The following podcast was recorded on Clubhouse as part of the 4x4 Innvoation Meet — Turn Ideas into Money series, moderated by Martin Schweiger. There were a lot of interesting aspects of how Intellectual Property (IP), such as patents and copyrighted items are being disrupted by the rise of Artificial Intelligence (AI).
It’s not perfectly transcribed which is why I recommend you listed to the podcast instead which has some really interesting twists on machine vs. human made innovation, and, whether innovation actually is even something new, or, simply a combination of things that have already been there — which would mean that innovation eventually is finite.
Martin: [00:02:19] Then. We don't want to lose time because I have two precious experts here. Today is Kim and Toby. Toby has long standing experience. The mission inventor, polymath who is used to who even sold chat bots for websites with his team. And I remember I spoke with you a few years back about Wright applying, and I still remember today it wasn't I was in a supermarket and you gave me 30 minutes lesson, which I will never forget in my lifetime because it was so insightful and with really hands on tips on how you should then use the chat bot because the goal was I wanted someone, a concierge from my website with a few hundred articles. So if people are there and looking for contents to, to guide them to write article. And yeah, what I understood is it depends all on the training of the chat bot. That's what I've taken with me from at the time. But surely now we also have some leaps in technology and that's where Kim comes in, because Kim is a professor for natural language. Language processing at the universities are clear in Paris, the biggest, largest university of Europe. Kim Right or not and.
Kim: [00:03:50] Big it's rather good but not we're trying to be small and good rather.
Speaker1: [00:03:56] Okay.
Martin: [00:03:56] Yeah. So I would say the people who are here, the CS Portie from India, Joe from the US is a talented engineer and then we have David, a patent attorney colleague.
Speaker1: [00:04:11] Who who.
Martin: [00:04:12] Talks like somebody from Scotland, but he says he's no longer from Scotland, married with a Chinese wife. Kim, that is you have in common. Kim is also married with a Chinese wife. Rich is from Chicago and there's a long standing expert when it is about managing innovation. And yeah, I believe that's just a nice round. We have Michael Ostrovsky coming in and our procurement manager. Then Karl Powerlifter, like myself from the Blue Ridge Mountains area in the east of the US Southeast. But please, Toby and Kim, the stage is yours and please entertain us. I can ask questions if you get very boring.
Toby: [00:05:01] So, Kim, do you want to go first?
Kim: [00:05:05] Yeah, well, I'm not really used to how these things work on clubhouse, so maybe. Martin, you want to ask a question, then we can. Yes. Dive right into.
Martin: [00:05:14] It. Yes. Kim, I catch you today on the on the wrong foot. I just heard that Google lost 100 million in value because they have a chat bot based on the GPT technology. And you will explain later what that is because they did some some customer demonstration in public and it went wrong. So what went wrong?
Kim: [00:05:37] Well, the there was just the fact was not right. They just asked something who was who photographed the first exoplanet. And it was a bit US centric saying it was the naysayer, but actually it was a European centre. It's not I mean, it's not that important I guess. But to that this came just up in the presentation made it look less good than what the church has been doing before. So people were disappointed I.
Martin: [00:06:07] Yeah, I can. That was the first thing. Because actually the one who introduced.
Kim: [00:06:12] The to.
Martin: [00:06:13] The chit chat chip was, was you. Right. And then I started to experiment and then doing other things. And what I found is it's very unreliable.
Speaker1: [00:06:22] To.
Martin: [00:06:23] To, to work with it when it is about facts and retrieving facts. But the PD is very good when it is about converting a text in one format into another format. Let's say I found if you say explain this, this very difficult legal text to a 12 year old and the outcome would be something that can really be understood by anybody on the street. And and in my firm, we are using that I'm very honest about for replying to emails, inquiries and also to. So we are using that actively in order to improve our language, make it more simple and to be I mean, Kim is it, is this why, why is it so so why is it so bad when it comes to facts and why is it so good when it is about making something that looks nice?
Kim: [00:07:18] Well, first, I wouldn't I wouldn't really agree that it's really bad. In fact, it's bad in some fact. It's not reliable. But it's really good in most facts, actually. So historic questions, technical questions, questions about code. It's quite impressive. So the question is always the reliability. But this is a problem, a more general problem that if you go in the Internet and if you use Google to look something up, you have the same problem of finding some half truth and fake news and whatever you find. So it's a it's just and this is what GPT three has been trained on. And so it got some something wrong and sometimes it just combine things in the wrong way. So I think it's it's not necessarily worse than looking, looking things up on Google because I mean the only advantage is you can somehow decide what the really the weak point of of this system today is that it doesn't cite its sources. So you cannot say, okay, I trust it because I like to trust the source. And this is something that Google is attempting to do to to somehow counter this this project of an AI by saying, okay, we're going to do the same thing, but we're going to give for each sentence we provide we give links to the exact web pages where this has been found, and you can verify for yourself if you if you want or if you don't, then you just keep it this way and you have the same problem as before. You just don't know exactly how reliable the results are.
