants to invest only into, preferably in startup companies that are owning the intellectual property. So the lesson that we learn here is that you need to write your prompt like you're explaining to a human being. For example, your colleague, if we take another shop, then you can see that we will get the right answer. I provided chat GDP with some extra details, like to assign intellectual property an investor, but also a threshold of 2 million. And then you can see that we have the right result here, and the class that it generated is continuing down the screen, but I could not fit the whole class readable on the screen. So it's very important to provide your tool with all the necessary details to get the best result.

Speaker 2  20:46  
The next example is a real example that I had with a colleague of mine. Colleague of mine, we received a legal inquiry of an internal client, and we did not really understand what it what it really meant and what they really wanted. So we had a meeting with the three of us, my colleague and I and the client. And during that meeting, the penny dropped. And while my colleague was still chatting with this with his client, I was asking chat GTP whether it could draft a disclaim for this particular situation. After the meeting was done, my colleague asked me who would take up this file, and I could, and I was able to send a disclaim like this. This is not really the disclaimer we're talking about. Anyhow, she approved the disclaimer. Only changed three words on it, and we could close the file within 15 minutes after the meeting. So it's very important to put background information into your prompt to get a more accurate result. And in this case, I provided jet GDP that the disclaimer was about medical quality documents that we wanted to share on our website. Other institutions could use them, but they only were able to use it for their own risk. And of course, we want to exclude our own liability, etc, etc. And it was a result. Another trick that I could provide you is that you could try the same prompt with a different tool for better results. If you use this prompt in for example, co pilot, you will get a different structures result even though co pilot is using chat GTP as his large language model, but it's the same. Elements are there, but it's differently structured. So in this case, I prefer chat TP because it's more copy, paste, double for a website. While co pilot is well differently, providing you the answer. So I was so we were using the result of chat GTP. The next example is an example of summarizing PDF files. I was using tenor shares, AI PDF tool. And I found a book of the business model nature and benefits of the Harvard University, and I asked this AI tool to summarize this PDF in very understandable language. And this was the result. I actually wanted to remodel my prompt by adding something like that. I want to have a result in bullet points, but unfortunately, I only had one free token, so I could not really adapt my prompts again, but I didn't want to withhold your of this result for you.

Speaker 2  24:11  
The next example is about the shopping basket, which is already expired from Amazon, and I asked Amazon to write me a teaser on the patent and also how this patent influenced competitors. This patent is, of course, interesting because it's a famous patent for software which is not always applicable patterns are not always applicable to software, and this is why this is a famous one that was granted but also expired. I'm very satisfied, or at least a little bit satisfied, with the result I got, but I actually wanted to have. A one pager for a business developer. So the business developer could use this one pager to attract potential licensees for an innovation for example. But as you can see in the last sentence of this result, how do you feel about the impact of such innovation on your shopping, shopping habits? It really misunderstood the word teaser. I think it's my fault. I had to elaborate a little bit more and better that I wanted to have a one pager and what purpose I wanted to use the result, but you can assume that this last sentence is a translation of a word teacher, so that's something that you can learn from. And yeah, can do better next time if you want to prompt something like this,

Speaker 2  26:01  
another example is very legal wise, and that's that you can that you can ask AI tools to make classes a receipt protocol. And here you got an example of a liability class. And in this class, it's all about the licensor liability, while in the second class, it's the like about liability of either party. I think you get the picture for this example. Now, every time I do this presentation, I'm generating a promotion video, and in this time I was I got aware of the fact that the autumn webinar committee was seeking for new members. So what I did was providing the AI tool in video, a title, a language. I copied some general information of autumn into this prompt. I also provided the tool with some information about the task of the webinar committee, and I finished off with some instructions. And as you can see, I only used 922, characters, while I could use a lot of more. And I have to admit, I used a free account so you could see still some watermarks, but it only took me a few minutes to generate this, this video, and I'm going to try whether I could play it. That's not going to work. I have to quit the PowerPoint presentation, and I

Unknown Speaker  28:04  
start the video tech

Speaker 3  28:05  
and love sharing knowledge that this opportunity is for you. Join the ATM webinar committee. Be part of a UTM global movement. As a committee member, you'll define cutting edge webinars. Connect with top experts and inspire 1000s dive into emerging tech global collaboration and inclusive innovation. Apply today and drive the future of technology transfer. Don't miss out. Join the revolution at if you are not interested in joining the apt webinar, you can always provide webinar proposals to the committee to organize. Feel free to reach out to the organization of a UTM for more information.

