In this episode of The People’s AI, presented by Vana, we dive deep into the evolution of decentralized AI with Ramesh Raskar, Associate Professor at the MIT Media Lab and advisor to Vana.

We explore why decentralized AI isn’t a rival to centralized AI—but a natural next step. Ramesh outlines a five-phase framework for the evolution of intelligent systems, moving from the “mainframe era” to the emerging “web of AI agents.” Along the way, we discuss how decentralized systems can address problems of data ownership, incentive design, verifiability, and user experience.

Whether you’re interested in AI agents, blockchain, data sovereignty, or the ethics of AI infrastructure, this episode provides a high-level vision with real-world applications... from a credible AI researcher who's far from a kool-aid drinking crypto-bro.

Timestamps:

4:46 – Mainframe vs PC: A mental model for AI evolution

6:01 – From intranet to internet to the “web of AI agents”

9:13 – Limits of centralized AI

12:08 – Market forces, decentralization, and the economy of AI agents

13:21 – The four pillars of decentralized AI

16:41 – Proof of inference, verifiability, and trust in AI

18:14 – Rethinking UX in an agentic future

21:21 – Do we need standards? Or will AI agents just adapt?

25:27 – A case study of how Vana addresses the four pillars

28:21 – A look at the future: agent schools, decentralized labor, and AI socialization

32:58 – How Web3 could accelerate progress in entrenched sectors

33:37 – The biggest challenges to achieving this vision

MIT Media Lab's Decentralized AI Project

https://www.media.mit.edu/projects/decentralized-ai/overview/

Horizon 2023 | Ramesh Raskar | MIT Decentralized AI

https://www.youtube.com/watch?v=SSSffQsbFo4

About Vana:

Vana's vision is for user-owned AI through user owned-data. Its mission is to be the world's first open protocol for data sovereignty.

https://linktr.ee/vanahq

Vana on Twitter/X:

https://x.com/vana

• Vana ecosystem: vana.com

Vana introduces VRC 20 Token