Toby: [00:08:48] Well, I think just just to add to that, I mean, we we had access in one of my portfolio companies. This is the chat bot company, Martin, that you referred to earlier, UIB in Singapore. We had access to this chat, GPT in form of open eyes, GPT three Library for well over a year now and we've actually tested it thoroughly. And I think, you know, when it comes to facts, they're really three components of, of, of facts at least this is the way I structure it in my human mind. Right? There are facts that you know, because you have you have some personal experience with it. It's not not a theory. I mean, it's like you have very personal, subjective experience. Ideally, objectively, you have some validation from a larger crowd as well. And then there is theory, which is like in the Internet, you read something and you can get infected one way or the other. You might be right, you might be wrong, the fact checker might be right, He or she might be wrong. Who the heck today knows, right? So the second the second law at the end of the day is is essentially what what chatbot today is fed with historically. Now, sooner or later, this is going to be real time. But anyway, whoever programs an AI or sort of an AI also has has a human side and has a certain bias, whether that's politically, whether that is religiously, whether that is just a mood. At the end of the day, nobody is perfectly neutral and nobody is all knowing. So at the end of the day there will always be some form of impact on the AI, as long as it's developed by humans, at least on on whether whether it is how can I put it nicely, whether it's truly, truly objective or not.
Toby: [00:10:45] So what we came up with and this is my third point in experimenting this one year with with chatbot, this B2B sort of solution is we realised that if you point Chatbot into a specific space, it is far more likely to give you. Correct answers. And let me give you an example. If today I would say give me that rubber, you know, depending where in the world you are and which English you have sort of grown up with, you might think of a rubber being a way to clean a pencil from a piece of paper. You might think of it as being a condom, or you might actually think of it being like a chain of auto houses in Croatia, changing your tires, or you might be an actual rubber plantation somewhere. So my point is, if you ask something with regards to rubber to chat TBD without telling it a context, you know, it's very easy to fool it, or at least you think you have fooled it. But a human is no different, right? I mean, when when I tell a friend in the US of of a rubber, he thinks of a condom. If I tell someone in India he thinks of the pencil example. So where I'm coming from with this example is we realised in this one year that whenever we submit a question to to church deputy, we do it in the frame of a We basically tell it to look in its own brain in a certain area so that we get a contextual reply to that question, which is more likely correct. And, and this has helped us tremendously. We call it our mind space to actually make better use of the tool. Does that make sense?
Kim: [00:12:39] Yes. For me it makes complete sense. And it's it's a bit like if you give more context, like every human will also give better answers. And so you have to set up this this context. And this is also called prompt tuning, right? That you have to ask the right questions in the right way and you get better results in specific ways. So people are working on this now to actually automaton this, to somehow use a fixed language model and then generate with possibly different model, different prompts, and then find the best combination of words that actually gives you the desired results. And it has the advantage of making use of really big language model without fine tuning is what is really expensive. So yeah, so in a sense it's the same thing. You give the right context, the right words that the system knows what it's supposed to answer.
Toby: [00:13:32] Absolutely. Absolutely. And it makes a huge difference. I mean, you know, why do we expect an eye to do something that a human couldn't necessarily write? I mean.
Kim: [00:13:43] Yeah, exactly.
Speaker1: [00:13:43] Yes.
Martin: [00:13:44] Yeah, we have tried this a lot over the past week that came together when it comes to applying the technology to drafting patterns. And yeah, I mean, it's a bit like you ask a student, we give him two to a four pages to read and then you ask him after 10 minutes. Now I ask you a question about in the about what I just gave you to read. And that's not much. Intelligent, I would say is a bit the intelligence is there that you within a very short time the technology can answer questions based on what you just gave you to read. So I wouldn't even talk about intelligence in that way. Is is more is the local training data that you create. Is it helpful to answer the question that you're asking after that and that that that confirms also what I think that there is no artificial intelligence because when when you ask the GPT to create something, then very often nonsense is the result.
Kim: [00:15:04] Well, I don't really understand. I mean, I think it's a common historically a common behavior that humans consider everything that the machine can do as non important or trivial or non intelligent. So we have gotten used to calculators that compute numbers much faster than us. And then we decide that it's not intelligence, even though before we thought that kids are smart when they can multiply it to numbers very quickly. So and I think.
Speaker1: [00:15:32] Now.
Kim: [00:15:33] The same thing may happen that we are saying that the machine can can do this. So by definition, it's not intelligent because we don't want to call a machine intelligence. But I think that's somehow tautological. I think, of course, these machines are intelligent to a certain degree, and I think the subject should and has to be what the consequences are of intelligent machines and not just talk us out of it and say, no, no, it's not intelligent. I mean, resuming a text of two pages, every schoolchild can do that, and that's not intelligent. Well, firstly, no, not every schoolchild can do it. And secondly, even if every schoolchild can do it, they are intelligent schoolchildren and its intelligence, I mean, and has consequences for our world if the machines can do that.
