Why Decentralized AI Is Not a Crypto Issue... It's a Business Issue | Jeff Wilser - AI-Curious

Hello. So, a trigger warning here. This might be the least popular presentation you were here here at all of ETH Denver. And I'm going to say some like non kind of like okay thoughts about where the space is and where we're headed and how we speak to things. So, as January mentioned, I'm the host of the AI Curious podcast.

I'm also the host of a decentralized AI podcast. And in many ways, I feel like I've been living these two lives in the past two or three years. These two lives that are almost at war with each other that have an almost internal AI struggle. And I'm going to call this the suits versus the hoodie struggle. You could also call it the LinkedIn verse telegram, right?

Everyone here you see Eth Denver when you meet you're going to like exchange telegram, right? Every other mainstream business conference in the world if you say here's my telegram like what the hell are you talking about? Here's my LinkedIn, right? So the rest of the quote web two world is all LinkedIn the suits our world is Telegram. So I have kind of embodied both these worlds.

There's a reason I'm mentioning this. If I can indulge in a bit of personal bio so you have an idea of where I'm coming from and why this could matter for the space. So in my hoodie life I've been covering web 3 AI, crypto AI, decentralized AI, whatever you want to call it for quite some time. I'm a longtime writer for Coindesk. I've spoken to many of you here for over the years for articles.

I see colleagues in the crowd. Hello Brad from Coindesk. So, a lot of Coindesk and Consensus background in my blood. Um, almost three years ago, I wrote a piece for Coindesk that at the time seemed kind of fringe of like 10 ways crypto and AI could work together. And at the time, people like what?

Like, how could crypto and AI work together? That was a novel idea. Uh, for the last three years, I've produced and hosted the AI summit at Consensus. I just flew back from Hong Kong. still a bit jet-lagged.

Uh I also host a podcast called the people's AI focus on decentralized AI and I do believe in these principles. This is not lip service. I think it's important. I appreciate the ideals that many of the people in this conference are working really hard to build. things like fairness, things like very reasonable concerns about big tech oligarchies, not trusting institutions, doing AI the right way, enforcing privacy, data sovereignty, user owned data.

These are all really important concepts. I am rooting for them. I I kind of view this as the I call it David's verse Goliath, right? There's like Goliath's really freaking big, but there are a lot of Davids with slingshots and can this kind of like to mix metaphors, there's a rebel fleet of decentralized AI projects, can those somehow band together see the underdog story to kind of topple the big tech AI? I'm rooting for it.

But a confession in my other part of my life I'm in very much mainstream AI. Do you want to call it business AI? I work with companies on AI strategy. Companies that have no interest in decentralized AI. They are not thinking about DEN.

They are not thinking about user sovereignty. They don't use that language. So in this other world, I speak at web summits on their AI stage. I host the podcast AI curious that has far more to do with mainstream applications of AI, talking about productivity, talking about businesses who want return on their investments. I interview folks on that show like LinkedIn's head of AI, IBM's chief scientist, Upwork CEO, and these business leaders, they are all focused on, and it sounds kind of corny and crass, but the bottom line, how to make AI work for me, how to make AI useful, how to do AI safe.

I also, and this is even more a field, do things like AI workshops and AI strategy sessions for companies again nothing to do with web 2 like normcore companies like construction companies and even dirty word in these parts like banks I literally have worked with banks on thinking about their AI strategy here though is the ultimate irony and this is what I wanted to share today these two worlds, these two like dueling sides of my life, the web 3 AI, decentralized AI, and then kind of business AI, normcore AI, I've recently realized they're not as far apart as we would think. So when we kind of drill down on the business AI concerns again, here's what they're focused on. How do I make sure the money I'm spending on integrating some agentic workflow, how is it paid for itself? Is it safe? If I'm a CTO, am I get fired if this goes sideways?

How do I ensure compliance? big big word in these circles, governance, right? They're worried about customer data. They're worried understandably about their IP, about customers IP. They're worried, how do I make sure my IP and my customers IP isn't gobbled up by open AI.

