World's Fair 2025
Real-time Experiments with an AI Co-Scientist - Stefania Druga, fmr. Google Deepmind
Overview
This talk introduces the concept of an AI co-scientist, analogous to a pair programmer but for real-world scientific experiments. The system integrates various sensors and cameras to collect empirical data in real-time, which an AI then analyzes to provide insights, generate hypotheses, and potentially accelerate scientific discovery. The presented system is built using open-source hardware and software, demonstrating a low-cost, accessible approach to AI-assisted experimentation.
Who should watch
- AI Engineers interested in building agentic systems for real-world applications.
- Product Managers and Builders exploring novel uses for AI beyond traditional software development.
- Researchers and Scientists looking for ways to manage data overload and accelerate hypothesis testing.
- Hobbyists and Educators interested in accessible, open-source tools for scientific exploration.
Key takeaways
- AI can act as a co-scientist, assisting with data analysis, hypothesis generation, and identifying blind spots in research.
- Real-time data collection from sensors and cameras, combined with AI analysis, enables dynamic experimentation and immediate feedback.
- The presented system utilizes open-source hardware and software, with a total cost under $300 and a build time of two weeks.
- Experiments demonstrated include crystal growth, where AI analysis of sensor data revealed that crystal formation occurs in bursts rather than gradually.
- The system architecture involves a React front-end, webhooks for input sources, and a backend communicating with an AI API (Gemini used in the demo).
- A unified context assembly dynamically builds prompts based on available modalities (text, voice, image, chat history) for the AI.
- The broader open-source ecosystem, including platforms like Jubilee Motion, supports the automation of laboratory tasks and scientific experiments.
- Future directions involve integrating simulations informed by real-world experiments to optimize conditions and accelerate discovery.
Notable quotes
*Think about pair programmers. How many of you use any form of copilot for coding? Right. So it's just like that but for doing things in the real world like science experiments.*
*The cost of all the parts is under $300 and it took me two weeks to build it and it's open source so you can play with it and hack it.*
*The insight I got from this is that the crystal formation is actually not gradual that it happens in bursts.*
Unofficial community note. Prefer the recording for nuance.