Future-Proofing Technology: AI and Blockchain Standards
good morning everyone uh Nal chrisan uh back again for uh round two of an uh uh engaging uh AI conversation um I'm a tech reporter with fed scoop and cyers scoop um and cover the the Nexus of tech and government with a focus on Tech policy issues like AI cyber security data privacy um and and such um uh and I wanted to quickly introduce uh my fellow panelists um starting with Shauna thank you so much for having us here today my name is Shauna Hoffman and I'm the president of guardrail Technologies and it was fun for me to hear guardrail mentioned so often earlier I did not pay Adele or ywan to mention it uh heyll so really excited about the opportunity to be here I spent over a decade in I uh working on Watson so it was fun for me to be kind of in the trenches um I had many like many of you seen Watson win Jeopardy so I thought ah this is great it's the answer to all my problems I realized very quickly after building AI solutions that it wasn't the end all be and so I got into blockchain uh back in 2017 and I've been very happy to combine the two so I'm excited to be here today thank you hi I'm Eric Lapin president of form free uh just brief history 25 years years uh 15 of that was in banking and capital markets as uh as well as technology and analytics and um you know I've been focusing on empowering the consumer with their own data direct from The Source uh uh and the lending business so um you know we've been around the company's been around for 15 years mainly in the mortgage industry and provided and validated uh income asset and employment direct from the source that consumers would control that data and we are now focused on um the web three approach where we're having that data tokenized and could be shared by any consumer without their personal identifiable information provided until they want to share it so it's empowering the consumer with their own data sharing uh that data possessing that data and then lenders make decisions based off credit profiles and nothing else uh no bias involved in that hi I'm Mark wer most of you know me because I was this CTO for GBA for a couple plus years what most people don't know is that at the same time I was working for the stealth startup AGI Labs out in Seattle and that was a project where I actually got to use professionally both GPT and blockchain for the past several years there was GPT well before chat GP came out thanks so much Mark so we're we're going to put up a couple of slides real quick right now to to lay the groundwork and give some um foundational facts and understanding of AI standards so real quick you know um I I the reason we're having this conversation right now and the reason uh standards for AI and blockchain matter is it's not just the folks who who who read uh you know my stories and Reporting which is you know mostly insiders within the government and tech companies or blockchain entities but the the The Wider sphere um of of the the general public has an interest in this as well because things like uh you know algorithmic discrimination data privacy safe and effective systems human Alternatives you know the key areas of of of of standards of of whether it's AI or blockchain are important because Beyond one of course there's you know human and civil rights concerns and people want to make sure that their data is protected that that things are transparent and so there's a consumer concern in in making sure that certain standards are in place but also money talks and the ways in which companies succeeded building AI or blockchain Technologies is when there's a clear standard that exists as to how tools and Technologies are supposed to perform where is the data that that uh that they build upon where does that come from what transparency is needed and standardization allows companies to have the trust that they need to have with their consumers and so if that trust doesn't exist they'll never make money they'll they'll never actually have paying customers and you know so so it's it's it's not just a sort of Pie in the Sky intellectual theoretical element it it has very real world consequences the standards that we're creating right now for AI and blockchain will have an effect for years to come on consumers but also on businesses and those trying to sell products because without these standards and Trust um you know you you just won't get consumers and widespread adoption in in in the first place and so just wanted to give everyone a quick lay of the land uh in terms of what are the standards in place right now in the in so in the EU which is one of the only places uh where we have uh a AI standards baked into the law which is the EU Digital Services Act the EU AI act um you know they they've been discussing this for some number of years and uh it it finally passed um at at the end of last year and we've had some developments this year and it's expected to be uh the compliance is supposed to start in in early 2024 and it covers uh areas like illegal content um transparent advertising disinformation and misinformation which are obviously big in in an election year and it requires explainability um blackbox explainability decision-making disclosures and advertising targeting practices um so that's that that's that's the EU in In America which is on our our next slide um you know we mostly have voluntary AI standards uh right now at uh on the national level and on the state and local level there's there's a few laws that have been passed but by and large it is voluntary um if we can move uh to our slide on uh us AI regulations um okay we yeah we we we can we we can start here so you know the the two key AI standards on the national level are the White House has its blueprint for an AI Bill of Rights which as I mentioned before is focused on safety and effective systems algorithmic discrimination data privacy notice and explanation and human Alternatives um