Toby: [00:16:22] I think the key point I would like to raise is, you know, taking an existing content and and rewriting it, whether that is from an already written text or from a database from a structured or non structured for that matter format. And putting this into into a new light, like into a new style. Right. Or into a new, how can I say in a more jovial, sound sounding poetry, let's put it this way, right. But without losing the essence of of of the of the of the factual content, probably a machine will be far more better at this than than a human. And the reason I say this is you can almost through algorithms, mathematically create what is perceived as being jovial, as lighthearted as, you know, a serious by simply inserting some words or changing the grammar. I mean, at the end of the day, the languages which have very, very strong and clear grammar, you know, they it grammar is like mathematics for languages in a way, right? So I mean that's why I think machines will be extremely good at that.
Martin: [00:17:42] Now try isn't, isn't this, isn't this a contradiction. Sorry to interrupt you because when you talk about grammar grammar, that that's a bit rule based. And what I understood is that the technology is just not rule based, but still deterministic, but in a different way.
Toby: [00:18:00] Right. All right. I'm not I'm not disagreeing to that. All I'm saying is the the natural affinity between artificial intelligence and generally I'm just talking not about the technical elements of it, just on the on the philosophical closeness of the subject matter. Right. The the the affiliate AI and a content is there but portrayed in different colors is is very close but just try to find a completely new point of view. This is where it gets hard a completely new point of view, a completely new way to interpret something or a completely new how can I say I mean completely new point of view that was not there before to say just a poem for, for example, that is going to be very hard for you to pull.
Martin: [00:18:55] Poem. Well, Kim.
Kim: [00:18:57] Yes. So I well, I made at some point we worked together, Martin and me, and he had to head off to to his workout. And I told him, well, asked me to write a motivational poem for your weightlifting. And he he did it. And I think it worked. Worked so well. He was no. Martin You were really strong this day because you got such great motivation from this wonderful.
Martin: [00:19:24] Poem, actually. Was that the result was. It was a poem. Yes, it was a poem. And I'm thinking about where did this machine get the poem from?
Kim: [00:19:34] So it created.
Martin: [00:19:35] It. That's what you say. But maybe it founded.
Kim: [00:19:40] Well, you can Google it. Google is three or four words and you will see that it does not exist on the Internet. So I think it didn't find it. It combined some stuff to make it.
Toby: [00:19:48] I think it combined some stuff to make it. And and maybe it has some idea of of when it combines A with B, we see what is the likelihood of of people liking this or perhaps even this making sense. Right. But that's not look let's let's I I'm a musician first of all I compose music. You know I I'm kind of a renaissance man, old school guy, right? I mean, so I, I would like to take a stab at this from a philosophical point of view. To me, every matter, whether it's content or whether it's an object or whether it's a company or a building behind everything, there is some form of you could call it inspiration or you could call it spirit, or you could call it, you know, the emotions that people have and creating something of of of value. Right now, I think we can agree that at this juncture, at least as much as we can say, and I does not feel any emotion while it is creating something. Well, we can debate that maybe in a different in a different clubhouse. But the the point that I would like to raise is if a human actually feels an emotion, it's it's real, it's there and somehow it enters its creation might be question mark is maybe there are people who don't sense that behind an AI created work whether this is there or not and other people they very much do sense it. And from a purely, purely from an artistic point of view, maybe it's very hard to distinguish between these two eventually in the future because it is like an invisible world. How are we able to tell? Right. But I do think that there will be a time where if somebody can prove that it was an authentic human who created something, it might have a different price tag on the art that is being sold then if it was a machine.
Martin: [00:21:53] Hmm. That goes deep, Toby, because you're talking with a patent attorney, and we are always discussing is a certain invention that is characterized by a number of features or elements. It's like a filter. You create a claim, a patent claim that you claim. This is all all, all objects that fall under this filter are protected by my monopoly. And then the question is always because in this world there is no no true invention, because all the inventions that I see, there are always potentially combinations of what exists out there. And then the question is whether or not the combination that you do between two earlier disclosures is obvious or not. And the the main the basic criteria for that is, was there a motivation to combine the two? And if not, this is the threshold for inventiveness in patent law. And probably the same applies to melodies. I don't know who is the why is one an artwork, one one music piece and artwork and the other one is not?
Toby: [00:23:09] That comes back to the spirit behind. Right. I mean, at the end of the day. And how do you want to check check on this from a from a from a pure patent or legal point of view? I mean, I have personally written 19 patent families. So I'm also and I do advise a lot of inventors on on IP. That's what I'm doing in Switzerland now professionally. But putting that aside, I totally agree with with with the challenge here. Right. All I'm saying is, is not it's not on the on the jurisdiction or the or the legal side, but more on the I can see an audience that is willing to bid for a pure I made a piece of art, let's put it this way. And I can also see an audience that is willing to bid on a. Piece of art that is 100% authentic human. And that I mean, how to prove it. That's a different story. I mean. Right. Because, you know, these worlds overlap. But, you know, even if the ideas are matched in the mind, but the pure execution and then then it comes, then some interesting questions come along, like did you when you wrote this poem, did you use it for any of the two lines out of the.
Speaker1: [00:24:27] 200, you know.
