Jon A. Chun
The hardest questions about AI can't be answered inside one discipline, one institution, or one sector, so Jon Chun builds the collaborations that cross all of them. For a decade he has co-built Human-Centered AI: turning the oldest human questions into quantifiable, testable models with frontier AI, across academia, industry, and government.
He co-leads the team representing the 25,000-member Modern Language Association at NIST CAISI (LLM evaluation and red-teaming) and is co-PI on Schmidt Sciences HAVI, building open-source AI to rescue endangered cultural archives. With Katherine Elkins he co-founded the world's first Human-Centered AI curriculum and lab at Kenyon (2016), and earlier co-founded and led SafeWeb.
Harder to fool than we are
Instruction-tuned language models are 110–300× more resistant to narrative manipulation than people, measured across healthcare, law, and finance. The systems we worry about being fooled are, on this axis, far harder to fool than we are. Read the research →
Work the field builds on
When FAccT 2025 researchers mapped how the field defines AI "openness," they named the ICML 2024 position paper Jon co-authored (an oral presentation, top 2% of submissions) one of three canonical openness frameworks. Trace the reception →
Three seats at once
Held at the same time: uncommon at any career stage, and rare from a small liberal-arts college.
Co-leads the team representing the 25,000-member Modern Language Association, the only humanities-led team in the federal AI-safety consortium.
Co-PI on Archival Intelligence, 1 of 23 teams selected worldwide from 600+ applications for the inaugural Humanities and AI Virtual Institute.
Meta Open Innovation AI Research Community; multi-year BWG forums with analysts, investors, and operators, plus a deployed multi-agent product in the market.
| Sector ↓ / Mode → | Built | Measured | Governed | Taught |
|---|---|---|---|---|
| Academia | First HCAI curriculum (2016) | SentimentArcs; ICML 2024 oral | Comparative AI regulation | 400+ mentored projects |
| Industry | SafeWeb; 2 US patents | BWG multi-agent product | SafeWeb → Symantec ($26M) | Open materials, 107K+ downloads |
| Government | NIST CAISI red-team tooling | Ethics-based LLM audits | MLA seat at NIST CAISI | Standards working groups |
| Non-profit | Archival Intelligence (New Orleans) | FATE audit (Notre Dame–IBM) | Human-Centered AI Lab | Helix Center AI+Medicine |
Featured: Co-PI on Archival Intelligence, 1 of 23 Schmidt Sciences HAVI teams worldwide from 600+ applications, building open AI to rescue New Orleans' endangered Creole/Cajun newspapers and early jazz artifacts. See the full collaboration map →
Find your way in
Quotable results (the 110–300× robustness paradox), named co-panelists, and where Jon has appeared in the press.
The publication record, the federal + foundation + industry triple, and how the work is cited across 70+ countries.
SafeWeb's $26M exit and first In-Q-Tel investment, two US patents, and a deployed multi-agent product.
A decade of undergraduates doing original AI research, open to every division of the liberal arts.
Built, shipped, and built on
The largest online-privacy service of its era, with roughly 2 billion transactions in 8 months. The first security investment from In-Q-Tel, the CIA-affiliated venture fund; two US patents on early SSL / clientless-VPN appliances; acquired by Symantec in 2003 for $26M, at 35× trailing revenue. The full venture story →
A multi-agent AI product deployed in the market, built through multi-year BWG Global forums with analysts, investors, and operators, alongside the Meta Open Innovation AI Research Community (since 2023). Frontier research grounded in commercial deployment. See the industry work →
The best 200+ AI CoLab projects on Digital Kenyon: 107,000+ downloads across 4,760 institutions in 198 countries, plus open tooling like SentimentArcs and MultiSentimentArcs (Frontiers in Computer Science, 2024). Inside the AI CoLab →
Work done first
Dates establish priority; the Reception page documents the uptake that followed. Most of the research below is co-authored with Katherine Elkins.
- 1996Co-architected one of the first web-based electronic health record systems at a major US teaching hospital (University of Iowa).