this is a voluntary framework that the White House has put out this is it's not mandatory for any body to follow it but a number of major companies Microsoft Google Amazon others have made voluntary commitments to incorporate parts of this into into their business model and then we have the nist AI risk management framework which is the National Institute of Standards and technology which has put out um a risk-based use case specific risk management framework for AI um which is which is uh is supposed to work complementary to the to the White House a Bill of Rights but some in Industry have said there have there are contradictions but these are the two main ones the AI Bill of Rights from the White House and then nist AI risk management framework uh and then we'll we'll move on to the next slide uh and so this is a real quick snapshot of us aai regulations um as I mentioned there's a sort of a patchwork of laws that are moving through State legislatures a very small handful that have actually passed you know we have uh New York City's local 144 law Illinois has its uh 820 ilsc Maryland has its uh section 3717 California has its AI bill um and so you know this sort of is an explanation of why we need something on the federal and National level is because if there's a patchwork of State Solutions it is confusing for consumers it makes it difficult for companies to have a standard approach across the country and so this is why we really do need to come together and and have something on the national stage which is why you know we have some action happening in the federal government like the AI research resource act um which which which Congress has has has passed recently uh the American data and Privacy Act which um is still in the works uh the algorithmic accountability act um and then FTC rul making and enforcement the Federal Trade Commission has taken a lead role in AI uh enforcement uh based on laws that already exist on the books um so hopefully that gives you a a a decent starting point and Foundation as to the conversation that we're we're about to have on Ai Ai and blockchain standards and why it is so critical both uh for consumers and for for companies and those creating this um so I think I think I think we we're good with the with the slides here um thanks Tech Team so I wanted to start off um if uh if Eric if you could give us a little layup the land I've explained the AI standard side if you could give us a La Land on blockchain standards and where they are right now and where where they're headed yeah and and and from a where we are today especially in the use case side of things um you know we we see a lot of it especially as it relates to fractionalization we see it a lot with I'll give you an example um real estate title insurance and lending is a great use case to discuss because you see a lot of um you see a lot of companies figuring out what blockchains are going to be used um how is the AR arcial intelligence work with that and as you saw in one of the slides there's uh one of the bullets was about blackbox modeling and that becomes extremely important as we utilize the blockchain because we want to make sure as best as we can that the data that's being um utilized um is Source data and for an example Source data uh an example would be um banking account information financial information that is permissible by the consumer um um for attaining a loan for one primary reason why you see that and you'll see a lot of that has to do with especially in the Banking and Financial Services space that identity is very important so from the banking side you want to look at kyc AML you're probably familiar with SSI self- Sovereign identity but there's one there's another piece of that called ssfi which is self- Sovereign Financial identity which is fairly new but being reviewed and looked at by the financial services industry to look at transaction data direct from the source and that digital ID is anchored to the chain so from the blockchain standpoint you're making sure that whatever that digital identity is verified that that chain is is you know number one interoperable number two that that it does have the standards to work with certain types of artificial intelligence and some of the changes that we're seeing especially United States and other parts of the world is removing blackbox modeling um especially when it has to do with inclusion because there's some bias that comes into that so with artificial intelligence using Linguistics you can do mathematical deterministic off of source data to figure out a true ability to pay for a consumer so as it deals with standards andal that you know blockchain and artificial intelligence working together um to really look at what data is being utilized where does that data reside and how is the uh Intelligence being used as far as removing bias from any type of decisioning but real quick Eric if you could give us a brief explanation on the national scale or or on the local side what what sort of blockchain Standards exist right now give us a sense of what some have called The Wild Wild West yeah um I mean we're we're really seeing that there are organizations that you mentioned FTC earlier but um you can take other organizations such as uh Resa Federal Credit Reporting Act um um they're utilizing standards that you really have to start with verifying who someone is and creating that digital identity and that's really the most important of what we're seeing to to get it out of the Wild Wild West and remove a lot of the fraud that you would see so that's I would say that's probably if I were to pick one that's the most important it is which one uh identity ID identity okay understood um and then Mark if if if you could maybe shed light uh on some of the if you could compare in contrast you know with whether it's Identity or some of the other standards that that Eric was referring to and then the AI standards that we mentioned