Martin: [00:24:29] Anyway, Yeah, yeah, yeah. If you draw patterns, did you use any to, to, to draft a paragraph about the advantage of a certain pattern claim.
Toby: [00:24:39] I did not. But have you checked out patent Pell.
Martin: [00:24:42] Yeah, of course. This is, you know, the, the Kim and I, we are part of the. The robot robotic patent drafting family in this world. We had a conference in September, World conference, and we were all there. And Chuck Sue, the owner or the founder of Patent PAL, is our friend. So we chat with him on not daily, but maybe weekly or every two weeks or so. Yes.
Toby: [00:25:11] So what do you think of it? I liked I mean, of course you're a little bit biased, but that's fine. I just I mean, when you say bad day.
Speaker1: [00:25:20] I mean.
Martin: [00:25:21] I hope what I can tell you is the following. Currently and last week we have drafted Human I, we have drafted two patents together and one of the two patents. I gave the input data, the same input data to check the patent pal and we can compare the outcome. And what do I think of it? Yes, it works it it would give you it would give you something that looks like a US patent. Yes.
Toby: [00:25:53] Mm hmm. Mm hmm. Mm hmm.
Kim: [00:25:54] So, yeah, I think the main point to take away is to come back to the combining. The two questions of being able to draft new patents and creating symphonies. Is that the result as such as indistinguishable? And then the question is what to do with it? I mean, people say, well, I heard it, but for example, there have been many experiments about experimental machine generated music. Probably Toby, you know, about this and also classical music. And they played this to people telling, not telling them that it was they said it was made by some, I don't know, obscure renaissance I don't know. Writer And, and then later they told them and first they asked them which they preferred and so on it very often the results are very simple, have been has been reproduced many times people cannot distinguish the music machine generated music from or generated composing from the the human generated composing. But they are really angry afterwards when you tell them. So they feel that they have been cheated. But that's about the spirit thing we talked about and it gets a bit it gets a bit weird when you talk about this, right? Because somehow why do they feel cheated? Because they like the music for us. So and I think similar question applies to the threatened.
Toby: [00:27:14] I think they also sorry to interrupt you, but I think it's not only the cheating is not that's one thing. I mean, nobody likes to be fooled, but I think there is also that element of shit. If it can do this, what else can it do?
Kim: [00:27:27] Yes, maybe just like some.
Martin: [00:27:29] Maybe replace myself. Right. So.
Kim: [00:27:32] Exactly. There's some some basic fear popping up here. So and with patents, I think it's the same thing. I personally think that that as Martin just said, that all ideas are combinations of existing ideas to some extent. And I personally believe that the space of ideas is finite for different reasons I can explain to you. But so I think that this space is getting less and less empty, so there's less and less ideas to to have simply by for reasons like combinatorics. I mean, if you can put the idea in, I don't know, one page, then these pages, many, many of these pages have been written so they are less available and the same as paintings. If you have a painting and you say, okay, I can take it to whatever resolution of pixels, 1000 by 2000 pixels, and you know how many paintings are possible. If you also have a measure of how what is distance between two paintings is if it's just the same very small difference in humans, I cannot perceive it should be considered the same. So if you if you consider this.
Martin: [00:28:39] I want I want to kill you for this. But finish first and then I kill you.
Kim: [00:28:44] Yes, but I think that. Well, I don't know. It kill me first and then we come back. I mean. Okay. What? What's wrong? What?
Martin: [00:28:51] I said it's wrong because I come from the area of anti-counterfeiting and I know the tools that I used there. And, you know, the QR codes and the QR codes, they are also finite. There's a finite number of QR codes. Qr codes are the square ones. Yeah, right. Yes. And what what the experts told me that that they can generate more QR codes based on the current technologies and the atoms in our whole. Is it universe or is it in the whole whatever exists out there? So you can.
Kim: [00:29:25] Label, but it doesn't contradict what I just said. I mean, maybe there are still many ideas left to have, but yes, it's yes, but it's not infinite. That's important. I mean, even the idea of creativity, I mean, no, we're not really talking about chat bots anymore. We're talking about machine creativity. Right. And I really like this field, but I really think that's great. I mean, the term creativity is something very recent. It was the 19th century before people did not believe that there was anything like creativity, and many cultures today don't believe it. So this idea that one human creates like ex nihilo, something that has not existed before, it's a very like 200 years old idea that has not really stood the test of time before. They are people thought about muses doing this or some. And so it's a new idea. And I think it's it's time to get over with it, because I think that it is an idealization of some creator that does something. And I mean, now these new tools show us Dali also for paintings and and of course it's GPT for for words, for texts. Show us that many of these human capacities are actually computable and we don't it's we don't need this creative, this little light bulb coming from above, from the muses or from inside or from wherever. It just it's a combination of existing ideas that gives new ideas. And these ideas are also patentable. And we don't need this idea of spirit there. I think it doesn't it doesn't bring us forward to just to bring up some bit of discussion where we don't agree maybe. No, no, no.
