Official ETH Denver AI Summit: Balancing efficiency, end users and decentralization

[Music] there you are we some energy Let's Get It Started should we start with a quick round of intro L this first sure yeah my name is Casey I'm the founding General partner of topology which is a New Frontier Tech firm and before this most recently I was an investment partner at paradig and then before that I was an engineer at Google hello everybody my name is David Snider I'm one of the co-creators at lit protocol it is a distributed Key Management Network that currently manages about $110 million in assets and as of late we've been building a lot of tooling for AI agents such that agent runtimes can transact on behalf of users while providing guard rails and keeping agents unrug so to speak hi everyone my name is John um I am a co-founder at a project called coin cap I also Al co-founded um shapeshift which some of you probably used in the past um and I'm also an adviser to uh Venice AI hey everyone this is Vader I'm the founder of Vader AI we're one of the largest agents on virtuals uh excited to be here nice so that's super exciting so for the people that didn't check the the title I'm just read it out because we we have this stage a summit so this panel is all about kind of transitioning from the kind of like the acceleration between decentralized how we balancing efficiency H and how we get to kind of like end user and decentralization infrastructure so I just want to ask a quick uh question like where where is this kind of like the state uh where are we today uh and what are the gaps do you see right now around this landscape feel free don't be shy go for it okay can you please repeat the question yeah yeah so it's about kind of like so uh where are we today around kind of decentralization uh decentralized systems and users yeah so I think we're far away from decentralization at the moment for AI and I think it should as my personal opinion it should not be the priority because I think first we need to figure out uh product Market fit and then decentralize it um so like we for example um try to implement decentralization and a proof of a a lot of the stuff a lot of narratives that came with decentralized AI and realized that neither end users nor are holders really value or price that uh so we refocus more on the product Market fit aspect of it but this is a bit more again a personal answer curious about what the other panelist thinks yeah I mean I think um it's hard to answer in a broad swath it's like what needs to be decentralized what needs to be secured for example if you use chat GPT from open AI most people are very comfortable trusting chat GPT with the context they ask it you know the chat gbt has context around their personal lives what's going on for them but what they don't do is they don't send their private key they don't send their credit card numbers to trat GPT and so like what is the actual sensitive data that needs to be protected I think is the right lens to take in the context of what needs to be decentralized and secured and then where is that just like kind of a a marketing benefit where it doesn't really matter so hard to answer in Broad strokes and I think we have to look at it from this specific places of um where the threats are as it relates to having decentralized or centralized components in the AI agent economy yeah I mean I think in general decentralization is always a gradient um you know even even like very decentralized to something like Bitcoin or ethereum but many other projects which are decentralized to some component but not entirely I think in AI we're still figuring out what what degree of central uh decentralization is actually helpful to users and what degree is an impediment and I also think it depends on the project and I also think things like decentralization and privacy are interrelated and privacy actually matters a lot um in my opinion in Ai and this is something I think a lot of projects are looking at especially things like Venice of how do we actually get the benefits of AI without giving up all of our data and I think that's that's very related to to what degree decentralization can actually bring benefits to users yeah this is actually a good point like where are the trade-off that you guys see right now uh building around kind of like a ey ecosystem that you see either uh going in the right direction or not in the right direction I don't know you have some personal opinions what what are the trade-offs that you guys had to do um we have to know yeah I mean we definitely built with uh ATL protoc call this distributed Key Management Network we've built with a user controlled view from the start I think on a more macro level uh I've been really impressed with the work of uh what the mother out team has done with lit protocol encryption and recall Network formerly ceramic and table land in the context of making decentralized memory for agents and so I think this question today matters potentially less for each individual implementation but the reason I mentioned the macro view is we out and look a couple years ahead it seems to me that there's essentially two options one is something that looks a lot like the way that social media looks which is we have various silos in this case those silos are like apple Google and open Ai and the alternative is a complete societal upgrade with things like collective intelligence um Collective cyber security threat detection intelligent protocols that everybody can contribute to but are functionally owned and controlled by no one and so really I think like the fate of the internet and the fate of society is actually in this data privacy and user owned data component associated with AI agents yeah so I think the tradeoffs are very clear it's cost versus decentralization and um is is that cost worth kind of the ROI is that marginal cost worth the marginal returns most of the time not like if you look at uh you know I love the idea of decentralization or like you know let's have a decentralized uh Facebook Instagram you know Tik Tok where like the data we actually own the data users own the data and they can monetize it but the reality is that people don't really care you know they still continue using the centralized social media I think it's the same for like