The Robot Revolution will be Powered by Decentralized Data, w/ PrismaX Founder Bayley Wang

Robots need decentralized data before they can work

PrismaX is betting that robotics is bottlenecked by data rather than hardware. Robots can already run and flip, but they cannot reliably work in messy real settings, and no central team can hire enough people to record every kitchen and warehouse. PrismaX wants a decentralized network where ordinary participants upload videos or teleoperate robots, with DAOs and a token coordinating access to the resulting data.

Takeaways

• Robotics hardware advanced before useful software. Wang says actuation work from about 2010 to 2019 lets robots do parkour and backflips in videos, but they are absent from the wild because automation remains unsolved. The physical world is diverse and demands accuracy far beyond a chatbot.
• Handwritten rules fail in unpredictable real world. Some teams use machine vision to list objects and a VLM to plan tasks, but those rules break when authors did not expect a scenario. LLMs rarely fail so badly because they learn from a cross-section of the internet.
• Robot data cannot scale like text. A text model can download the entire internet. A robotics model needs hundreds or thousands of people to collect data. Wang says robotics companies pay teleoperators $50 an hour, and billion-dollar companies run eight-person data collection teams.
• Centralized data collection bakes in manager bias. Even a project with money to hire a million people would not know who to hire, Wang says. Managers' bias produces a robot tailored to a committee of 20 or 30 people. Diverse datasets and operator environments make models robust.
• Robots will augment workers, not replace them. Wang predicts people will become the back end of robots, handling teleoperation and reliability. One person will do the work of seven or 10, so robots raise output rather than eliminate the human role.
• Kitchens and restaurants lead near-term robot use. Wang is most excited about kitchens and restaurants, where automation lets restaurants serve more interesting dishes while humans have more time to be creative. She expects general domestic chores to take about a decade, with high-value cleaning sooner.

PrismaX's robot data playbook

1. Participants upload videos of repetitive tasks such as sheet folding and restocking to add diversity across environments.
2. Others remotely teleoperate robots, a task Wang says robotics companies pay $50 an hour for and cannot scale with small teams.
3. Guilds organize operators and buy robots that mine incentives, which lets robotics companies sell hardware at cost instead of raising a manufacturing round.
4. DAOs manage access to and sales of the data, so PrismaX does not have to manage all of it itself.