- 2001The first security investment from In-Q-Tel (the CIA-affiliated venture fund), for SafeWeb.
- 2016Co-founded the world's first Human-Centered AI curriculum and lab at Kenyon, with Katherine Elkins; the field's first peer-reviewed account followed in 2023.
- 2019One of the first methodologies for sentiment analysis of narrative, introducing "middle reading," between distant and close reading.
- 2020The first writer's Turing test of a large language model: "Can GPT-3 Pass a Writer's Turing Test?"
- 2021SentimentArcs, the ensemble method for narrative sentiment analysis, weighing dozens of models against one another, and the first use of dynamic time warping in diachronic sentiment analysis, which lets story arcs of unequal length be compared by distance. It also reported that state-of-the-art transformers can struggle to find narrative arcs.
- 2023One of the first explainable-AI methods built for narrative rather than image classification, and the solution that lets stories of unequal length be compared on shared coordinates.
- 2024The first multimodal measure of sentiment-arc coherence in long-form film: MultiSentimentArcs (Frontiers in Computer Science).
- 2024The first ethics-based audit of moral reasoning in deployed LLMs, and the first systematic EU–China–US regulatory comparison after the EU AI Act.
Breadth, quantified
- 3 seatsFederal (NIST CAISI), foundation (Schmidt HAVI), and industry (Meta / BWG), held at the same time
- Only oneHumanities-led team in the NIST CAISI consortium, representing the 25,000-member MLA
- ICML 2024Oral presentation, top 2% of submissions
- 110–300×How much harder instruction-tuned LLMs resist narrative manipulation than people
- 61% / 90%Women / non-STEM students in the gateway course, broadening who does serious AI work
- 70+Countries where the research is cited (see Reception)
Cited, quoted, and built on
"GPT-3 writes better than many people."
Luciano Floridi & Massimo Chiriatti, Minds and Machines (2020), citing Elkins & Chun's writer's Turing test"Debating the system of values we wish these tools to align with is the first step."
Tanya Klowden & Terence Tao (Fields Medalist), engaging Chun & Elkins, IJHAC (2023)"The human-centered AI curriculum at Kenyon encompassed the true essence of a liberal arts education: using a wide range of academic disciplines to discuss world-changing contemporary issues."
Raul Romero, Kenyon College Class of 2022- Forbes
- NPR
- Christian Science Monitor
- Al Jazeera
- Chronicle of Higher Education
See full scholarly reception → · See press coverage → · Grants & recognition →
Common questions
What is Human-Centered AI as Jon Chun practices it?
It uses state-of-the-art AI and real engineering to turn the oldest human questions (what it means to be human, what a good life and a good society are) into quantifiable, testable models. It builds, measures, and governs. It is distinct from human-centered UI/UX design, from non-technical AI-ethics or STS critique, and from low-code digital humanities, and it works through radical collaboration across disciplines, industry, government, and non-profits. Jon co-founded the world's first such curriculum and lab with Katherine Elkins at Kenyon's AI CoLab in 2016.
What does Jon Chun do now?
He co-leads the team representing the Modern Language Association at NIST CAISI (LLM evaluation and red-teaming) and is co-PI on Schmidt Sciences HAVI (Archival Intelligence), one of 23 teams selected worldwide from 600+ applications.
What was SafeWeb?
An internet privacy company Jon co-founded in 2000 and later led as CEO. It received the first security investment from In-Q-Tel, a nonprofit strategic investment firm affiliated with the CIA, and was acquired by Symantec in 2003 for $26 million.
What is the confidence-scoring method for auditing language models?
Introduced in "Informed AI Regulation," it measures how firmly a model commits to a moral judgment versus hesitates, giving a way to compare normative certainty across models. It has since been applied across 1,613 social-decision scenarios (COLING 2025) and included among 69 foundational works in the AAAI 2026 "Beyond Verdicts" survey. It is co-authored with Katherine Elkins.