About

James A. Evans is a professor at the University of Chicago, serving as the Director of the Knowledge Lab and the Faculty Director of the Masters Program in Computational Social Science. His research focuses on the collective system of thinking and knowing, exploring how attention, intuition, ideas, and shared reasoning develop and influence processes of agreement, doubt, and understanding. He is particularly interested in innovation, examining how new ideas and practices emerge and the role social and technical institutions such as the Internet, markets, and collaborations play in collective cognition and discovery. Evans's work has concentrated on modern science and technology, but he also investigates other domains of knowledge including news, law, religion, gossip, hunches, and modes of machine and historical thinking. He supports the creation of novel observatories for human understanding and action through crowd sourcing, information extraction from text and images, and distributed sensors. His methodological approaches include machine learning, generative modeling, and social and semantic network representations to explore knowledge processes, scale interpretive and field methods, and develop alternatives to current discovery regimes. His research has been supported by prominent agencies such as the NSF, NIH, and the Air Force Office of Science Research, and has been published in leading journals and featured in major media outlets. Evans also founded and directs the Computational Social Science program at Chicago and teaches courses related to augmented intelligence, the history of modern science, science studies, and Internet and Society.

Research topics

• Computer Science
• Artificial Intelligence
• Sociology
• Political Science
• Social Science
• Data science
• Epistemology
• Mathematics
• Psychology
• Knowledge management
• Natural Language Processing
• Biology
• Engineering
• Positive economics
• Medicine
• Communication
• Economics
• Philosophy
• Operations research
• Social psychology
• Linguistics
• Mechanical engineering
• Theoretical computer science
• Management science

Selected publications

• Socio-Epistemic Bubbles and Tacit Confidence in Randomized Controlled Trials

  Social Studies of Science · 2026-03-12

  articleSenior authorCorresponding

  The paradigm of scientific medicine is among the most influential epistemic shifts in the past century, wherein randomized controlled trials (RCTs) represent the impartial arbiter of medical claims. Nevertheless, not all RCTs agree, and systematic reviews are invoked to reconcile them. We theorize how tacitness-beliefs, implicit assumptions, expectations, heuristics, unrecorded routines-accumulates within 'socio-epistemic bubbles', continuous regions of social density that decrease diversity and increase unwarranted certainty about healthcare studied by RCTs. To assess our theory, we analyze 20,117 meta-analyses extracted from 1,962 Cochrane systematic reviews. We find that being closer within 'social space' inscribed by scientific collaboration increases agreement across RCTs. Our analysis suggests that this amplified certainty can drive premature convergence affecting medical practice and population health. By increasing researchers' ability to perform an experiment in the way required to achieve an expected result, socio-epistemic bubbles represent the dark matter of the experimental process often obscured. Our findings imply hidden limitations associated with unmeasured social influence across the policy sciences, through which conflicting claims perpetuate and highlight the necessity of accounting for them to improve collective certainty.

  PublisherDOI

• Designing AI-Agents With Personalities: A Psychometric Approach

  Personality Science · 2026-01-09 · 1 citations

  articleOpen accessSenior author

  We introduce a methodology for assigning quantifiable and psychometrically validated personalities to AI-Agents using the Big Five framework. Across three studies, we evaluate its feasibility and limitations. In Study 1, we show that large language models (LLMs) capture semantic similarities among Big Five measures, providing a basis for personality assignment. In Study 2, we create AI-Agents using prompts designed based on the Big Five Inventory-2 (BFI-2) in different format, and find that AI-Agents powered by new models align more closely with human responses on the Mini-Markers test, although the finer pattern of results (e.g., factor loading patterns) were sometimes inconsistent. In Study 3, we validate our AI-Agents on risk-taking and moral dilemma vignettes, finding that models prompted with the BFI-2-Expanded format most closely reproduce human personality-decision associations, while safety-aligned models generally inflate ‘moral’ ratings. Overall, our results show that AI-Agents align with humans in correlations between input Big Five traits and output responses and may serve as useful tools for preliminary research. Nevertheless, discrepancies in finer response patterns indicate that AI-Agents cannot (yet) fully substitute for human participants in precision or high-stakes projects.

