11 hours ago
You’re swimming successful data. You’re creating caller information each day. If your wellness app counts your steps? That’s caller data. The Oura ringing that’s tracking your bio-metrics? Valuable data. Your societal media posts, adjacent the anserine jokes that got zero likes? More data.
This is each information that AI companies would emotion to harvest. You can’t physique bully AI without bully data, which is wherefore galore presumption information arsenic the “new oil’ successful the contention for AI. The problem, though, is that portion your information is invaluable in theory, the world is that it’s hard to monetize your ain idiosyncratic data, arsenic you person nary leverage arsenic an individual. (Open AI isn’t knocking astatine your doorway to bargain your aged tweets.)
Enter Vana. “I deliberation information is this cardinal assets powering the adjacent procreation of AI, and truly the adjacent procreation of our integer economy,” says Anna Kazlauskas, co-founder of Vana and CEO of Open Data Labs. “A batch of radical frankly conscionable don't recognize that they really ain their data.”
But you bash ain your data. And it’s valuable… if you tin someway articulation forces with millions of others who besides ain their data. This would springiness you bargaining power. And that’s the ngo of Vana: To make an ecosystem for user-owned data, which successful crook fuels user-owned AI.
That ecosystem involves a premix of Data DAOs (a “labor union” for data), decentralized information marketplaces, the precocious launched VRC-20 token, and a caller collaboration with Flower Labs to physique the world’s archetypal user-owned foundational model. (Exhibit A that Decentralized AI is creeping into the mainstream: The Vana/Flower collaboration was covered by WIRED.)
Kazlauskas volition springiness a keynote astatine the AI Summit astatine Consensus 2025 outlining this vision, and she gives a glimpse here. And she sees the momentum shifting. “We're already starting to spot this displacement wherever much radical recognize that, ‘My information is truly important to AI’ and ‘I’m really the proprietor of that.’” She predicts that successful a fewer years, implicit 100 cardinal users volition beryllium onboard. In 10 years? “World population. Above 10 billion.”
Interview has been condensed and lightly edited for clarity.
Why is user-owned information truthful important to you?
Anna Kazlauskas: Most radical presume information is owned by the platforms that it's sitting on, but that's not the case. In the aforesaid mode that erstwhile you enactment your car successful a parking lot, the parking batch doesn't ain your car. You tin ever instrumentality it back. You person afloat ownership implicit it.
And there's a immense magnitude of wealth being made today, mostly by large tech companies, disconnected of that data, but users are the ineligible owners. So I deliberation it's important that we reconstruct that ownership, some from a idiosyncratic position and from a developer's perspective.
Can you link the dots of however this helps developers?
As a developer, particularly successful an AI world, having entree to the close information is truly important. And it's ace hard to bash close now, due to the fact that astir of the information is locked up wrong the walled gardens of large tech. So galore of my truly astute friends who bash worldly successful AI spell enactment astatine the large labs, due to the fact that that's wherever the information is and that’s wherever the compute is. But that doesn't person to beryllium the case.
How bash Data DAOs acceptable into this imaginativeness exactly?
So a DataDAO is benignant of similar a labour national for data. Where fundamentally you person a ample radical of radical who excavation their information together, and past tin marque corporate decisions implicit what happens to that data.
The crushed wherefore that's important is that your data, connected its own, is not that useful, right? It's overmuch much utile erstwhile there's a large excavation of it. When there’s capable of it to bid an AI model.
What are immoderate of the Data DAOs you’re astir excited by?
There are a fewer successful the wellness abstraction that are truly interesting. There's an aboriginal 1 that's really doing afloat exports of diligent aesculapian records, which I deliberation tin truly assistance beforehand a batch of probe successful the space. There's immoderate related to biometrics, sleep, and health. There’s 1 with the DLP [Driver Loyalty Program] Labs; they’re gathering car data. And wrong their data-set, the Tesla information is truly absorbing due to the fact that astir radical deliberation astir Tesla arsenic invaluable due to the fact that they person a information lead, right? Actually, the users tin get a batch of that data-set.
You’re pivoting from mentation to signifier with the caller collaboration with Flower Labs to physique COLLECTIVE-1. What’s the extremity there?
COLLECTIVE-1 is the archetypal user-owned instauration model. Usually erstwhile radical deliberation astir a instauration model, they typically deliberation of 1 institution moving a precise ample grooming occupation successful a azygous information center, right? Like OpenAI. And the crushed wherefore it's typically done successful a centralized mode is due to the fact that it requires, one, a full batch of compute power, and two, a full batch of data.
Flower AI is benignant of the person successful federated [decentralized] training. They've done a truly large occupation of gathering these large unfastened root libraries. They’ve travel successful from the grooming broadside and the algorithm side. And with Vana, we truly absorption connected that information piece, right? So we fundamentally person each this information that radical tin bid on. Then you springiness users end-ownership of the model, and users tin determine connected what the exemplary is allowed to do? So this is the archetypal instauration exemplary of its kind.
And the mentation is that eventually, with amended data, you tin physique AI that’s not conscionable competitive with the cardinal players but better, is that right? So it’s not conscionable astir ideology, but besides performance.
Exactly, yea that’s 100% right. From a decentralized context, I deliberation often radical hold successful rule that, “Yes, we should person AI that's owned by the people. We should person decentralized AI.” But what’s the happening that we tin really bash amended successful a decentralized context? Data is the answer. For each company, they lone person their azygous portion of a data-set. Apple’s got their data. Google’s got their data. But if you’re going done the user, you tin chopped crossed platforms and really physique amended data-sets than immoderate azygous company. Data is the concealed condiment that makes it each work.
Love it. Thanks Anna, spot you astatine the AI Summit successful Toronto.
Jeff Wilser volition big the AI Summit astatine Consensus 2025, and is big of The People’s AI: The Decentralized AI Podcast.
English (US) 