llms you know you have Venice AI you have chat GPT or perplexity and people are still using those ones so it goes back to the tradeoff of like costs versus benefits basically I think that's right frame just to follow up on that like in our world specifically we're very focused on policies essentially like abilities and Scopes that a given user can grant to the agent and functionally there's three methods there's one you could do it totally centrally with centralized enforcement with vendors like TurnKey and privy two you could do it entirely on chain in a smart account but then your capacity or the benefit is limited and then what we've tried to do is find the balance where you can have policies over any kind of transaction cryptocurrency transactions across any chain transactions through API for e-commerce buying stocks um using a browser that runs into te and having those policies be on chain and enforced through a distributed Network we think is the right balance because frankly Vader to your point the capacity is or the benefit is what people care more about than the decentralization but that's not to say that the decentralization isn't important important and like what we've seen with the safe hack is that it takes a traumatic event for people to then care and focus and I think we'll probably see something similar in the AI agent World although hopefully not as catastrophic yeah I think in general you make a good point is like decentralization is very much one of those things that you don't feel like you need until you really need it because something terrible has happened um you know hacks in general tend to show that out but I think in regards to AI it will likely take some sort of traumatic event like what you're talking about before people actually realize oh yeah maybe we shouldn't be doing this all in an entirely centralized Manner and giving all of our data over to open AI to do whatever they want with without even thinking about it and things like that you know I think Ju Just Add on like that will definitely happen but then we will go back to default just like you know what happened with the elections with like Facebook intervening and then there were a lot of talks you know we should not use WhatsApp anymore we're going to use signal and then you know everyone's still using like WhatsApp or like certain these like centralized stuff so I despite agreeing with you I wish it would happen but I don't think it's it's going to go to decentralized yeah I again I think it's a gradient it doesn't have to be 100% decentralized or 100% centralized I think there are aspects of decentralization that we can build into AI that are actually beneficial um and I I appreciate the projects that are working on that but to your point that doesn't mean that the masses are suddenly going to start caring about it so um but that doesn't mean it's not useful you know just in the same ways like many people in the early days of Bitcoin did not necessarily understand the benefit of decentralization but you know over 15 years we realized that like if Bitcoin was decentralized that would be a very bad thing and it probably still wouldn't exist so there there will be AI systems that get built that probably won't be able to be built without decentralization or won't continue to exist without decentralization and so for those things it will be really important but that doesn't mean every single AI app or any every llm or every AI application has to be that way it depends on what the goals are I mean the only things I want to say also like users mostly care about speed and simplicity right I don't know what what is your opinion on that because I'm also asking myself which are the main uses uh use case that we see around kind of subcategory of AI where there are there will be some trade-offs around that I don't know what's your experience yeah I'm curious on Casey's thoughts here on the tradeoffs um I mean in general like what because you you probably talk to more Founders than than we do like how are they thinking and how are they um kind of managing this I think that decentralization is changing so much right now with the government like I just can't even stress that enough how different it is and how different of a situation we're in there's like a few reasons I think over the course of the last few Cycles people decentralize things like one thing is privacy but that's not synonymous with decentralization at all like you can be centralized and private and I think like open AI is already trying to fulfill that hole and and we're going to see if that comes to fruition in a way that Society trust but that's one thing another thing is regulatory Arbitrage which historically has been a really interesting thing and maybe is not interesting anymore and then there's like this third bucket in my mind of why people decentralize things or Founders decentralize things which is more of like an ethos of ownership and power and all of that which still holds true and I think like I mean I just kind of painted three pillars I just think now Founders are relying Less on the regulatory Arbitrage is even been a reason to do this I also see a lot of like traditional AI people now coming into crypto because they think um with the regulatory landscape cleared up it actually is less of a risk to them to go into the space so they have like a philosophical alignment and they feel they can express that without the possibility of going to jail yeah and that and that definitely makes sense I mean I'm I'm a Believer though in general of like well while things may be better on the regulatory side for us in the crypto and or Industries right now like you never know what the future's going to hold so I very much a believer of like build things that cannot be destroyed by who's sitting in some chair at some given point in time and things can change so quickly exactly so uh you know it may be great right now but we don't know what it'll be like in 2 four years from now necessarily and so we we shouldn't forget the lessons that we've had to deal with over the last 48 10 years especially on the crypto side and let that you know some centralized governments effectively take control of AI because we built a bunch of systems where they can do that so we just have to be careful on that front I I