Jiacheng Miao

I am building Agentic Science as a HAI Postdoctoral Fellow at Stanford, working with Jonathan Pritchard and James Zou.

I am currently interested in:

• How do we make scientific knowledge agent-native?
• How might Agentic AI change or distort science?

My work has been featured by Anthropic, The Scientist, and Stanford Medicine, and recognized with the NOMIS & Science Young Explorer Award.

Email me at jcmiao[at]stanford.edu.

Google Scholar  /  Github  /  X

---

Selected Work

Agentic Science

• Agentic Bootstrap preprint, 2026

  Shows how AI agents can reach different scientific conclusions from the same data, and introduces the m-value to place a reported conclusion within a multiverse of plausible analyses.

• Paper2Agent [code] > 2.3k stars preprint, 2025

  Turns scientific knowledge into executable paper agents that can collaborate autonomously.

AI4Science

• POP-GWAS Nature Genetics, 2024

  Enables reliable genetic discovery using AI-predicted outcomes. (My favorite result is Theorem 2—an AI-powered Gauss-Markov theorem.)

• PSPA Journal of Machine Learning Research, 2025

  Introduces new theory for reliable statistical inference with AI-predicted data.

• PSPS NeurIPS, 2024

  Lets researchers safely use AI predictions with almost any existing data-analysis method.

Human genetics

• Nature & Nurture Science, 2025

  Explains how statistics and AI can help us understand how genes and environments work together to shape health.

• M-Learner In revision, Nature Genetics, 2025

  Uses genetic information to predict why the same treatment may help some people more than others.

• PIGEON Nature Human Behaviour, 2025

  Measures how genes and environments work together using large-scale genetic summary data.

• X-Wing Nature Communications, 2023

  Improves genetic risk prediction across people with different ancestral backgrounds.

• QUAIL PNAS, 2022

  Finds genetic variants that affect how much a trait varies across people, not just its average.