World's Fair 2026
Autonomous Agents for Scientific Tasks - Sina Shahandeh, Radicait
Overview
This talk explores the application of autonomous agents to complex scientific discovery tasks, moving beyond simpler coding puzzles or optimization problems. It highlights the necessity for agents to engage with real-world measurement data and employ a scientific method, including hypothesis generation, model implementation, and learning from failures. The core thesis is that significant advancements in agent performance for scientific tasks stem from formulating accurate hypotheses about physical systems and correctly implementing them.
Who should watch
- AI engineers and researchers working on agentic systems.
- Product managers and builders exploring applications of AI in scientific domains.
- Individuals interested in autonomous systems that can handle open-ended, real-world problems.
- Those seeking to understand how agents can assist with scientific hypothesis generation and validation.
Key takeaways
- Autonomous agents can tackle scientific discovery tasks that involve real-world measurement data, which are inherently open-ended.
- A scientific method, encompassing hypothesis formation, mathematical modeling, and execution on data, is crucial for agent success in these domains.
- Agents must be capable of searching over methods, priors, data preprocessing techniques, model classes, and hyperparameters.
- Learning from intermediate failures is a key component of the agent's problem-solving loop.
- An ontology-based memory system can significantly aid in hypothesis generation, a critical factor for agent performance.
- The presented demonstrations are derived from actual scientific problems encountered in industrial and applied research settings.
Notable quotes
*For an autonomous agent to assist with a scientific discovery task, the problems must come from real measurement data of the world and they are highly open-ended, requiring a scientific method in the solution loop.*
*Often, a step change in agent performance comes from forming an appropriate scientific hypothesis about how the physical system behaves, implementing that hypothesis correctly in a mathematical model, and executing it on the existing real data.*
Unofficial community note. Prefer the recording for nuance.