BBW24 – Day 3 – Blockchain and AI: Balancing Innovation With Ethics

so hello I'm Yuri catalo I am a co-founder and general partner at Athenian Capital um I also am a research scientist at Autodesk and I focus on the business side of emerging Technologies so everything from web 3 AI to the metaverse um and looking at where the company could potentially go in five to 10 years and how we can make you know money in the future so I'm it works yeah yeah yeah okay like this so I'm joho and I am I am the co-founder of Nova and with Nova we are uh trying to to build an AI that is able to accelerate the research process uh so this is my first ad my second ad is that I'm building one of the biggest Community here uh in AI we are at cic it's called EOS and we have now uh more than 40 startup in our space and AI startup and more than 200 members and we expending in Berlin so in less than six months so it's very successful and uh the last thing that I'm doing is I I am an artist and I'm exploring all the new technologies so the metaverse the the AI so yeah it's fun awesome um I just spoke a couple minutes ago so you guys probably know me but Jesse uh CEO of trust wear tangentially interested and uh uh now let's get into it so um first thing wanted to ask you guys first question ensuring fairness and transparency in AI systems through blockchain let's talk a little bit about this let's talk about what what you guys are seeing in this space maybe start with you okay so we we have a lot of uh ethical problem linked to to to to Ai and the first thing that you um probably heard like 100 times it's uh it's about the the bias right so um we cannot avoid the bias uh in AI but what we can do is try to minimize them uh we you probably heard about the the story about the Apple credit card in 2019 uh where uh the limitation the credit limitation was not the same for women and men uh so this is the kind of of things that you can find in Ai and very easily uh discrimination um so how we avoid this so I guess that in the model you need to plan diversity but this is not easy so the things that I want to say is that most of the time we think AI as a fixed model but it is something that is in constant Evolution and that that that needs to to to grow with the different Insight from people so yeah so uh so one area where I've noticed a couple of things happening is actually in clinical trials for drug development um and so I'm I'm on the board of a of a new cro that's kind of working on this but what's Happening Now with uh the use of AI is you can ReDiscover or reuse drugs that were thrown away through clinical trials uh so they can fit a better subset of individuals and so what's happening now is a better Advent of personalized medicine so there's a company actually working on this I'm not filiated with them but it's uh care. trials and they are U utilizing web 3 in the way that it should be used uh which is to create a safe secure transparent way for researchers who are using clinical trials and Drug development to find better subjects that may not fit the traditional you know white male type of uh setup when the when the you know the higher ups actually usually do clinical trials and so what it does is individuals from around the world can come together opt in and there's an AI system that explains exactly what they're opting into and so if you go through a pharmaceutical company they're not going to necess explain to the average person and a lot of people feel like they're statistic or they feel like they're being like tested upon the way they have this set up is the exact opposite and so they use AI systems to explain ways that the participant can utilize they also can increase the diversity of participants by focusing on a a wider you range of individuals um and because it's a web 3 native web 3 place um again the individuals their their data is secure they get to decide who gets access to what parts of theirs and they know they're not being prayed upon by traditional systems yeah that's so I think one of the things that I take away from that is along the lines of the origin and authentication as well when you're looking at things that are cross uh cross compatible across different ecosystems especially in the healthcare side I think that's one of the things that we St we talked about originally and that's very interesting I want to really touch on that and one of the key points that we had L looked at which is the validity and Source um of data so um Yuri do you have any other thoughts on that in terms of the in terms of validity and Source yeah so we talked a little bit about um and I think there a longer this actually been I think coming up multiple times today about the election um you know a lot of memes and and different types of uh misinformation that's coming out there and so looking at the validity of sources and actually love there was a startup earlier today that was pitching about this was ABS phenomenal but the idea of again we all live in our own bubbles uh we have different sources of Truth and things that we believe believe into but having a single source of Truth where things are kind of stamped um again really love what that startup was doing in internally at like larger companies it's less of an issue because we have multiple systems in place where we sort through there's again people whose full-time job it is to sort through our data to to you know quantify it to categorize it to find out what's true and what's not to get rid of the duplicates but when you start going to the larger area of like open source or even again what's happening now in the political Arena of uh different types of information and different demographics particularly so I'm I'm again I'm from Indiana uh my parents are are aging Boomers and there's a lot of things that come across their desk that they will believe as truth because they have a different kind of filter they're not used to questioning a lot of things and so having some kind of stamp of approval will help with this kind of I guess re-education of just of us asking critical questions of like just because I saw it on the internet doesn't mean it's true and how do I see with my eyes and believe my eyes yeah a