before in what ways are AI standards and blockchain standards different from each other and what ways do they have overlap and similarity so so I guess what would be most helpful is to talk about the artificial intelligence blockchain maturity model supplement um there really are no standards that attempt to address both artificial intelligence and blockchain there are a good number of standards and they're being developed in real time for artificial intelligence particularly after the generative AI explosion of last fall uh there are definitely blockchain standards for particular Industries as Eric mentions but AI is not one of them um the blockchain standards that the blockchain maturity model um appendices put out is basically how can blockchain be used to make this particular type of application easy to promote sharing make it trustless make it private and also make it so that these systems can be evaluated um in one sense AI is a standin for virtually any software product that could be dangerous that you know has lots of people working on it where we've got to have privacy where we need to be able to evaluate the results and change course in the middle um so that's what we're striving for here um AI is also a particularly good example because it separates cleanly into two radically different types of systems we have the type of gradient descent neural networks that we've been familiar with for the past decade plus and then we have the new generative AI systems and for the neural networks we actually do have Source data um with the problem being that at times it's data that needs to be hippoc compliant um and with the generative AI we really don't know what the source data is now both of them most of the time um also have black box problems so these are the things that the standard the problems that blockchain can help solve and what we want to do in the AI blockchain maturity model supplement is to point out all these shortcomings of AI that blockchain can help resolve and how we can actually resolve them totally so I think that brings me to to to my next question as as we sort of turn from understanding the to to to Solutions you know I think the key challenges as you outlined with AI or blockchain are we want standards that allow um you know consumers and and those who create these tools to have control we want some sort of confidentiality or privacy we want the ability to make quick uh updates and iterations and we we want an element of transparency so I'm curious how can blockchain standards address this this problem give us this this set of tools which over were you sh yeah absolutely so um I wanted to mention one thing uh so I was the chair of the cftc's distributed Ledger technology committee years ago and the CFC is oh sorry uh us Commodities Futures Trading Association thank you yes thank you very much are you from there no I was gonna say stand up and turn around so and it was it was fun to be part of industry and be brought in on those meetings early on when blockchain was something that was really not talked about a whole lot and it was an educational opportunity for the Commissioners to ask questions to those of us actually Building Solutions we are seeing that same thing now where we have many committees on the hill um even my own Congressman Darren SoDo I've had great conversations with him about where we are in regards to AI regulation and how this works so um my background is in legal but also of course in technology uh my prior to guardwell Technologies I was the chief technology leader for for legal and strategy at Dell and one of the things that's always seen is it all starts out with what we call the legal industry soft law it's the guidance it's the standards and then it moves into regulations many years later right now everyone in this room should be part of these conversations that are happening with every every agency if you don't have one in the list may I ask you today to please because they need to hear from you my local Congressman I mean it was he was absolutely busy and all of a sudden he heard Ai and I'm forming a company in your town I was the first one to be able to sit down with him in that uh session that we were at and actually have have him hear what my concerns are responsible AI is so important to me um and it's because I've been in AI for almost 20 years now ai is more than 90 things when we started doing this I mean it was developed in 1956 this is not new what's new is it has been released to the entire world something that I used to charge $20 million for not joking um when I was at IBM is now free and so we have everyone in the entire universe if they want to able to get on to chat GPT for example and put in any question you know one of my friends had called me she's like Hey we're gonna go out for dinner Chad GPT I asked it where we should go and it stated where we should go and it wasn't a real place it was some madeup place yeah actually that would be funny um no so so we have some issues with a system that was made by fallible humans by humans for humans and it gets things wrong about 26% of the time and so what do we need to do today we need to put standards in exactly so you know where I'm going with that um so I wanted to mention just how the law works why it's important that every person in this room becomes part of these conversations because the one thing you do not want anyone to regulate is the technology itself and the reason I say that is because technology the moment we take a breath is going to change it's probably 91 things now 92 93 um so as we start to Think Through where do we want Society to go I'm more worried about the outcomes and worried about biases and discrimination and so where I see blockchain fitting in is to make sure that every piece of data that is used in any training model is understood is known there's no more black boxes and it can come out to the surface uh there's a great example that happened