Applied AI Researcher | Interdisciplinary Human-Centered AI Leader | Innovator

Co-Founder with Katherine Elkins -- Kenyon Human-Centered AI Colab

Google Scholar | Academia.edu | LinkedIn | Twitter/X

• Overview
• Current Research (2025--2026)
• Published Research (2019--2024)
• Open-Source Software
• Grants & Affiliations
• Education & Curriculum
  • Courses
  • Student Projects
  • DEI & Impact Metrics
  • Program Timeline
• Industry & Entrepreneurship
• Collaborations
• Organizations
• Patents

Research scientist and AI leader specializing in interdisciplinary ML/AI applications across high-impact domains. Over the past decade, I've mentored over 300 original AI/ML projects spanning nearly every academic discipline -- from narrative and medicine to security and political science -- with work downloaded nearly 100k times by institutions including Stanford, MIT, Berkeley, Oxford, Amazon, and the Chinese Academy of Social Sciences.

I am currently serving as co-PI for the 25k-member Modern Language Assocation team in the NIST AI Consortium (former US AI Safety Institute) for LLM red/blue team testing. As of Jan 2026, I am also PI for one of 23 top AI research teams worldwide to have been awarded a Schmidt Sciences Humanity AI Virtual Instititue (HAVI) grant. As chief AI Architect, our interdisciplinary team from Columbia University, Berklee College of Music, and Louisiana State University is implementing a multilingual and multimodal archival digitial, restoration, and integrated graph database for early 1900s New Orlean Jazz collections. Previously, I lead research into tabular LLM recidivism prediction benchmarking for Notre Dame's Tech Ethics Lab on an IBM grant using multi-agent debates simluations of US bench trials. I have several papers for review at top-tier 2026 venues including ICML, FAccT, UAI, CogSci, TMLR, DMLR, JCP, and JAIR. Research focuses on AI safety and security, AI auditing and benchmarking, machine psychology/Theory of Mind, Affective AI, multi-agent workflow automation, and international AI policy and regulation.

Technical expertise spans LLM development and auditing, multimodal AI, automated agent frameworks, game theory simulations, and applying SOTA research to real-world problems in finance, healthcare, law, and narrative. I translate cutting-edge research into measurable products and services, having co-founded the world's first human-centered AI program bridging academia, research, and industry best practices.

Background includes Fortune 500 Director roles, Silicon Valley startup co-founder/CEO, and international experience across the US, Japan, and Latin America. Two US patents on network security. Published researcher with patents spanning gene therapy to AI policy.

Languages: English (native), Spanish (US Foreign Service Exam certified), Japanese (JLPT certified), French (professional working proficiency)

Back to Top

Schmidt Sciences Humanities and AI Virtual Institute (HAVI) (December 2025--Present) Principal Investigator and AI Architect -- 1 of 23 HAVI grants awarded to leading AI research teams worldwide (top 2-3%)

"Archival Intelligence: Can AI Rescue Endangered Archives" -- Co-PI and AI Architect leading an interdisciplinary team of scholars in jazz studies, musicology and CS from Columbia University, Berklee College of Music, and Louisiana State University. A team of AI researchers, archival scientists, jazz historians, and New Orleans cultural experts racing to preserve endangered small community archives using only smartphone photos, with AI to recover portions lost to damage. This 18-month pilot tackles the hardest cases by focusing on voices systematically excluded from historical archives: multilingual newspapers documenting Creole and Cajun communities as well as early jazz materials.

Pending

While LLMs are widely documented to be sensitive to minor prompt perturbations, we uncover a striking "Paradox of Robustness": instruction-tuned LLMs exhibit 110--300x greater resistance to narrative manipulation than human subjects. Using a controlled perturbation framework across healthcare, law, and finance, we find a near-zero effect size (Cohen's h = 0.003) compared to substantial human biases (h = 0.3--0.8).

Submitted to ICML 2026

Introduces Syntactic Framing Fragility (SFF), the first framework for quantifying decision consistency under logically equivalent syntactic transformations. SFF isolates syntactic effects via Logical Polarity Normalization (LPN) and provides the Syntactic Variation Index (SVI) as a CI/CD-ready robustness metric.