picture is worth thousand words a video is worth way more than that and then what happens when the video is a little wonky and yeah what are your thoughts so different things on on this topic the first one is that um as the poison can be the remedy also uh this is the same thing with ai ai can spread the the the the the this misinformation everywhere but it can be also used to um uh spot the misinformation so my company uh Nova was was was basically doing this when we we start to to accelerate the research process we went like oh but we should also provide to the researchers something that uh a feature that can be a detector of verif verified information they Tred to to attach it but didn't work you're good just enough um so yeah so this is this is something so for example just just just some numbers um when AI appeared in the research papers we came from 2 million a year to 5 million a year can you imagine so second of all um if you if you I'm I'm sure there is some researchers here close by MIT and all of these so um when you when you when you do your your your research sometimes you will quote a paper and this paper that you quote you quote it because it was quoted by many persons so the value of the paper is a quote but it doesn't mean that it is better sometimes it isn't it has been retracted from magazine and you don't know it and you continue to quote so that's a big problem and it it was very difficult with my company to to to to to manage this problem we we we still didn't figure it out because first of all we were like but how we Define what is true or not this is a pH philosophical problem and it's always like this in technology you have a philosophical problem and you need to translate it in technology that's very difficult so what ensure you that you have the truth or not sometimes just perspective when we discuss there is no truth it's just perspective and we build the truth uh um so that's very interesting for for this point of view and after we were like but we have some criteria of verified information and when we mix all these criteria maybe we don't Define the truth we Define what is is reliable different yeah it's very interesting I mean I I I've never really thought about it in that way and I think that these types of discussions really start getting me onto the more theoretical side of things one of the things that we did talk about before that re that that leads right into is the value of human uh oversight when we're dealing with AI right if you're looking at truth what does that actually mean and what is the combination between essentially The Human Side and the uh and kind of the AI side the automated decision-making Etc so um thoughts on that no yeah okay so there is something I I was at MIT like three days three days ago media lab and there was really really uh great panel uh about uh interactive AI so interactive versus generative AI so first of all you can use generative AI in interactive mode right so you you probably all uh try this you use chat GPT simple prompt and it's garbage literally okay so if you are in education and your student did that with a simple prompt you're like no this essay is not good it's logical but it's not good so um so that's a problem you need to be good at prompting and you need to be good uh how to to interact with AI to build something something meaningful something deep so for example when I'm writing an article um I'm not doing it in in English in in general I'm writing it in French and I'm translating in English and after I refine the English that's the the first thing so it is an interactive manner to use the AI same in art you can just play and roll the dice you have uh pictures but is it art I don't think so what makes it art is the effort is the construction with a machine so if you put for example your first drawing and put a nice promp you have an a picture that really look like it the same thing that if you drew it but after you can just take it back and uh and improve it like manually this is this is what is interesting with AI this interaction but just to come back to my to to my MIT thing um they are working to build real interactive AI especially in music so for example you have this you can jam on an AI and the AI will adapt in function of what you're playing so it is the AI is really uh a person kind of uh with who you can interact she adapts or it adapts or E whatever the pronoun and you adapt also okay so it's it's really interesting I mean that's me I I think when we talked about this first it was the way that all of us kind of agreed on one thing which was it's it's in conjunction with the a that makes something beautiful um but I want to touch back on what you just said about it's like it's almost like a human right and go into uh something that you talked about Yuri which was checking biases right with like like essentially that that part because it kind of ties into the human side yeah uh so it's you know so you know as humans we are all of us have some kind of biases built in whether or not we're we're obvious to us or not they all are it happens a lot with um you know people who are also building out these different uh algorithms thejan gave an example of what happened at Apple there's numerous cases at like Amazon where they were trying to fire like do a faster way of of hiring so a bunch of Engineers got together and they set up this new HR thing and all of the the people who set this up these Engineers were you white men who went to prestigious universities and so they Ed their own resumés as the training data and so everybody who came back like looked just like them and everybody else was rejected it's just because that's what their training data was and they didn't realize what their bias was because they didn't check it um and so there's a lot of human interaction that has to happen that checks all of our biases one things that U we're doing at Athenian is we're going to we're doing the opposite um so again VCS we we are trying to look for patterns to make an outsize return and so you look for obvious what can be obvious patterns to what is already won we have created our own um data model that looks at Founders in different areas predominantly about