years ago with um with face well with Amazon actually um they ended up training their system using of course AI machine learning um year I think it was like in 2018 and they trained it all off of um resumés it was their recruiting system but it was all majority of men's resumés and so what did it do women like me are ousted out of the system because I did not fit whatever model they needed we have to make sure that our data is not biased that our training um happens from so anyways I just wanted to mention those few things and uh hopefully that's a bit helpful for those in the room and gives you a little kick to go out and help certainly thank you and and so I'm I'm curious you know to to sort of build off of um the the the the great Foundation that you've given us what is the role of blockchain in AI when it comes to um securing machine learning algorithms and transparency and and availability and uh privacy concerns and algorithms and uh algorithmic bias and discrimination in in what way specifically can blockchain standards play a role in alleviating this and do we see any examples of this already occurring so fundamentally blockchain should be the substrate and the control system around AI models um there are plenty of malicious Act in the world so we've got to be able to for example record what was the data that the model was trained on um again it may be hippoc compliant but that is fine because you can actually set up um certain blockchains so that these stored procedures can see and operate on Hippa data but no one else can actually get to that dat so you actually can do some incredible things in the medical field as Marquee told you and still maintain confidentiality um you need to have documentation of the results of what this model comes up with um if possible um you want to have transparent reasons not blackbox reasons um that may be too much to ask for but honestly there's no reason why there can't be some post talk analysis done by something other than the neural network itself um you also need to be able to evaluate systems again you're getting back into smart contracts um there are a lot of algorithms that can be run on data so that it is um not blind to particular variables because that's problematic because a lot of the time other things stand in for those variables for instance race is easily replaced by income or ZIP code or something of the sort what these algorithms actually do is they pick spe specific variables and ensure that basically the results for that variable are the same for all the potential values so that way you're ensuring that minorities get exactly the same result distribution that for example you know whites get that poor people get exactly the same distribution that rich people get um which is really important because you start looking at things like law enforcement um and there are plenty of programs now AI does sentencing and it's an absolute nightmare because if you're poor you are going to get a much worse outcome and this this is the type of thing that blockchain can help with if we know how the models were created what the data is if we can actually throw fake data sets against it and rely upon the results and see the bias if we can also so use these algorithms all of this becomes public data it's very easy then to start evaluating and we can ensure that we get the results that we want and again the blockchain maturity model really addresses results um it can make suggestions as to this is one or more ways in which this can be done but that's more to prove that it's POS possible and to inspire thinking understood so you know I you know there have been a series of challenges in coming up with AI standards as we've said on the federal level on the national and local level I was hoping whether it's with the blockchain maturity model that you just mentioned Mark or or or others if any of you can peel back the layer into what it is like to create and come up with blockchain standards and what do those conversations look like behind behind behind closed doors and what are some of the challenges and successes in creating said blockchain standards yeah and and I think we're all saying the same thing which is great because we all come from different backgrounds um but at the cor at this at the core of all this this is conscious capitalism um this is intention economy and intention economy was a term coined back in 2012 the author of the book is Doc surl highly recommend you read it um but it's basically it's taking information in data making sure it's accurate and its source data and making sure that decisions that are made actionable on top of that are accurate and when we're looking at these when we're talking about the models that we're speaking about here um at the very core of all this it all comes to inclusion and as a society we've lacked that since the beginning of time and one of the things that this solves from the blockchain and AI standpoint is the standards is looking at the data that you know where do the data reside how do you making it actionable can the consumer control the sharing of this data and we're seeing that switch from the web to where we are today where you have the big corporations you know the fangs Facebook Apple Google all those guys utilizing data but the cons me the consumer all of us in here the consumer don't know where it's being used how it's being used and we're certainly not making money off of it so change that flip the script on all that now put it into the web three space where we're going where that sharing of the data now can be utilized by the consumer the standards around that I think number one the first standard is there's this organizations Mark talks about Hippa um you talked about when you were at cftc I was very fortunate when I was at credit Su to work with cftc we didn't know each other back then but um worked with rag um retrieval augmented generation and that was really being used for