Martin: [00:31:06] No, no. I mean, is you are entirely right when I say spirit, then the first thing that comes to my mind is religious literature, right? Because in all religions, you would say that there are some actually many. Actually, I know only one that has that the Bible is is inspired. This is what what they say. And when you read the Bible and you have you have you have the gist of it, then you would would realize what is Bible and what is not. You can you can see it immediately from the from the from the text there. But that's why I didn't want to dive into the spirit thing, because I would already be happy if I find a good concierge for my website or a good patent robot for draft. I don't I don't even want to turn all the people into religious followers of what what I what I say that also it would be nice right let's let's let's now look at the technology we have started before to be brought up grammar which is clearly rule based and I'm following language processing for a very long time, maybe 30 years if it's enough that the first translation programs up and as a patent attorney, you always work with translations and then there were translations today are much better than 30 years ago, of course, especially the automatic ones. And also there's the GPT solution. And when we draft our patents with the cotton, then there are parts that are rule based and others that are not. So who is the more competent of YouTube when it is about the technology?
Speaker1: [00:32:50] Hmm. I don't know. Toby, you go first.
Toby: [00:32:55] You go.
Kim: [00:32:55] First. Well, just maybe I can talk a little bit about myself, because I've been a mathematician by trade and I've studied mathematics and then moved over to linguistics, and I worked on language, on grammar, first on rule based grammar, and now more and more on language model based grammar. So I think that's already an interesting question when you say that grammar is necessarily rule based. Well, I think there's it's not so clear. I think that we like to think of it this way and we would if you want to learn a new language and you want to know about what what the order of the words is and whether there is a case system or not. And these kind of things. These are, of course, rules. But all of these rules have so many exceptions that the best encoding would probably not be rule based, but it's always like we like to think of it as rule based. And then we we try to pack a group. The the observations in the that we have in the language into packages that are human size so that the humans can understand it, but it's necessarily a simplification of what's really going on. And that's the problem that I mean, look at history of machine translation. We already step further than in actually in generation. So before we try to build rules, I mean, I've been working in the end of 1990s in the first speech speech machine translation system built at the German Center for Artificial Intelligence in Saarbrücken, and we built a system that's completely rule based, that generated it, analyze it, generate and so on.
Kim: [00:34:33] And all these everything we did and many other people did is completely, I mean, overtaken by, by, by, by a large measures by this new language model. So machine translation has changed and it has changed by ignoring all rules. And so there is if you look at deep altcoin, the best system, this this doesn't use any rules and all is based on training on different types of training models. And and they and so it's not so clear. I mean why is the translation better if we ignore our rules? Probably because the rules had too many exceptions to actually encode them. Well, by human, we cannot deal with this anymore. And the machines do it better by just giving them lots of translated data and they just learn this. So no, I don't think that language is actually rule based. We can give abstractions like simplifications of language that is rule based, but this is maybe only useful for grammarians and for second language learners. But if you want to do any system that works, then you don't need these rules anymore.
Martin: [00:35:51] Toby.
Toby: [00:35:54] Really? Nothing much to add, I think. You know, unless we are developing a completely new language.
Martin: [00:36:01] I mean, that's the reason why you two are here. Because I'm a little patent attorney with some experience in programming, and I'm seeing this technology coming up and now becoming very tangible for my own job. And we have this initiative. We founded the Robotic Patent Drafting Family International. And and now this technology is so at hand that you. It improves. I mean, you can turn a technician into a very good patent attorney even, but just giving him the today's chat GPT for rewording what he tries to describe in his own words because the chat tippity is already much better than a typical junior college student when it is about writing a text that can be understood by others.
Kim: [00:37:04] Yes, I think this is really the interesting question. The interesting question is for this for the patent attorneys. I mean, I see all attorneys actually even I mean, all lawyers actually as being translators. They translate something into legalese, into some language that fits the requirements of the of the law. And and now if this translation, for example, you get an invention, disclosure or a scientific article, a preprint or something, and you translate this into a patent application, are you doing anything actually more than a translator does? And if the answer is no, then let's look at what happened to translators in the last years. They didn't disappear, but they are completely machine based. How they do they they post edit machine translation. That's the main job of translation I think. I mean, I think I'm not putting myself too much danger if I predict that something similar will happen quite quickly to a patent attorneys.
Speaker1: [00:38:06] Yes.
Martin: [00:38:08] It says the patent attorney here in the round. Yes, definitely to the average patent attorneys. But what what.
Speaker1: [00:38:16] Now.
Martin: [00:38:18] If I can add a very specific field that is really my favorite is about patenting algorithms. The the I mean, in Germany, the requirement for for any patent subject matter is that it's technical. Technical means that you can achieve something without the interaction of a human. So and in very simple words, you can patent something if it falls on your foot. So if if you if, if it falls on your foot, it is then in certain. Then it is certainly basically patentable if it's novel and inventive over the prior art. So having said this, if you have an algorithm, then you must make this algorithm somehow technical and technical means. In that way, every jurisdiction has developed their own criteria. And what I know from my experience over the past. Now 30 years this year, is that the patent attorney needs to become creative in order to turn an algorithm into something that is patentable because you need to to to fit it into a into a scheme or into a structure that enables the patent examiner to tick a certain number of boxes convincingly. And and that is not something that can easily be done by a machine because the machine doesn't know about convincingly. The machine assumes, as you have said, Oh, this might work. So I try my best here. And if it's not convincing, okay, then I try next time. But the difference in patents, when you talk about patenting, you cannot afford as a patent attorney that you you give it 30 or 300 or 30,000 tries. You expect it to come up with something that is pretty close to 100%. And I can't see that in the same.