  PublisherDOI

• Simulating Subjects: The Promise and Peril of AI Stand-ins for Social Agents and Interactions

  2025-04-02 · 2 citations

  preprintSenior author

  Large Language Models (LLMs), through their exposure to massive collections of online text, learn to reproduce the perspectives and linguistic styles of diverse social and cultural groups. This capability suggests a powerful social scientific application – the simulation of empirically realistic, culturally situated human subjects. Synthesizing recent research in artificial intelligence and computational social science, we outline a methodological foundation for simulating human subjects and their social interactions. We then identify nine characteristics of current models that are likely to impair realistic simulation human subjects, including atemporality, social acceptability bias, uniformity, and poverty of sensory experience. For each of these areas, we discuss promising approaches for overcoming their associated shortcomings. Given the rate of change of these models, we advocate for an ongoing methodological program on the simulation of human subjects that keeps pace with rapid technical progress.

  PublisherDOI

• Simulating Subjects: The Promise and Peril of AI Stand-ins for Social Agents and Interactions

  2025-03-31

  preprintOpen accessSenior author

  Large Language Models (LLMs), through their exposure to massive collections of online text, learn to reproduce the perspectives and linguistic styles of diverse social and cultural groups. This capability suggests a powerful social scientific application – the simulation of empirically realistic, culturally situated human subjects. Synthesizing recent research in artificial intelligence and computational social science, we outline a methodological foundation for simulating human subjects and their social interactions. We then identify nine characteristics of current models that are likely to impair realistic simulation human subjects, including atemporality, social acceptability bias, uniformity, and poverty of sensory experience. For each of these areas, we discuss promising approaches for overcoming their associated shortcomings. Given the rate of change of these models, we advocate for an ongoing methodological program on the simulation of human subjects that keeps pace with rapid technical progress.

  PublisherOA PDFDOI

• Using human mobility data to quantify experienced urban inequalities

  Nature Human Behaviour · 2025-02-17 · 38 citations

  reviewSenior author
  PublisherDOI

• Evaluation of a novel capsule sponge triage pathway for patients routinely referred with reflux symptoms: safety, long-term outcomes and impact on endoscopy from a large volume single site cohort over 4 years

  Frontline Gastroenterology · 2025-06-10 · 2 citations

  article

  Objectives To prospectively evaluate a real-world pathway for routine reflux investigation using non-endoscopic capsule sponge (CS) triage with respect to impact on endoscopy, histology and long-term safety. Methods Patients with reflux symptoms received CS as part of a triage pathway in a UK hospital over 4 years. Only patients with abnormal CS, inadequate samples or ongoing symptoms had endoscopy. Clinical outcomes and patient evaluation were assessed. Results 871 patients had CS, 87.6% providing adequate samples. 540 (62%) did not require endoscopy and 82% were discharged. Patients with normal CS were significantly more likely to have minor findings/normal endoscopy (p<0.001). 86 (9.9%) had abnormal CS: 1 patient was diagnosed with oesophageal cancer, 2 with Barrett’s dysplasia and 34 with Barrett’s oesophagus (BO). Malignant/premalignant pathology and endoscopic yield of major findings were significantly increased in abnormal CS (p<0.001). The positive predictive value for histological BO in abnormal CS test was 43.5%, NPV 98.2%. Follow-up was 2078 patient years, median 27.24 months (range 12–48). Gastric cancer was diagnosed in one patient 2 weeks after normal CS due to concerning symptoms at appointment. Only six (1.8%) patients were found to have Barrett’s/atrophic gastritis in those with a negative CS who had endoscopy for persistent symptoms. 97.5% patients found CS acceptable and 94% would have another CS. Conclusion CS pathway is acceptable to patients, safely identifies pathology, augments the proportion of significant endoscopic diagnoses while ensuring appropriate endoscopy and discharge of low-risk patients. These findings could inform a patient-friendly and resource-efficient service for routine reflux and aid appropriate endoscopy utilisation.