definitely see a trend around kind of like devl and building uh the developer tooling specifically right because right now with all those AI tooling and before in this before this panel that was speaking about csor lovable a lot of like Frameworks are definitely accelerator accelerating the D development life life cycle so people will kind of like push those tradeoff go as fast as possible to like product product Market fit and then we'll say oh wait do we need to be regulated or somehow like be compliant on uh on I don't know like uh data data privacy or data kind of like feeding of of my models yeah yeah I mean it's like move fast and break things music is My Philosophy you do stuff it can be illegal doesn't matter um you make it legal basically it's like the Uber style um so I think when we zoom out you know if crypto AI wants to win we need to attract the web to AI Engineers top ml researchers like all these entrepreneurs and I personally think that decentralization is not the way like I think they're going to come for two things one is relatively quick funding and the second one is the ability to experiment um with more freedom and with more creativity so second comes with agents or AI being able to transact in crypto without having a kyc without being a kyc so they can do whatever they want and the first one like probably a lot of Engineers who are good at being an engineer are not very good at say fundraising or the business side of it and I think the whole speculation because if you think about like crypto and mem coins it's basically like a global casino and if you're able to gain attention and use the speculation to leverage and fund your operations you suddenly have this opportunity for entrepreneurs that now they don't need to go to the traditional VC route anymore and they can just raise from all the speculation all this liquidity so I think those two are the main um pillars that can help crypto win crypto AI win against web to a 100% I mean one biggest things that I see in this cycle is the brain Capital injection that is coming from like uh like AI like tooling developers from Web Two to web 3 right it's it's impressive like all these the most of the AI agent builders that I see building strong automations are actually from like strong phds and and the best way to also expand is they're coming into web 3 because the web3 rails nature you have data on chain you have kind of like you know in web to banking if you want to build a defi agent it taks like years to have open API like it it taks five years or something but I see we have one minutes and 40 seconds what is the main things uh that you're most excited about coming in the next uh let's say six months Max yeah um I mean just duv tailing off of that the thing that I'm really excited about is the that we've been doing with the lit agent wallet and it very much takes that perspective of trying to attract web2 developers whereby a transaction is a transaction whether the uh transaction is settled on bitcoin ethereum salana Visa Shopify stripe Charles Schwab whatever right like from the perspective of an llm from the perspective of a web 2 agent developer these are all just tools that the llm is leveraging and continuing to ratchet up the abstraction stack such that somebody can come into the space and not be concerned about the specificity of what's the difference between a smart account on blockchain a versus blockchain B like we have to move beyond that and that's definitely been a core part of our design lens with the lit agent wallet Casey or Vader anyone want to go next yeah I think agents have the potential I think in crypto there's only two product Market fits so far one is payments the other one speculation I think agents have the potential to be the third one and I think for that um we basically like I think it's very exciting because we have the potential to do this the agent theyel up into Frameworks like Eliza game they're nowhere near uh Lang chain or others right now but they're getting there I think as they get better we're going to see like these entrepreneurs like web two people AI Engineers getting into web 3 I personally on a mission to find those folks and like help help them fundraise and and basically grow um yeah I'm excited about that yeah I wish we could actually have more time and speak about uh a agent coordination agent to agent coordination that's a big topic you want to go next yeah I I think in general there's I think just the intersection of crypto AI in general it's kind of become a meme but at the same time there's actually a lot of interesting applications that can be built off of that and I'm really interested to see what consumer applications we can do on this crypto AI crossover whether it's on you know actual crypto related apps whether it's games whether it's other types of applications that are facing the user I think that's where there's just a ton to be explored and we just we don't have a ton there yet I think agents in general again are are very interesting but the the actual applications of what we can do with this thing is what excites me over the next number of years yeah I I'll wrap with two things one is I think the agents V1 that we've seen was really akin to nfts and the reflexivity and if we do which we will but when we have like another agent Uprising we really need to break into util IL or it's not going to last so that's like what needs to be done if it's going to work and not just be like another Crypton native asset class that goes in Cycles with with ex extreme volatility the second thing I want to say is like in the crypto AI category I think the thing that isn't talked about enough is how distributed architectures can be leveraged in a compute sense to drive down cost and increase efficiency and this expands way past crypto like real AI people real front tech labs all of them as we consume more compute really need to think about the right architectures and stuff like distributed training decentralized training distributed inference all of that uh it's just like there's no doubt in my mind it gets absorbed somehow into the global economy like the details are still TBD but it's a real problem um that we're posed to solve sounds good you already did here at e Denver 20125 cool ladies and gentlemen big round of applause