Speaker1: [00:40:20] Martin.
Toby: [00:40:22] And I think. Martin, we lost him.
Speaker1: [00:40:26] Hello, Martin. Can you hear me?
Toby: [00:40:30] Yeah, but actually, I'll jump in here. I agree with Martin, because one of the things that I did when I when I moved to Switzerland, I did exactly that. I mean, I mean, focusing on helping inventors, basically putting putting their ideas in a way that, a it's almost like a translation layer from a creative translation layer. Let me put this is not a translation layer. It's more like it involves business modeling. It involves, you know, this how to describe something that's in your head, not into legalese, but into into something that then someone can take and make into into a legal language that is that is acceptable for it to be filed as a patent. So actually, I think the niche here from a from a patent attorney's or generally IP point of view is is actually preparing a possible inventor or a possible invention or innovation before it would go to to a lawyer or to a I system to write out the patent. That's where the where the real beauty of almost the composition of intellectual property lies. At least that's my, my, my experience. Yeah. Happy to be. Uh, listening to to other experiences.
Kim: [00:42:01] Well, I think that we should not underestimate what the translator does. I think translating is also highly creative because you have to fit the story of one context into another where you can actually understand some of something. If you translate, I know a story from one context to another. You have to choose the right words. You have to restructure the stories, and it's a creativity with some constraints. And what Martin just said is he said it's very creative, but then he said, you have to check boxes in the end. Well, this sounds to me like a I mean, I don't want to take the term creative away. I'm just saying it's it sounds to me like something you can train the machine on whether already I mean, we're trying to we're trying to do this, for example, for ideas to which extent they are patentable. And then we can try once we know it's patentable. I mean, I completely agree with Toby that it's there's a different job that is a business direction. What is good to patent, what would be a good direction to develop and all this stuff that's exciting questions but and it's also work of the patent attorney and maybe this will remain and then only the writing process will be automated. But I'm not absolutely convinced that this is the hardest part. It's probably the most fun part, the most interesting part. But I'm not absolutely convinced that this is going to be the the part that will be preserved from automation.
Toby: [00:43:27] I think I think the key key message is there are two parts that at the moment are being one. You know, I think the entire process of of coming up, say with a patent, right. So far it is it is actually two steps that are clubbed under one roof at the end of the day. And I can definitely see how these are being split into an area where we where we say, okay, we need a human, maybe with some help of an eye, but we need a human in this part. For the second part, it's actually almost better if an AI does it, because I will definitely not miss to tick all the boxes.
Kim: [00:44:09] Exactly. And that's actually the selling point of Cadent. Where we're working at is that we say we take the boring part out of patenting and we leave you with the with the interesting choices. And in a sense that also Martin just said, depends on the patent lawyers. Some will still subsist a bit like translators, for example. I know poetry translators are probably not using the bell, but most translators use it today. And so maybe there's also some different classes of translators and there's some different class of attorneys, but all of them, at some point, if they want to give legal advice to their clients, then they have to decide whether it's a good idea to patent something. And what is the business development? What is the choice of what should be patented, what should be hidden? And this these are probably harder questions than they actually writing process. So I think the writing process at least that we are working on should be automotives, because I think no attorney enjoys putting reference numerals into their texts and and, and checking whether all the words have been introduced in the right definition section and all this stuff. I think that's I think no one is actually proud of being able to do this well. But these choices are we mentioned that it's this business advice and in the in the invention industry, inventive industry, that is something that it's probably harder to automotives, but it's probably also be machine aided sometime soon, I think.
Speaker1: [00:45:42] Uh.
Martin: [00:45:43] Rich would disagree, and I can tell you why. Because turning an innovation into something that willing the people are willing to pay money for is an entirely different level. It's not it's not about has nothing to do with innovating. An idea is cheap. And and the problem is how do you sell it to the people? And we are talking here about a sub problem of a problem of a problem. Of a problem. We are talking here how how can we simplify drafting patents because a patent alone is not is not a value it has the value is zero unless you do what is described and protected by the patent. And so yeah, people always forget that. And in the innovative process, the intellectual property namely how to protect what you. If just conceived as a product from being copied by the competition. The IP part is the least important one, says the patent attorney. The most important one is not the conception, but the most important one is finding out what the people the market wants. Without that component, you can scrap your entire innovation pot. Because if people don't buy what you've just invented, then you can also not just not do it. The same outcome. Different, by the way, from writing music or another artwork.
Kim: [00:47:18] Why is it different? I didn't get this. I mean, music is also about being able to sell it, right?
Martin: [00:47:22] No, music is about. Music is about Tobi. What is music about? About the spirit. This is where you start.
