Steve Ehrlich hosts Ben Fielding, CEO and co-founder of Jensen, a decentralized AI platform building infrastructure for distributed machine learning. Fielding has a PhD background in neural architecture search and AutoML, previously researching embarrassingly parallel techniques for optimizing deep neural networks.

Jensen launched Delphi, an information market platform built on the OP Stack layer-two, representing their first mainstream application. Unlike traditional prediction markets, Delphi allows anyone to create markets and query information bi-directionally.

The conversation explores how blockchain primitives enable machine learning models to maintain identity, communicate peer-to-peer, and establish programmatic trust for autonomous trading of information and computational resources.

From Neural Architecture Search to Distributed AI Infrastructure

Fielding's PhD research focused on neural architecture search using evolutionary algorithms that are 'embarrassingly parallel' - they can run on distributed devices without depending on each other during training

"Machine learning needs a horizontal scalability moment" similar to Google's MapReduce, which converted PageRank from vertical scaling to distributed processing across many devices - Ben

Jensen discovered blockchain through research papers on solving disputes between technical devices without human intervention, leading them to use consensus algorithms for programmatic trust

"We need to be able to take that security and translate it down into our technology" by building machine learning operations that blockchains can verify through consensus - Ben

Information Markets vs Traditional Prediction Markets

Information markets trade a resource required for machine learning - information itself, which is created by applying intelligence to raw data

"Information markets are bi-directional" - anyone can create markets to buy information, unlike prediction markets where only centralized companies create markets - Ben

Machine learning models can programmatically access markets, gather data, process it into information, and trade it within milliseconds rather than requiring human intermediaries

The system creates a "queryable model of the world" combining free market economics from Hayek with machine learning world models

Competing Against AI Giants Through Open Infrastructure

"Those companies will continue to make enormous amounts of money, create walled gardens, but they won't scale as far as a truly open technology can scale" - Ben

Jensen compares current AI companies to AOL in the early internet era - building walled gardens that eventually get commoditized while open infrastructure achieves greater scale

Open information markets can access "every single corner of the world" through incentivization, while centralized companies are only incentivized to gather big, profitable information

Real-World Applications and Market Mechanics

Example use case: A shop owner creates a market asking when construction on a nearby intersection will finish, incentivizing anyone with relevant information to trade

Information markets enable small players to hedge risks previously only available to large institutions - like orange farmers hedging drought risk without going to JP Morgan

"When information markets get to a certain scale, the financial markets are incentivized to come in and operate within the information markets" - Ben

The platform focuses on "the long tail of markets that do not exist right now" rather than competing for flagship markets like presidential elections

Token Economics and Platform Security

The AI token captures fees from information markets, with most fees going to market creators and a small portion used to buy and burn tokens, creating deflationary pressure

"There are no rewards in Delphi markets right now. The value that goes in is the value that comes out minus the fees" - economically balanced system prevents wash trading - Ben

All smart contracts are audited by Trail of Bits and the network uses two-of-two security from launch, avoiding the one-of-one vulnerabilities that affected recent exploits

Jensen uses machine learning oracles for market settlement instead of stake-weighted voting, claiming stronger truth commitments than existing prediction market platforms