simple things back this was like 12 10 12 years ago or so but recently I don't know if anyone read an article but Bloomberg just uh has really in the financial services space come up with something that basically said um from a commodity standpoint utilized 50 million data sets uh 1.2 million in hours of retrieving this data and putting Artificial Intelligence on it but not using a blackbox model so we've all three set it blackbox blackbox models will bring in the bias to any sort of decisioning whether it's going to be with HIPPA respa uh anything with cftc OCC um yep gdpr uh good friend Omar right here that we're on a committee with mizmo together which is in the Housing Industry he works with Jenny May which is part of housing uh a HUD housing Urban and development and standards we have a standards group that we work with there and really a lot of it is making sure that the data sets that are being used are Source data that it's accurate data because anything that's then going to be placed on the chain and anchored on the chain is going to have accuracy bias removed inclusion at the core um and then artificial intelligence that you use on there what we're seeing a lot recently is the mathematical determination of someone ability what of someone's ability to pay we have 5050 million Americans in the United States that have low or no FICO because everyone's judged by a three-digit number that says I have to be in debt to get more debt that's what it is it's wrong it doesn't work it's a vector you still need it but you also need to look at someone's cash flow discretionary income other things that are happening in that person's life bring it all together now you have a open aperture of really determine someone's ability to pay so that's an example of a black box that maybe decades ago was helpful but it's not helpful anymore you got to we got to stop and look at that so the standards are let's take more of a Linguistics approach or natural language processing approach that you know I'd like to hear your opinion on that but you know we're seeing more and more decisioning that's made without the bias when you look at it from that that side thanks Eric I think very shortly here we're going to open things up to to questions from the audience but I just wanted to give everyone one last chance uh if anyone has any closing statements or want to add anything regarding real life use cases of AI standards that you have seen professionally or personally um before before we move it over to the audience shaa green are you guys hearing me I know okay that's much better all right I've been fortunate to be be part of um two different AI ethics committees one when I was at IBM and also the other one when I was at Dell and I love the big corporations who are really jumping in you know like Oracle also putting these AI ethics standards out there and they're putting guidance they're being the um those who are leading the charge I would Absolut these are public models take a look see what they're doing and also bring those inhouse to your corporation your department um when you're you using generative AI uh you know it's available of course to everyone make sure that you are putting those standards in place and you checking your work um one of the reasons why I formed guardrail um and renamed the company that is because I think it's extremely important for us to have those guard rails to make sure that we are factchecking looking at contradictions we not oneandone using the AI systems um I think an important standard for us is to make sure that we are looking at the AI and those results um there's a very good explanation of that um uh I don't know if you all saw ma versus aanka case um those in the legal realm we all were pretty mortified that there was an attorney who had went to chat GPT requested a motion to be created and the motion ended up citing five cases chat GPT created two of those on its own so three cases were real two were not they turned this into the court of course the clerk who asked to check all of the Motions that came in said who whoa hold on two of these are not even real they asked the attorney where did you find them now he said a very reputable company name and unfortunately that caused lots of issues I am surprised that the sanction was not more than $5,000 but he was first in the industry unfortunately to get caught I knew somebody would um so again it's factchecking it's checking our work making sure that what comes out just knowing first off it's gonna have biases it's going to have problems it's going to have hallucinations up to 26% of the time check your work um so that needs to be a standard uh that comes in into play very one thing that's very worthwhile to note is that AI standards tend to be of two different types one is is quality and that's something that the blockchain maturity model can help with another is where AI can or cannot be used and that's a problematic um area for standards anyways and it probably cannot be controlled by such things um also with regard to chat GPT and hallucinations there are numerous ways in which that problem can be handled um chat GPT even has the capability to go out and check for Source material and verify any of its facts so if you have it set up correctly that shouldn't be that much of a problem um there are a whole bunch of other things that can be dealt with with anciliary systems um a lot of the time for decisions for chat GPT if you one of the things that we worked with um out at AGI Labs was making it so that decisions could be explained in or AR English whenever they were made by a generative AI system um these are not things that are impossible a lot of these problems are because many of the companies currently are insisting on one big system rather than separate best of breed systems to do some of the jobs I mean the whole idea of reinforcement learning by human feedback is a nightmare um as long as the companies