Toby: [00:47:35] No, no, no. I would. I would. Some music. This is. Let me take this one. After. After doing music for so many years. So music is to different people, different things, first of all. So. And that's why the answer is probably not as straightforward as you like. And it is not about the spirit. It is more about emotion, right? I mean, you know, everybody in the world has emotions. And so even some I would say some parts which I would not consider music to other people, it stirs up some some emotions. And to them, it's music, right? To me, it's just noise. But and this has nothing to do with culture or taste. It's it's the emotion inside the music that that matters. And why does it uplift us or why does it depress us? These are all fair questions alongside the question of music, but music is probably the most universal language at the same time of of stirring up different emotions in different people. You know, it's probably the most universal language there is, you know, because you you literally I mean, an A is an A in America, in China, in where wherever it is, Right. I mean, it's it's it's quite universal. Right. And and but, you know, at the end probably, by the way, this is one of the reasons why, you know, when they send all these, you know, human history and and human accomplishments out in space with the hope that some form of E.T. one day contacts us, that they send different types of music, you know, whether it's a mozart or a Beatles or whatever, maybe maybe somebody feels, you know, impressed by one and others are feeling impressed by others.
Toby: [00:49:21] But cutting a long story short, I think music with music, there is something in the music which either touches your you or it does not. So it's about capturing that audience. And and I have yet to see honestly, I mean, people talk about I generated music. Let me come back to the I element of this and move a little bit away from the intellectual property part. But the reality is I tried so many I music generators and they are crap. You sent me one, I tell you. And I agree that people are disappointed when they figure out, you know, oh, and I made this. But there is no and excuse me, we use the word spirit and I'm going to use the word soul. But it's so soulless, this music. And maybe maybe people have an antenna for this or they don't. I don't know. Maybe people are just different. But I composed three pieces in the last three years. Totally. But I would say 30, 40 minutes, every note, handwritten. And I know why I write this note and I know why I write it in this place. I know why I put this phrasing. I know all these things. Now you I just cannot believe and maybe this is my ego, but I cannot believe that, you know, eventually I know is the same. So I don't know whether it will be able to capture an audience from an emotional point of view or. Yeah. I mean, as it would be, you know, with pieces that some of the old masters composed.
Martin: [00:50:58] Well, Toby, this.
Speaker1: [00:50:59] Was.
Martin: [00:51:01] I should I should get this type because the, the last paragraphs is just more the older pattern the tourney's carry inside. You said I, I've, I've written three and translated into my language. I've written three patents over the past three years. Each one of the patents has about 80 pages, more or less. And each word in the patent applications that I've written has its place, and I know why it's there. And you cannot replace it with another word because I'm so convinced that this is the right word there at this place. And and if you asked 100 patent attorneys, 90 patent attorneys would just sign that statement because they say, oh, yeah, that's true for me and and because I'm the. And that also explains what I've seen or what I've heard in the last conference. The plurality of of patent attorneys thinks that their own work is a masterpiece and an artwork. Well, what the others do is only crap. That's the that's the sentiment in the in the patent industry. And that's exactly what you described.
Toby: [00:52:12] But let me let me add something to this. Now, if you would take another composer and I would show this person my piece, he or she would basically say, I totally disagree with you.
Speaker1: [00:52:24] They would.
Toby: [00:52:25] They would not.
Speaker1: [00:52:26] Yeah.
Toby: [00:52:26] You see what I mean? They would absolutely not agree. I mean, they would even perhaps not even I don't know. I mean, we are getting a little bit sidetracked here, I guess. But no.
Martin: [00:52:37] That goes to the very core, I tell you, because the problem is not so much will artificial intelligence once replace as humans or not? The question is, will we be able to use the artificial energy, artificial intelligence to just take away all those boring tasks that we don't like? Like, as Kim said, putting reference numerals into 80 page text without making a mistake.
Toby: [00:53:03] So so let me let me give you that example with the piece that I composed. So generally when I compose, I first can play before I put it in in an actual script into the into the music, which you can see and follow on the screen. It's in my it's in my head, it's in my fingers, it's in my feeling. So I can actually play the whole thing. But, you know, it takes me another, whatever, two weeks part time at least, you know, to write a bloody thing down now. But I'm already playing it on a on a midi enabled keyboard. And I tell you what the there is no I at the moment at least, I mean because you know, sometimes a quarter note is not a quad node, it's probably a quad node plus, you know, just a little bit more or you use the pedal left pedal, right pedal. It's basically it's dumb as a rock when you when you try to get the music that you're playing onto sheet music, it's really stupid. Even I could actually do that as an assist. That would be extremely helpful to to a lot of composers around the world. I mean, they would probably save, you know, 30 to 50% of that time. But that's solution currently.
Martin: [00:54:13] Yeah, but that's what we are currently doing in the patents world. And 20 years ago we started to do that in the latest world. And I can tell you 20 years down the road, if you contribute to that development, that will be something that would help you to to take away the the sweat work that you don't like. And I'm very sure. Or Kim.
