Uploaded October 2026 | Updated November 2026, 1 day ago
AI is transforming how scientific research is conducted, but what happens when AI can do much of the research itself?
In this MIT Initiative on the Digital Economy Lunch Seminar, Chenhao Tan (Associate Professor of Computer Science and Data Science at the University of Chicago, Director of the Chicago Human and AI Lab) breaks down the future of AI in science across three acts.
Act 1 — Automating Verification: With NeurIPS receiving 40,000 submissions in a single year and human reviewers unable to keep up, AI agents can now replicate papers, check code, and find errors that humans miss. Tan shares results from replicating 168 ICML oral papers with these agents — only 34 reproduced over 40% of their claims. He also introduces "Living Science," a project that keeps research alive by re-running analyses with updated data using AI.
Act 2 — Autonomous AI Scientists: Tan introduces Neurico, a system that can take a research idea and write a full paper. He explores what happens when AI agents form communities, build on each other's work, and conduct open-ended exploration.
Act 3 — The Future of Science: Who decides what research gets done — and who pays for it? Tan argues that AI scientists cannot be accountable for resource allocation, meaning the role of humans will shift from production toward selection and evaluation. He introduces a "KitKat" analogy for the future of scientific workflows and poses a provocative question: what if AI gets so good that the best thing a human can do is just click "yes"?
Speaker Bio:
Chenhao Tan is an Associate Professor of Computer Science and Data Science at the University of Chicago, and directs the Chicago Human+AI Lab. He earned his PhD in Computer Science from Cornell University and dual bachelor’s degrees in computer science and economics from Tsinghua University. His research focuses on human-centered AI, communication & intelligence, and AI alignment. His work has been covered by major news media outlets, including the New York Times and the Washington Post. He also won a Sloan research fellowship, an NSF CAREER award, an NSF CRII award, a Google research scholar award, research awards from Amazon, IBM, JP Morgan, and Salesforce, a Facebook fellowship, and a Yahoo! Key Scientific Challenges award.
0:00 - Welcome & Introduction (Peyman Shahidi)
1:21 - Talk Overview: Three Acts of AI Scientists
3:07 - Act 1: The Paper Explosion Problem
6:55 - Evaluation Framework: Coherence, Reproducibility & Generalizability
10:54 - Key Finding: Failures Are Prevalent
18:39 - Scaling Up: Replicating ICML Oral Papers
22:06 - Concrete Example: The 8x Math Error
31:25 - Living Science: Keeping Research Alive
36:35 - Act 2: Autonomous AI Scientists (Neurico)
42:07 - Community of AI Scientists
44:48 - Act 3: The Mirage of Autonomous AI Scientists
48:14 - Science as Resource Allocation
50:52 - The KitKat Analogy
55:00 - Will Humans Just Click "Yes"?
56:30 - Closing Remarks & Q&A
AI is transforming how scientific research is conducted, but what happens when AI can do much of the research itself?
In this MIT Initiative on the Digital Economy Lunch Seminar, Chenhao Tan (Associate Professor of Computer Science and Data Science at the University of Chicago, Director of the Chicago Human and AI Lab) breaks down the future of AI in science across three acts.
Act 1 — Automating Verification: With NeurIPS receiving 40,000 submissions in a single year and human reviewers unable to keep up, AI agents can now replicate papers, check code, and find errors that humans miss. Tan shares results from replicating 168 ICML oral papers with these agents — only 34 reproduced over 40% of their claims. He also introduces "Living Science," a project that keeps research alive by re-running analyses with updated data using AI.
Act 2 — Autonomous AI Scientists: Tan introduces Neurico, a system that can take a research idea and write a full paper. He explores what happens when AI agents form communities, build on each other's work, and conduct open-ended exploration.
Act 3 — The Future of Science: Who decides what research gets done — and who pays for it? Tan argues that AI scientists cannot be accountable for resource allocation, meaning the role of humans will shift from production toward selection and evaluation. He introduces a "KitKat" analogy for the future of scientific workflows and poses a provocative question: what if AI gets so good that the best thing a human can do is just click "yes"?
Speaker Bio:
Chenhao Tan is an Associate Professor of Computer Science and Data Science at the University of Chicago, and directs the Chicago Human+AI Lab. He earned his PhD in Computer Science from Cornell University and dual bachelor’s degrees in computer science and economics from Tsinghua University. His research focuses on human-centered AI, communication & intelligence, and AI alignment. His work has been covered by major news media outlets, including the New York Times and the Washington Post. He also won a Sloan research fellowship, an NSF CAREER award, an NSF CRII award, a Google research scholar award, research awards from Amazon, IBM, JP Morgan, and Salesforce, a Facebook fellowship, and a Yahoo! Key Scientific Challenges award.
0:00 - Welcome & Introduction (Peyman Shahidi)
1:21 - Talk Overview: Three Acts of AI Scientists
3:07 - Act 1: The Paper Explosion Problem
6:55 - Evaluation Framework: Coherence, Reproducibility & Generalizability
10:54 - Key Finding: Failures Are Prevalent
18:39 - Scaling Up: Replicating ICML Oral Papers
22:06 - Concrete Example: The 8x Math Error
31:25 - Living Science: Keeping Research Alive
36:35 - Act 2: Autonomous AI Scientists (Neurico)
42:07 - Community of AI Scientists
44:48 - Act 3: The Mirage of Autonomous AI Scientists
48:14 - Science as Resource Allocation
50:52 - The KitKat Analogy
55:00 - Will Humans Just Click "Yes"?
56:30 - Closing Remarks & Q&A