Submitted to FAccT 2026 (ACM Conference on Fairness, Accountability, and Transparency)

Audits seven language models for narrative vulnerability -- the tendency to let emotionally compelling but policy-irrelevant content shift rule-bound decisions. Demonstrates how aggregate and scenario-level analyses can diverge in ways consequential for deployment.

Pending

Submission for UAI 2026 (Conference on Uncertainty in Artificial Intelligence)

Introduces a game-theoretic evaluation framework grounded in quantal response equilibrium (QRE). Theory of Mind benchmarks for LLMs typically produce aggregate scores without theoretical grounding; our approach offers three methodological advances addressing whether high performance reflects strategic reasoning or surface-level heuristics.

Submitted to CogSci 2026 (Cognitive Science Society)

Uses large language model failures as a novel probe of pragmatic architecture. Compositional generalization tests reveal that models fail to bind features appropriately: accuracy drops 13--14% on novel context-utterance combinations, with 69% of predictions anchored to utterance patterns regardless of context.

Pending (Dataset Track)

Presents the Contextual Emotional Inference (CEI) Benchmark: 300 human-validated scenarios for evaluating how well LLMs disambiguate pragmatically complex utterances. Each scenario pairs situational context and speaker-listener roles with explicit power relations against an ambiguous utterance.

Pending | Extended version of AAAI-26 LaMAS accepted paper | IBM-Notre Dame Tech Ethics Lab Grant

Role-structured, multi-agent debate framework providing transparent and controllable test-time reasoning for recidivism prediction from tabular data. Benchmarked across ~90 unique model-strategy combinations on the NLSY97 dataset.

Pending

(In Review)

Tests whether LLM decision-making aligns with human values and whether that alignment reveals biases typical of human cognition. Measures confidence levels across moral scenarios while testing perturbations along syntactic framing and empathic backstory dimensions. Finds more performant models more closely mirror human biases, raising safety concerns about persuasion, manipulation, and deception.

ArXiv Preprint (December 2025)

Auditing 23 SOTA models over 14 ethical scenarios and four controlled framings (39,975 decisions), we find widespread inconsistency: many models reverse ethical endorsements solely due to syntactic polarity, with open-source models exhibiting over twice the fragility of commercial counterparts. Chain-of-thought reasoning substantially reduces fragility.

Accepted: AAAI-26 LaMAS (Singapore, January 2026) | IBM-Notre Dame Tech Ethics Lab Grant

Courtroom-style orchestration with explicit agent roles (prosecutor, defense, judge), 7-turn structured debate, and private reasoning strategies. Structured multi-agent debate provides more stable and generalizable performance compared to single-agent reasoning across ~90 model-strategy combinations.

Back to Top

• "Comparative Global AI Regulation: Policy Perspectives from the EU, China, and the US" -- SSRN and ArXiv
• "AIStorySimilarity: Quantifying Story Similarity Using Narrative for Search, IP Infringement, and Guided Creativity" -- ACL EMNLP/CoNLL, Miami
• "MultiSentimentArcs: Affective AI and Multimodal Diachronic Sentiment Analysis" -- Frontiers in Computer Science
• "In Search of a Translator: Using AI to Evaluate What's Lost in Translation" -- Frontiers in Computer Science
• "Near to Mid-term Risks and Opportunities of Open-Source Generative AI" -- ICML 2024, Vienna (with long form version)
• "Informed AI Regulation: Comparing the Ethical Frameworks of Leading LLM Chatbots" -- ArXiv

• "eXplainable AI with GPT4 for Story Analysis and Generation" -- Springer International Journal of Digital Humanities 5, 507--532
• "The Crisis of Artificial Intelligence: A New Digital Humanities Curriculum for Human-Centred AI" -- International Journal of Humanities and Arts Computing 17, 147--167

• "What the Rise of AI Means for Narrative Studies" -- Narrative 30, no. 1 (2022): 104--113
• "SentimentArcs: A Novel Method for Self-Supervised Sentiment Analysis of Time Series" -- ArXiv (2021)
• AI Improv DivaBot -- World's first live Human-AI improvisation with GPT (2021)
• "Can GPT-3 Pass a Writer's Turing Test?" -- Journal of Cultural Analytics 5, no. 2 (2020)
• How Artificial Intelligence Tells Stories: NLG and Narrative -- Narrative 2020, New Orleans
• "Can Sentiment Analysis Reveal Structure in a Plotless Novel?" -- ArXiv (2019)