Kim: [00:54:35] You have so many things. I could say this. I mean, really interesting ideas coming up here. I mean, I think that of course, I agree that the next step will be a Toby said I helped composition, but just give you one key word there. That's a big thing in AI research now it's called explainable AI. That means we want the system actually to explain why they did something. So for example, for classifiers, it's a very classical task. So for example, you you ask you give photos to the to the machine, ask them to identify cats and dogs or whatever. And and then you ask the system, okay, why do you say it's a cat? And one way of of course, one could just to explain it by words. But one simple way is, for example, show me the zone that made you think this is a cat. What is the most distinctive zone which actually helped you distinguish it? And then if it shows you the wrong area, you're probably it probably learned some, some some, some bias. So for example, the cats were always photographed in the grass and the dogs unleashes or something and that's why it identifies this way so explainable. The AI is actually trying to to make the decisions that have been made by the by the AI system explainable. And this is really important. If we want to use AI in critical tasks like, like like cars driving self-driving cars or even some intelligent decisions on in jurisdictions and so on.
Kim: [00:56:08] So it's a law. So it's a really big question of to know whether the system, how the system decided and whether it decide on a good basis, on a bad basis. And so, for example, you could really well imagine that for composers it tells you, well, here you put this here because that and that. And I think this is not science fiction. I mean, this sounds like this. And the alternative would sound like this if we could put this into words. But another thing let me think of what you said is that something we had in current is actually a problem is like our business model is we also propose an editor that catches errors in your in your patent application. And Martin explained to to us that no patent lawyer wants this, no patent lawyer even thinks they can make errors and then to to try to sell them a tool that catches the errors is completely non nonsensical. So it would be the same to a composer to tell them why you should put it in C and not an A would sound better. I think toI would be offended and again, again in a parallelism between between Martin and Toby probably.
Martin: [00:57:15] Now please play the millennial, be offended. The language helped you. You can hurt your feelings, right?
Toby: [00:57:23] I'm too relaxed. Go ahead. Go ahead. No problem. Bring it on.
Kim: [00:57:28] No, no, I just said, I mean, what would you like? A system that not only helps you, but even criticizes you and says, hey, this a just doesn't sound right. Put a C here.
Toby: [00:57:38] Yeah, I would love that. I mean, look, I mean, it depends. I mean, look, here's the thing as a let's put it as a composer, right, as an inventor slash composer. But generally, these types of people, they have extremely. Strong opposing opponents opinions about themselves or what they do.
Speaker1: [00:58:02] You know.
Toby: [00:58:03] You don't do these things without having a strong opinion about something, right? Because you can never materialize it. You know, you you lose your focus and something or someone or whatever disturbs or disrupts you and and and, you know, your whole thing is gone. So. So you need to be pretty firm in, in, in your mind or mindset. So the good thing is you're quite stable. So even a guy comes along and says, you should do this. Actually, you look at it and say, Oh, maybe what an idiot. You know, you can look at it from from both angles. But there are if somebody is truly inspired, he doesn't have the ego to be offended by something or someone. If somebody is very, very, very strong minded, not saying strongly inspired, just strong minded, he will have a strong ego, but it will very it will be very easy for that person to dismiss that without being offended as well. So to me, I mean, to me, unless this disturbs my flow, I would I wouldn't mind that at all. I think the problem is, does it disturb my flow? And what will happen is the software or whatever tool at the end of the day it is will basically say, would you like to get more of those types of notifications, you know? And you're like, No, no, no. Oh yeah, yes, please. You know, and it learns from what what you want to know and what you don't want to know. I think that's how I look at it.
Martin: [00:59:39] Yeah. That also because Kim, you're always very harsh with me when I say something. Then this is not cast in stone. So please take it with a grain of salt whenever I do something. So it's not about we Patent attorneys do not like to get hints. And you also see, because we have now fed the system with patent claims of other patent attorneys. And when we talk about the patent claims, we clearly see that they also agree that this was a mistake. And then, of course, if you if you see that this is a mistake, of course, would you like to have someone who tells you, hey, this is questionable what you're doing? Are you really wanting to do this? Of course. And but of course not in my own patent application. Right.
Kim: [01:00:27] Exactly.
Martin: [01:00:29] Hey, people, we are reaching the end of the hour. We cut off shop. That's part of our format because we have people here who need to go back to work. This is international. We have people from New Zealand until California and in all sorts of time zones, and not everybody can afford to spend more than an hour. And that's why Kate is here. Thank you so much, Kim and Toby. We didn't get we didn't get even 10% as deep as I wanted. I believe we need to repeat that. And let me think about that, how we can get this done, because you two are also very under high time constraints. And I see that this is very difficult, But I thank you anyway, in the name of everyone who here, I think this was a very enjoyable discussion to listen and thank you so much and see you next week.
Kim: [01:01:24] Yes, very inspiring. Thank you for the invitation.
Speaker1: [01:01:27] Yes, thank you. Thank you, Martin. Thank you, Martin. Thank you. Thank you. Evening. Thanks.
The 4x4 Innovation Meet — AI & IP