their um opportunity so it looks at their um whether or not they they have a it gives them a resilient score gives them a score on their uh interest in their own projects as well as how their teams are reacting um as well as their persuasiveness because there are some key elements with Founders particularly at seed preed stage uh where you have to look at some more we'll call them like less tangible ways and it allows us to find startups that don't fit the obvious mold of like did you again did you come out of and I'm I'm happy to say that because I went there but it's like come out of these obvious places and it gives a more wider field um part of what caused my co-founder and I to do this is so again I'm from the Midwest my first startup was in Indiana no VC would ever look at me particularly my real background is in theater so I used to work on Broadway as a designer nobody's going to look at a someone who lives in Indiana with a theater background who's building this Tech startup never going to happen uh because I don't follow the average pattern um so yeah so like finding ways to check your balances either using an AI system or you being the human in that is very very important and we're seeing that more and more and more yeah I mean and I don't know we we deal with a lot of VCS and I don't know a single one that that I've ever looked at that looks at that type of criteria so Props to you for kind of Reinventing it and also doing it for a personal reason um we talk a lot about AI here last thing I want to say on this subject is like what we what we're looking at kind of is from from both of your perspectives or from both of your you know examples is like garbage in equals garbage out right and so that kind of brings I think one of the big big things that at least I see blockchain being able to do which is ensuring data quality right if there's a like a validity or a stamp as you had said before what does that do so that we understand where this data is coming from or we understand what we're training with um so yeah wanted to get into that agree 100% yeah yeah well I just want to say that like so the I guess the the main key thing here is um I know we're at a Tech conference it's very much blockchain focused like blockchain is not a not a Panacea it's not going to solve all Mankind's problems same with AI it's very much important to think of these as tools and whether or not they are applicable and so like in this instance if your data is correct and it's already good data you can then verify it but blockchain itself won't tell you if it's good or bad but it can tell you the validity of it so particularly if you're not using like some internal structures where you've created it yourself so I think so there's a good some good points of that as well as like you know just realizing it has its limitations and you can't just say well we did a blockchain solution so sold yeah so it gave you history and and origin of data which is good for example uh in art again so I will I will take this example you probably heard about this this artist he he did this art contest and he won the art contest with the AI his name is Jason Allen um and uh after the the the the people in the competition asked him to retract his uh his work but he wanted to sell it um and uh that was where uh came the problem of copyright so and the problem of transparency also of the model so he he tried to to to go in court to try to prove that he he built it with different uh with this interaction but how you can measure it without the blockchain with the blockchain you would have been able to tr track the history and say okay there's 50% AI but 50% human and after you can discuss about AI oh I did it with the AI but you have still your 50% of human you can discuss but um just with this uh AI tool they just took the picture and say oh no we don't know how mid Journey uh is functioning this is not transparent uh this doesn't belong to you and we will not have this discussion now but just a question for you this is when we can ask ourself if because of AI we are forced to rethink our classic Paradigm for example do we need to have a copyright why copyright so for example I I will have ideas from this panel or from your discussion I will come back home and I will write an article will I quote you no no no and I'm sorry just because I maybe I don't know your name or just maybe because at the end of the day I think that it is my idea and we are all the same in this I'm sorry to say that this way but it is true and we are quoting in um when we are doing our research when we have something written that we can verify but I can tell you my first job was to be professor in philosophy I can take you pandar from uh Greek word and n and have the same same sentences just because he translated but nobody could verify this so you have this kind of problem already but we didn't have the technology to verify it it's true and maybe we should stamp that exact quote onto an nft so we have it immortalized in time right yeah I mean but in addition to that so like if you trra play even further like look at so Microsoft right now they're making all their money doing on co-pilots with people so and I hear this all the time at the corporate level because we're building our own large language models you guarantee that like one of our customers does not want to see their design show up on their competitors customers and so like that's internal data we're trying to figure out how we deal with it I don't know how Microsoft does it but they're like is trying to Microsoft is trying to run the world with their co-pilots but I can guarantee you if they had some kind of authenticity system internally that could verify like listen as if you're one of our customers I guarantee that your competitor is not using your data and your models like the you know the constant joke we use so I also work in a it's like give me a light version of zaha ID's design and so you can have like zaha ided knockoffs happening all over the world they're like this phenomenal design firm um and they have a very distinct style but it is replicable if you start breaking it down even more so how do like these types of of companies and I