• Lamping KG, Rios CD, Chun JA et al. "Intrapericardial Administration of Adenovirus for Gene Transfer." Am J Physiol. 1997. PMID: 9038951
• Flanagan JR, Chun J, Wagner JR. "Evolution of a Legacy System to a Web Patient Record Server." Proc AMIA Annu Fall Symp. 1996. PMID: 8947740

SentimentArcs is the open-source code for The Shapes of Stories by Katherine Elkins (Cambridge UP, 2022)

Back to Top

• SentimentArcs -- Largest ensemble for diachronic sentiment analysis; basis for The Shapes of Stories (Cambridge UP, 2022)
• MultiSentimentArcs -- First fully open-source diachronic multimodal sentiment analysis framework
• AIStorySimilarity -- Benchmark for long-text story similarity using narrative theory
• LLM Ethics Audit -- Ethics-based audit framework for leading LLM chatbots

Block Diagram for SentimentArcs Notebooks

Back to Top

Back to Top

I creatively apply industry best practices and state-of-the-art AI/ML techniques to high-impact interdisciplinary research. Our paper The Crisis of Artificial Intelligence, published in the International Journal of Humanities and Computing, outlines the challenges and opportunities AI presents to higher education, as well as the theory, structure, and goals of our human-centered AI approach.

Our human-centered curriculum is designed to be collaborative at multiple levels and interdisciplinary while maintaining academic rigor. Students from virtually every department across campus create a unique classroom learning and research environment. Research from our AI Colab has contributed to collaborative projects with institutions including Oxford, UC Berkeley, CMU, and Harvard Business School. We focus on productive collaborations to operationalize solutions to the "big questions" central to the Liberal Arts experience, including partnerships with industry leaders (Meta, IBM), government agencies (NIST AI Safety Institute), and non-profits (AI Alliance, MLA, US Chamber of Commerce, The Helix Center).

Sample projects using SentimentArcs:

• Literature: Sentiment Analysis and Nabokov's Pale Fire
• Translations: Cross-Linguistic Survey of Kafka's Trial
• TV Scripts: Storytelling and Sentiment Analysis in Shark Tank
• Medical Narratives: Five Stages of Grief in End-of-Life Memoirs
• Social Media: Sri Lankan Protestors and Political Change
• Elections: Quantifying Polarization around Election Denial

Sample projects from Frontiers in Generative AI (Fall 2024):

• AI-Powered MLB Roster Construction
• AI-Powered Appointment Prioritization for Resource-Poor Settings
• PitchAI: Streamlining Investment Banking Pitches Using LLMs
• Ancient Greek Parsing with AI: Agentic System
• AI Chatbot for College Disability Services Support
• Agentic Framework for Data-Driven Cricket Player Scouting

More projects: IPHS 200 | IPHS 300 | IPHS 391 | IPHS 484

Progress on gender diversity since 2017 (61% female as of Spring 2022)

Over 90% non-STEM majors (Kenyon College Institutional Research)

• 1992--99: IPHS established a computer lab for DH scholarship under Director Michael Brint
• July 2002: Katherine Elkins joined Kenyon, mentoring computational projects in IPHS
• May 2003: Launched Symantec Clientless VPN appliance as Director of Development; relocated from Silicon Valley
• March 2005: Proposed human-centered AI curriculum with Ewing Marion Kauffman Foundation grant
• August 2015: Formulated detailed interdisciplinary AI curriculum
• March 2017: Led Kenyon Team at HackOH5 Hackathon
• August 2017: First Programming Humanity course
• August 2018: First AI for the Humanities course; Katherine Elkins awarded NEH Distinguished Professorship
• August 2022: First Cultural Analytics course; first Industrial IoT collaboration
• 2023: Joined Meta Global Scholars Research Group
• 2024: Co-PI for NIST AI Safety Institute Consortium; Co-PI for IBM-Notre Dame Tech Ethics Grant; first Frontiers in Generative AI course
• 2026: Co-PI for Schmidt Sciences HAVI grant

