Vana Empowers Users with Ownership of AI Models Trained on Their Data

In an era where tech giants dominate the AI landscape, leveraging user data for model training without equitable compensation, Vana is pioneering a shift towards user empowerment. The decentralized platform allows users to own a piece of the AI models trained on their data, marking a significant departure from traditional data monetization practices.

The inspiration behind Vana stems from the growing concerns over data ownership and usage. The platform was conceived as a class project at MIT, driven by the vision of giving users control over their data and how it’s utilized. This approach directly addresses the imbalance highlighted by recent events, such as Reddit’s $60 million deal with Google, where user data was sold without user involvement.

Anna Kazlauskas ’19, co-founder of Vana, emphasizes the necessity for a better system: “This data is needed to create better AI systems. We’ve created a decentralized system to get better data — which sits inside big tech companies today — while still letting users retain ultimate ownership.” Vana’s system operates on a user-owned network, enabling individuals to upload their data and collectively govern its use. AI developers can propose new models to users, who then decide whether to contribute their data for training in exchange for proportional ownership in the resulting models.

Vana leverages a little-known law that allows users to export their data from major tech platforms, which they can then upload into encrypted digital wallets within Vana. This data can be used to train models, creating data pools called data DAOs (decentralized autonomous organizations). These data pools can also be used to create personalized AI models and agents, all while preserving user privacy by avoiding the exposure of identifiable information. Users are rewarded proportionally each time a model is used, based on their data’s contribution to its training.

One notable example is the AI model trained to generate Reddit posts, where over 140,000 Vana users contributed their Reddit data. Similar initiatives have been launched using data from X, Oura rings, and other sources, fostering collaborations that combine various data pools for broader AI applications.

“Let’s say users have Spotify data, Reddit data, and fashion data,” Kazlauskas explains. “Usually, Spotify isn’t going to collaborate with those types of companies, and there’s actually regulation against that. But users can do it if they grant access, so these cross-platform datasets can be used to create really powerful models.”

Currently, Vana boasts over 1 million users and more than 20 live data DAOs, with an additional 300 data pools proposed by users set to go into production this year. Vana aims to foster generalized AI models, personalized medicine, and innovative consumer applications, overcoming the data silos created by tech giants.

“Today, big tech companies have built these data moats, so the best datasets aren’t available to anyone,” Kazlauskas says. “It’s a collective action problem, where my data on its own isn’t that valuable, but a data pool with tens of thousands or millions of people is really valuable. Vana allows those pools to be built. It’s a win-win: Users get to benefit from the rise of AI because they own the models. Then you don’t end up in scenario where you don’t have a single company controlling an all-powerful AI model. You get better technology, but everyone benefits.”