Those are very real questions your average VP has at a Fortune 100 company. really any size company. I mean, I even spoke to a working kind of a senior manager at a food distribution company and they're asking me, Jeff, I want to use AI. I want to really use AI in our workflows, but I'm not quite sure how do we make sure like the databases we have, this is like a small mediumiz company, how are databases like safe? So if you drill in a little bit deeper on these, I'll call them enterprise AI pain points, I'm going to say there are three main buckets here.

Uh the first bucket is leakage and liability. And there are some stats here that help kind of uh convey this point. Cisco had a study founding that 62% of employees are actually uh entering info uh into Genai that they should or shouldn't do. Uh there are net scope net found hundreds of AI link data policy violations. 2,000 incidents in one month.

Source code is being put into these chatbots. God knows how secure this is. IP like sensitive IP is often being put in these chatbots. These are the kind of leakage that keeps CIOS up at night. This is what makes them this is what makes them pause on their AI adoption.

You might have seen some study last year from MIT that over 90% of AITIC projects were failing to get a higher ROI. Now, frankly, you can debate how solid that study is. There's some arguments out of context, but I think the point still remains. businesses, companies from top enterprise down to SMBs are struggling with adoption largely often because of concern of its leakage and liability. Essentially their question is how do I know my data is safe?

Similar bucket two governance and auditability. How do I know what's happening here? How do I know who's using this? How do I know where the data came from? And so again, the Cisco study, uh, the majority of companies understandably have limitations on what data can be entered.

Maybe it's okay to enter data such as your emails, but not okay to enter your customer's proprietary IP. 61% have limits on which tools can be used. Again, quite reasonable and a quarter companies have banned gen AI altogether because concerns. McKenzie found that 44% of companies said that at least one negative consequence. Again, this is all to say companies have really deepseated concerns about how this is being done.

So the kind of convention narrative on the headlines, OH MY GOD, ALL CORPORATE AMERICA IS RACING to use AI like kind of, but they also have a lot of really tough questions. They're throwing a splash of cold water in it space as they should. Bucket three, vendor lock in concern that they spend all this money to develop AI solutions only to realize, ah we're kind of stuck with these guys indefinitely. And by these guys, that can mean a few different things. That could mean contractors, that could mean SAS pipelines, data storage, or open AI.

So, you can see where I'm going here. That was all suit talk I just gave. Those were all like extremely like meat and potatoes, almost boring, mundane business concerns. Virtually all of these questions, virtually all of these problems are precisely the problems that many decentralized AI projects are racing to solve. that many of the people here at East Denver are sprinting to solve.

They're just using different language. And I would argue the addressable universe is massive. And there's potentially an opportunity to shift our language, shift our messaging, shift how we market this to reach this massive audience. So I won't read through these in detail but the gist the kind of business AI way of asking a question is how do we use AI without leaking sensitive data. This is a question that is essentially being worked on by privacy preserving computes by confidential computing different ways ZK style zk style AI developments auditing compliance guess what crypto and blockchain is really good at figuring out how to audit things we're really good at that we know how to do that So we know on the rails on chain how to show receipts.

How do we collaborate across organizations? Again interoperability is something a space has been working on for years that's already in development. the question of vendor lock in crucial question for business and that's also something that I mean look at Ethereum look at so much of in a way you could argue like crosschain bridges that's a crypto version of saying avoiding vendor lockin so these kind of cryptographic solutions and decentralized solution are ultimately about avoiding vendor lockin ultimately avoiding consolidation of power and influence. Decentralization is a different way of saying let's not get stuck with one company, stuck with these guys or giving these guys too much power. So why does this matter?

Apart from my own therapy session of squaring and reconciling my two lives, you guys don't care about that. But big picture, I mean, I don't mean to be flip or reductive, but we use the phrase web 2. And psychologically, that's appealing because it kind of implies, well, that's cute. Web two, those guys are ahead buried in the sand. We're on to web three.

And it almost suggests that like we're so past that But real talk, web 2 is pretty much the entire world. And people say web 2 AI almost suggesting it's some backward antiquated version of AI. That's pretty much most of AI. So if we want decentralized AI to make its way to the rest of the world, it could be useful to start acting and speaking in a way that appeals to the rest of the world's a 124 trillion market cap as compared to a two and a half trillion market cap.

Thank you so much. Thank you, Jeff.