AutoDiscovery

AutoDiscovery is an AI-powered data exploration tool developed by the Allen Institute for AI (Ai2) that autonomously analyzes datasets to surface genuinely surprising findings. It leverages Bayesian surprise to identify discoveries that meaningfully change existing knowledge, rather than simply confirming what is already expected. The tool is designed to inspire new lines of inquiry by challenging assumptions embedded in data. AutoDiscovery is part of the AstaLabs experiment initiative at Ai2.

• Bayesian Surprise Engine: AutoDiscovery quantifies how much each finding updates prior beliefs, ensuring only genuinely novel insights are surfaced.
• Autonomous Exploration: The tool independently runs experiments across your dataset without requiring manual hypothesis specification.
• Dataset Analysis: Upload or connect your dataset and AutoDiscovery will systematically explore variable relationships and patterns.
• Shared Sessions: Explore example sessions (such as the National Longitudinal Survey dataset) to understand how the tool surfaces insights before using your own data.
• Sign-in Access: Sign in to get started and run your own discovery sessions on custom datasets.
• Research-Oriented Design: Built by Ai2 researchers, the tool is tailored for academic and scientific data exploration workflows.