know we talk a lot with ar about this of like with their distinct Styles not showing up on someone else's buildings and without being attributed or even like you know leaking accidentally through our system or through another system so being able to stamp that is a huge deal uh with trust and transparency which all companies are like terrified of like it takes you know forever to build your trust but then a moment to lose it so if a large corporation suddenly loses that trust no one's going to use them and now they've just killed them themselves I think it's a it's an incredible point and when you look at kind of the way that things are going with with the adoption of these systems into the corporate world just want to ask those like thoughts about kind of the barriers I mean we're kind of in in overtime here we've answered most of the main questions but what are you guys thinking about kind of the next couple of years and what we're going to be seeing here as well oh so my my fear and and it it comes just right away from yesterday so I didn't think really about all of these but I met one of my friend and he's working with live AI I didn't even know that it existed this so he's showing me this so you take your phone and you prompt and at the same time the space is transforming I can take his face and start to speak with his voice in live so it's it's much more it's it's deeper than than the Deep fake it is a live fake really uh so I I see like really uh good things about this like if you if you try to create with a voice for example multimodal AI you you speak with a voice your your your uh painting is changing and you write at the same time and you use all the the possible canels to create something in a real time but if you if you misuse it that's that's a disaster how many people could could use it to Ru ruin your reputation in five second usually I I'm doing I I told them I'm doing the this presentation the show with me my Avatar behind speaking and and at the beginning you have this really philosophical discussion and suddenly he's saying whatever he wants and I'm like what are you saying right now and it's me how do you recognize that it's not me without the blockchain technology yeah this goes Way Beyond just moving the cursor every couple minutes to show that you're online I mean you're showing up to meetings that it's you but it's not you I mean the implications are a little crazy Yuri I know you have some thoughts on this uh sure sure so we talk about more like like corporate adoption it really comes down to like what the average person thinks is is Law and so I uh at at Autodesk I spent years trying to build up our web 3 part of it wrote a bunch of white papers get presentations to the CEO and then suddenly Tera Luna like messed everything up everybody because then that was a chain reaction and they're like no our customers will never trust this so it's dead it's not even like things you can bring up anymore because it's so recent so part of this and and it's like it's sad kind of watching some corporations drop this Starbucks had a phenomenal web 3 component that they didn't call Web 3 but it was a way for users to interact and to talk about coffee and then they gave you stamps which are nfts but they didn't call it that and that was a great way to do that it came from their marketing department not from their researchers um or if you look at do swoosh swoosh at Nike still doing well unfortunately Nike has fired all of their researchers uh who built that out so I don't know what's happening on Nike side of it but it really comes down to and I think it was either your keynote or another person's brought this up it's like blockchain's going to happen in the background they're not going to let their users know about it because there is this other lying sentiment about when someone says blockchain or web 3 they atically think of crypto scams and until like the whole Market shifts that then other corporations are not really going to adopt this type of thing because it's not worth the risk now if you say you've got an AI component your stock is going up everybody loves it I mean that's not always going to be the case but for right now ai is the cool sexy thing and web 3 is not the cool sexy thing yeah yeah except for us nerds here yes exactly awesome I want to give a little time for questions I know we're we're we're at the end so yeah questions could you touch a little bit upon trademarking IP for AI I was reading a little bit about it and there was some issues around doing that but is that something that's being pushed forward both you know in Europe and us for for you know AI creation I I'm not a lawyer yeah that's not my thing I'm not a lawyer also so I have no idea about this um intellectual property I know even less than that soorry yeah um on that particular subject no I I never heard about this to be honest and I think there is again there's too many problems he he said it very well I mean there is too many problems with with the with the data origins in a large language model that we are using using I mean this is why open AI is sued by New York Times right so um when it's come when it comes to intellectual property when it comes to copyright it's this is a real problem and and and and I don't think we will be able to to manage it without the blockchain this is this is my real feeling this uh great conversation guys what are your thoughts on the advancements with digital identity um and kind of how that's all going to come to play with this and data being the new gold um you know utilizing AI to kind of keep this open source and for example in clinical trials right you know you can use ZK to mask a lot of the personal information and then smart contracts to verify these things so that you know you have these truly decentralized economies that are able to still share data and move the science along much quicker but you don't have to take into account the copyright because a lot of the information maybe not even seen so what are your thoughts with that yeah I mean we were talking about this exactly on our setup of like proof of personhood but I'll let you because you I'll let you go under that one so I I this uh this is