Speaker 2  28:56  
Okay, I hope you could all we're all able to see the video, and like I said, it only took me a few minutes to generate this video with the prompt that I just showed you. And of course, you afterwards you can add some logos and change, well, a lot of things. But like I said, it's only took you, yeah, it will took me only a few minutes to generate this video before I continue with awareness session. I want to say that I did not provide you an example of generating pictures because generating images, or pictures is a true art. I think I could give a whole presentation about generating images, because there's a lot of technique coming with that you can you could put things within a prompt of. Between brackets enable you to underline some aspects of your prompt, for example, if you want to generate a blue bike and a red car, it will enable you to make sure that their car is red and not the bike, for example. And it will take people quite some time to generate the right picture and having the right prompt for all their for all the future pictures. So it's a true art, and well, perhaps you could dive into that for yourself. And now I would like to continue with the last material part of this, this webinar, and that's about awareness, I will touch upon topics intellectual property, confidential information and privacy, but also with the quality of data. When it comes to intellectual property rights, I'm going to ask four questions, and the first question is the limitation of using results generated by AI tools. Well, the ability to use the outcome of a AI tool is really depending on terms and conditions of the AI tool. As you can see in the terms and conditions of terms of use of chatgpt, there is a rule that states that you may not use output to develop models that can compete with OpenAI. But there's also a statement that they will assign everything to you. That means that the ownership of the output is yours, and I think that has something to do with the responsibility that comes with it. And now they assign everything, including the responsibility, to you, but be also aware of the fact that AI X could limit the uses of AI as well. For example, in the European AI Act, there is a limitation in using deep fakes, deep fakes that requires you transparency. In case you are using deep fakes, you need to ensure that for everyone, it's clear that it's about deep fake another real video, for example, could a generative AI tool qualify as an owner well and generally not? Like I already said, an AI tool does not have a own free will and the conditions of both copyright laws, but also patent laws, require human aspects. And a famous case about about the human aspects of intellectual property is the debit case. In the beginning, there were two countries that granted a patent application, and I think from the top of my head was South Africa and Australia. But in the end, there, they were all declined based on the fact that it's missing a human component, which is required in the patent law.

Speaker 2  33:30  
Could you influence third party IP rights? I think yes, you can. There's no guarantee that you cannot inference any intellectual property right, although you will limit the risk by generating, yeah, images, for example. But it's also, it's all based on trained data. So there's also a possibility, always a possibility, that you will influence third party rights, and you need to compare, like any other work, and the normal legal IP as concepts are applicable, could prompts be protected by copyright? Well, it's same counts here that the normal legal IP concepts are applicable. So it could be that your prompts will be protected by copyright. But in most countries, the copyright law requires an original, creative and personal stamp of the author. So if you are, yeah, if you are very creative in your within your prompt, then it's possible that it could be protected by copyright law. And I think it might be applicable when you are prompting for images, for examples, which, like I already said, it's a true art. The next awareness session is about confidential information and privacy and Well, I think. Common Sense. I really thought that the first time I gave this presentation, you should not put any confidential information or personal data into public generative AI tool. However, two months after I gave this presentation, for the first time, the Dutch authority announced that a general practice fed private information about patients into a chat GTP based program, which was then stored on a tech company service and potentially used to train the software and well. So what we thought it was common sense. It happens anyway. And if I ask you whether you should put confidential, confidential information or personal data into a public generative AI tool, you will all answer, of course, we won't do that, but it when does this go wrong? And it really does, like you already showed you, it goes wrong in this in the situations when a employer does not provide its employee, employees a closed environment for using AI tools, while the employees really have a strong desire to use these AI tools. A personal experience from a colleague of mine is that a colleague offered to translate an agreement or a document or an IDF form into a foreign language or other way around which well, of course, could cause a threat of trade secrets or any anything. And another example of of a risk is that you could that you are willing to summarize a large document or want to create minutes of a lengthy board meeting that really needs to be kept secret, and you will put that into an AI tool. These are the common situation that I can think of when this still goes wrong, although we might think it's common sense. The last awareness session that I want to talk about is that the basic rules that artificial intelligence is only as good as its data. Be aware of the fact that the quality of data or the lack of it could cause a bad outcome, but also be aware of the fact that the data could be biased, or the model could be biased because it's trained by one sided information. Every story has two sides, but if only one side of that story is active on the internet, you will get not the full picture or the right picture as an outcome of the two. It can also be that the information or the data that is put into the model was wrong in the first place, and that could also lead to ethical risk. And the last remark that I want to give you is be aware of the fact that you don't know how the model got his answer. The Black Box is actually complex decision making with a lot of parameters, which makes it impossible to trace back, or almost impossible to trace back, how the AI model got his answer. I

