Jacob E. Thomas

Jacob E. Thomas is an epidemiologist, data scientist, and AI engineer who builds tools for evaluating historical AI personas, researching public discourse, and investigating conflict. His central concern is how persuasive systems distort evidence while appearing trustworthy.

Thomas earned an MA in clinical psychology from Columbia University and a PhD in health behavior from the University of Texas at Austin. His research examined depression and nicotine dependence and psychological factors affecting burn survivors’ community integration. He later led analytics research in recruitment marketing and was affiliated with Results Generation at AI Engineer World’s Fair 2026.

• Miranda distortion: Thomas argues that historical AI personas often reproduce culturally dominant portrayals instead of the documentary record. An Alexander Hamilton shaped by Broadway can sound convincing while flattening contested evidence about slavery; conventional personality benchmarks and automated judges reward that familiarity instead of detecting its inaccuracies. Preference optimization can compound the problem when evaluators share the same assumptions.

• Epistemic simulation: His alternative anchors personas to primary documents and specific historical moments, with domain experts defining evaluation standards. The configuration combines source materials, prompts, temporal constraints, an off-the-shelf model, and human curation. Keeping documents in context preserves provenance and makes personas inspectable, revisable, and accessible without specialized training infrastructure. His open-source COMPANION framework implements this approach.

• The Prism Experiment: Thomas designed a preregistered study comparing primary-source grounding, modern biography, and unanchored generation across four periods of Abraham Lincoln’s life. Historian-authored questions probe issues such as executive war powers and slavery; responses are assessed for anachronism detection, documentary consistency, and contextual plausibility. The study is a proposed protocol, not a completed finding. Domain specialists build the evaluation and review consequential cases without supervising every interaction.

• The Threshing Floor: Thomas created an open-source Reddit research platform that collects, organizes, analyzes, and exports public conversations. Its design emphasizes provenance, default username anonymization, and research ethics; when automated access became unreliable, he adopted browser-based manual imports instead of evasive scraping.

• IranWar.ai: As principal investigator of IranWar.ai, Thomas leads an open-source project organizing information about the 2026 U.S.–Iran conflict. Its public codebase exposes research materials, communicates uncertainty, invites corrections, and warns against relying on the dashboard for financial, military, or life-safety decisions.

Thomas’s interest in trustworthy personas also reaches elder care: an attempt to support communication with someone experiencing advanced dementia underscored the danger of substituting a plausible synthetic likeness for the documentary record needed to represent a real person faithfully.