follow this it will be easy to jailbreak chat GPT and cause it to give harmful answers and hallucinate um so the standards really have to be about results and where we can influence results is where our standards need to be do anyone any questions good morning my name is Ryan Cooper I represent DeVille crypto Solutions and I just had a question um are you all segregating your uh attacks at the solutions for sexism racism and classism because they all kind of fall in the same category but they all have very um different solutions and I'm just wondering if you're um solving them in a segregated manner good question um so the use case that I was talking about earlier is is solving for that and also bringing into what the panel members were saying about it's it's about the results and it's about the standardization of the data that's being used so just let's take an example from wealth generation or ATP which stands for ability to pay um big issue that we've had in United States and the rest of the world but I I'll tackle United States for now is uh R sex gender race um have had decisioning that's been detrimental to a lot of people and by the use of blockchain and the use of artificial intelligence by looking at removing the blackbox model in this case because the blackx the blackbox model has been used by um automated underwriting decisioning it's been used by credit scoring um it's realized a lot of people understand that this has to change so there are a lot of things that are change what what's happening right now as we speak is basically taking information that is direct from the source of a person that permissions the bank accounts open banking platforms if you're familiar with that ficity um plaid a bunch of them out there when the consumer permissions that permissions that data to be utilized the artificial intelligence can now come on and have Linguistics utilizing Mor morphology utilizing looking at the grammar the syntax really determining the ability to pay and it becomes a tokenized so that information is a token which is the representation of the value of that ability to pay and it removes all the bias and the decisioning where you have lenders that can say I like this credit profile I like this um I like this alternative scoring model to know the ability to pay here's our Loan offers and then once the Loan offers are excuse me the loan office the loan offer is accepted by the consumer then the consumer shares their token hit share token and that information is shared and your pii which is personable identifiable information is then shared then it's Coupe Coupe lives here this is where he's employed and waiting to share somebody's race or sex and it's not done until after the offer is accepted and made on the merits of the ability to pay and it removes the bias thank you uh I think we have time for just one one or two more questions so we'll we'll go yeah yeah please I'm I'll share the mic right after you sorry I I had um like an aha moment as I was coming in I was listening to the radio and the radio said that 80% of the educational colleges are using AI to make decisions now for re uh reviewing applications of students that are coming in I that's a use case situation that I want to see how how do we incorporate this combination of blockchain with AI to not have the bias that you're talking about be applied to admissions in colleges and universities because if we have a bot looking at an application and not taking into the humanity which is who we want to be our next generation of innovators I'm not sure what our how that we're going to solve this but I that's a an application that's not business-wise but it is education business so um throwing it out there how do you see we solve this problem so there are actually two different answers to both your question and his question if you're talking about racism sexism and economics um for the neural network gradient descent type AI the ones that tend to give either scores or yes no answers those are actually similar problems from that viewpoint so solving that problem with the algorithms that ensures that you get the same result you know regardless of you know what race you are what economic status you are that works it's pretty much a solved problem except for the fact that we don't implement it in 99% of the cases we should there's also the problem of the large language models and if you go to Google and you type in you know something like you know a doctor is a you know male or female you'll get male a nurse is a female there's so much embedded in our language in the internet and we actually don't have um ideas or we've got ideas but we certainly don't have um established standard operating procedures for how to deal with that and and in particular it would take a good amount of work um you'd basically have to build um special type of large language model that recognizes traits that you think are positive or negative um this is actually something that we were attempting to deal with and are still attempting to deal with out in Seattle um that is nowhere near a solved problem um it's effectively giving the AI a rubric in the same way that you would give a human being a rubric if these things are mentioned give them this much score if these things are mentioned you know or if these things are mentioned but those are open questions okay for our last question did you want to comment I'm sorry no we actually don't have enough time for a lot of off I can take off over time so just the question quickly okay is there a certificate Authority thinking in terms of blockchain uh instances using a thirdparty nonprofit certificate Authority that has no skin in the game whether it be evaluating this or that is that concept being thought about I think you just answered her question that's literally what I was going to bring up was so the answer is yes there are blockchain standards that are concerned with that all right
2023-11-12 14:46