Back to Top

• SafeWeb (Co-Founder/CEO): Built world's largest online privacy service (2B transactions in 8 months). Secured $11M from Wall Street funds + first security investment from In-Q-Tel (CIA). Pivoted to enterprise SSL VPN; drove $26M acquisition by Symantec at 35x revenue. (Products)
• Symantec (Director): Launched Clientless VPN Gateway in one quarter, under budget, with 2/3 team. 400K LOC release, zero attrition through acquisition integration.
• Matrix Absence Management (Director of IT): Architected B2B systems for 200+ high-tech clients (Intel, Dell, Cisco). Led $25M Unisys due diligence.
• Latin eVentures (CTO): Multi-million-dollar B2B portal-ERP across Latin America in 3 languages. Saved $500K in vendor renegotiations.

Back to Top

Government & Policy

• US White House AI Safety Policy / NIST AI Safety Institute (NIST, MLA)
• Comparative Global AI Regulation (SSRN, ArXiv, Journal of Cyber Policy)

Industry & Research

• Meta Global Scholars Research Group
• AI Alliance Essential AI Competencies Guide (Meta, IBM, Intel, US Chamber of Commerce)
• BWG Strategy -- Tech strategy and AI investing (NYC)
• IBM-Notre Dame Tech Ethics Lab

Academic

• Collaborative AI Safety and Open Source Policy (Oxford, Berkeley, UVa, Berklee/MIT)
• AI and the Future of Higher Education (Notre Dame, CMU, UC Berkeley, Harvard Business School)
• Computational Law, Ethics and XAI (ACM, Northwestern Law, IBM, Notre Dame)
• Multimodal Affective AI (Meta, NEH, EMNLP/CoNLL)
• Narrative and Affective AI (Yale, Harvard, Narrative Conferences)
• NLP and AI Research (ICML, ACL, ACM, CogSci, UAI, AAAI, FAccT)
• Theoretical Computer Science (McGill, UPitt, CMU)
• Psychology and Cognitive Science (Ohio State University)
• Literature, Language, Linguistics (MLA, Cambridge UP)
• Theatre and Film (Denison, NEH, LA Theatre)

Other Domains

• Medicine / FDA (AFDO/RAPS)
• Physics and Engineering Simulations (NAFEMS)
• IoT Predictive Maintenance (Kenyon AI Colab, Weyerhaeuser Paper)
• Computer and Network Security (USPTO, Symantec)
• Electronic Medical Records (U of Iowa, American Medical Informatics)
• Financial Information Systems (Bloomberg/McGraw-Hill Tokyo)
• Semiconductor Research (SEMATECH, Lawrence Berkeley Labs)

Back to Top

Lead AI Safety Researcher (2024--) for the Modern Language Association (25,000 members worldwide)

Announcement: The MLA is joining more than 200 of the nation's leading AI stakeholders in a US Department of Commerce initiative for trustworthy and safe AI. The MLA-sponsored team, led by Katherine Elkins and Jon Chun at Kenyon College, evaluates model capabilities with a special focus on linguistic edge cases and ethical frameworks.

Founder (January 2024--)

A volunteer, wholly virtual non-profit facilitating interdisciplinary AI collaboration between researchers and domain experts on safety, bias, explainability, ethics, and policy.

Executive Board Member (November 2022--)

Interdisciplinary roundtables drawing together leaders from arts, humanities, sciences, and technology. (Living in Difficult Times, November 2022)

• Interdisciplinary AI/ML Projects
• Kenyon Digital Colab
• Kenyon Interdisciplinary Human-Centered AI Program

Top 10 Institutions reading our AI DH Research -- digital.kenyon.edu/dh

Eurasian Institutions

Institutions from The Americas

Back to Top

1. Chun, Jon A., Stephen Dao Hui Hsu, James Noshir Hormuzdiar; Symantec Corp. "Intelligent Secure Data Manipulation Apparatus and Method." US Patent 7,730,528 B2 (June 1, 2010). (Patents folder)

2. Stephen Dao Hui Hsu, James Noshir Hormuzdiar, Jon A. Chun; Symantec Corp. "Method and Apparatus for Encrypted Communications to a Secure Server." US Patent 8,065,520 (November 22, 2011).

Back to Top