Speaker 2  39:03  
now, if we think back to the case and knowing all the advantages but also the risks of using AI tools, do you think it was a good idea for the judge to use GPT for collecting information while deciding on an actual court, court case, my personal opinion was, well, the first time I heard heard this headline, I thought, What is the problem? I think AI is very beneficial for any professional, including judges. However, if I when I read the articles and the FOSS around it more closely. Then it seems to me that the judge was relying on the AI to blindly, and it's even unclear whether the parties were able to object to. The interpretation of the judge of the information, that of the information or that it got from the AI tool. And in that sense, I think the judge should have known better. And of course, you can use chat, GTP or any other AI tool as a handy tool, but you should not rely on the AI tool blindly.

Speaker 2  40:30  
Then I want to finish this webinar with three websites to pioneer for yourself, and I'm fully aware of the fact that this might be something for our software development minded tech transfer professionals among us, but let's decide for yourself. The first website that I want to share is huggingface.co that's not a typo, and it's an AI community with all kinds of Yeah, it nerds, I guess, sharing models, data sets and applications that you can try online, but even download on your personal Computer yourself, and you can see that there are lots of different types of applications available. The next website I want to share is LM studio. LM Studio is a program that you can download for free, at least for personal use, which enable you to run AI models offline, or at least locally, on your personal PC and in a closed environment. Like I said, it's free to use for personal use. If you want to use it for professional use, you need to contact them, and it might require a different user license. I have to admit, I did not try this one myself, but, well, it's at least something to to worthwhile to mention. The last website is w3 schools, and it's actually a website that will learn, will learn you to code. You can see that you can learn to code in many different languages, including Python. You can see it in the black bar in the top of the website. I code in Python myself. And the beauty of this website is that it has its own online IDE, so you don't have to install an IDE on your personal PC. You can learn to code online, and you can run the program online as well. But the best part of this website is the section that I copied here on this slide, and that's the section about machine learning, and it really provides you a introduction in artificial intelligence, or at least, machine learning, and how does it work, and how the decision making of an AI model works

Unknown Speaker  43:19  
so well you might need some basic coding knowledge, but like I said, you can learn this from the same website.

Speaker 2  43:35  
This was my presentation in advantages and risk using generative AI tools. Are there any questions? And I have to admit, I cannot see any of you, but perhaps I can stop sharing the screen so I can see everyone i

Speaker 2  44:15  
i see One one question about the slides, and I think whether the slides are made available, and I think that's I have the intention to make them available, but I think that's autumn has the same intention.

Speaker 1  44:38  
Yes, the slides will be made available after this webinar.

Unknown Speaker  44:43  
Okay, any other questions?

Unknown Speaker  44:58  
I don't see any. I.

Unknown Speaker  45:05  
Okay,

Speaker 2  45:13  
I see one question. Any suggestion on best tools for contracts, analysis, for example, inquiries not generative for drafting language. Well, like I already said, we are still investigating contract management systems for ourselves, so I do not really have a good suggestion. At the moment. So no, I cannot provide you one. I

Unknown Speaker  46:08  
Okay?

Speaker 1  46:18  
On behalf of the autumn, I would like to thank Lawrence for the informative presentation on today. And thank you again to our sponsor, Marshall Gerstein. A recording of this webinar will be available for viewing in the autumn Learning Center within a week of this event and is included in your registration. Please be sure to complete the webinar evaluation, which will open immediately when you sign off this session and thank you for being a part of today's presentation, thank you again for joining us, Lawrence, and have a great afternoon. Everyone. Bye bye.

Unknown Speaker  46:50  
Thank you. Bye bye. You.

Transcribed by https://otter.ai