something I'm saying all the time uh about like digital identity I don't even know why we didn't have U earlier this conversation when internet came because all these I don't uh does somebody have a a big B social network here no like if you have so for example if you have a big Instagram if you have a yeah Instagram most of the time you start to have a lot of hateful speech okay it's it's it's impossible to manage that when you are arest on internet I mean it happened to me and it is a hell to get out from this a hell um nobody's really able to to to help you in this the the police is completely like uh uh out of the game in term of Technology it is a hell and um so so I don't the only option for this is a digital identity and this is a clear um use case for for blockchain this so I I think that the the next 15 years this is this is something that will happen we will have a digital uh identity and I think it is normal we can we cannot dream about a word where we are uh more free without the responsibility that comes with the freedom I mean for me this is crystal clear this that's that's a amazing way to end this off yeah I I agree oh yeah go for it oh sure hi so we're talking about the biases the different ones and uh as an international student have believing that and suffering from that especially with the different with my resume maybe because there are many many steps and the first step is the resume being analyzed by an algorithm and my question today is how could we be fixing all these biases I know that um in machine learning we're training the machine in a certain way and as you said previously we're giving inputs and how could we fix the inputs and how can we make the system better today so I have an idea but go go ahead on this one yeah so so same um that's very good because my wife has been in in the same kind of situation because she has a she has a Russian name and uh and I mean because of the war and all of this she's not Russian but uh there there is a problem and if you start to look at her CV she's coming from Stanford I mean she she has an exceptional CV she she talked at United Nation one of the five expert uh who was uh invited to to to to to write the the EU chart uh for AI so and she's not able to find a job and I'm like how this is even possible this so I hear you and most of the time it's because we are replicating in the AI system the the same mistake that we did in our human word so I don't know if it happened to you but I was witness of this you have this guy yesterday was working at McDonald I don't have any anything against people who are working at McDonald good but um yesterday he was at McDonald suddenly is uh HR in a big corporate for AI and is judging your uh curriculum so that's absolutely wrong there is another way to do that uh it's not to have somebody who was interviewing with all the bias human bias this time not artificial intelligence biased uh but it is about um giving an answer and you have the skills that will appear and that will fit to the job or not so for for example if this is a model that we Implement in the HR uh AI system we have more chance to avoid the biases but if we if we still keep our old school manner to think and to hire people and we see that it doesn't work in our world HR is a big big problem uh we will not solve it yeah so that's it I think it's a human problem before oh definitely is so like think about incentive structure so I guess and I have a very untraditional background um I just happened to into a job at a corporate environment but I applied for many and I'll never get picked up because I have a weird weird background um HR is incentivized to bring on people that could fit a role and when you write roles and I've done many of these you like write out a list of qualifications and these are just ones that you are not say guessing but you're just like Bas on other people I know who can do this job this is what I think they should do and that's like it translates all the way down to an H manager who then looks at this and says well I am incentivized to hire somebody as quickly as possible who fits this criteria and they don't get creative even though there are many elements and phenomenal people who can do similar roles who don't have the traditional like a b c d d DD all the way down and so you have to like start from the very beginning of like hiring managers the people from the top down who are willing to look beyond the buzzwords that that are created to hire people and to think like outside the box it is difficult to scale in the beginning but the companies who are doing it do a phenomenal job of bringing in Talent from other areas and these individuals are thriving and succeeding then in other parts they wouldn't have because they don't have like again they don't check all the boxes and so unfortunately now we're trying to get AI to do a lot of things AI is not a human it just replicates and looks for patterns and we're feeding it patterns that are not helpful so you almost have to like scroll all the way back and start from something scratch and have like individuals who care about finding people and then let that be the new Norm but it's like it's like again it's a human shift it has to happen and humans have to get involved in this part I I mean this is last thing I will say on this like I'm in the position where I've been hiring both at the like centralized exchange and and now and I try to look for people that are go-getters above just like the CV like one of my favorite things to do is on the first interview just like take the physical piece of paper resume and toss it because I've already read it right I know who you are right you have the interview but it's about like can I see and we've we've hired four in four different countries right we six people on the team four different countries right and it's it's it's like that's why I'm so attracted to what you guys are doing on the VC level is because it's looking at the it's some almost the intangibles and looking at the at the way that we can bring that in um but I don't know how that gets extrapolated into the bigger AI system but I I I think that you guys are right about about the human interaction there um