  PublisherDOI

• Designing AI-Agents with Personalities: A Psychometric Approach

  2025-09-22

  articleOpen accessSenior author

  We introduce a methodology for assigning quantifiable and psychometrically validated personalities to AI-Agents using the Big Five framework. Across three studies, we evaluate its feasibility and limitations. In Study 1, we show that large language models (LLMs) capture semantic similarities among Big Five measures, providing a basis for personality assignment. In Study 2, we create AI-Agents using prompts designed based on the Big Five Inventory-2 (BFI-2) in different format, and find that AI-Agents powered by new models align more closely with human responses on the Mini-Markers test, although the finer pattern of results (e.g., factor loading patterns) were sometimes inconsistent. In Study 3, we validate our AI-Agents on risk-taking and moral dilemma vignettes, finding that models prompted with the BFI-2-Expanded format most closely reproduce human personality-decision associations, while safety-aligned models generally inflate ‘moral’ ratings. Overall, our results show that AI-Agents align with humans in correlations between input Big Five traits and output responses and may serve as useful tools for preliminary research. Nevertheless, discrepancies in finer response patterns indicate that AI-Agents cannot (yet) fully substitute for human participants in precision or high-stakes projects.

  PublisherOA PDFDOI

• Missing vs. Unused Knowledge Hypothesis for Language Model Bottlenecks in Patent Understanding

  ArXiv.org · 2025-05-18

  preprintOpen accessSenior author

  While large language models (LLMs) excel at factual recall, the real challenge lies in knowledge application. A gap persists between their ability to answer complex questions and their effectiveness in performing tasks that require that knowledge. We investigate this gap using a patent classification problem that requires deep conceptual understanding to distinguish semantically similar but objectively different patents written in dense, strategic technical language. We find that LLMs often struggle with this distinction. To diagnose the source of these failures, we introduce a framework that decomposes model errors into two categories: missing knowledge and unused knowledge. Our method prompts models to generate clarifying questions and compares three settings -- raw performance, self-answered questions that activate internal knowledge, and externally provided answers that supply missing knowledge (if any). We show that most errors stem from failures to deploy existing knowledge rather than from true knowledge gaps. We also examine how models differ in constructing task-specific question-answer databases. Smaller models tend to generate simpler questions that they, and other models, can retrieve and use effectively, whereas larger models produce more complex questions that are less effective, suggesting complementary strengths across model scales. Together, our findings highlight that shifting evaluation from static fact recall to dynamic knowledge application offers a more informative view of model capabilities.

  PublisherOA PDFDOI

• The (Short-Term) Effects of Large Language Models on Unemployment and Earnings

  ArXiv.org · 2025-09-19 · 1 citations

  preprintOpen accessSenior author

  Large Language Models have spread rapidly since the release of ChatGPT in late 2022, accompanied by claims of major productivity gains but also concerns about job displacement. This paper examines the short-run labor market effects of LLM adoption by comparing earnings and unemployment across occupations with differing levels of exposure to these technologies. Using a Synthetic Difference in Differences approach, we estimate the impact of LLM exposure on earnings and unemployment. Our findings show that workers in highly exposed occupations experienced earnings increases following ChatGPT's introduction, while unemployment rates remained unchanged. These results suggest that initial labor market adjustments to LLMs operate primarily through earnings rather than worker reallocation.

  PublisherOA PDFDOI

• Learning from one and only one shot

  npj Artificial Intelligence · 2025-07-14 · 4 citations

  articleOpen accessSenior author

  Abstract Humans can generalize from only a few examples and from little pretraining on similar tasks. Yet, machine learning (ML) typically requires large data to learn or pre-learn to transfer. Motivated by nativism and artificial general intelligence, we directly model human-innate priors in abstract visual tasks such as character and doodle recognition. This yields a white-box model that learns general-appearance similarity by mimicking how humans naturally “distort” an object at first sight. Using just nearest-neighbor classification on this cognitively-inspired similarity space, we achieve human-level recognition with only 1–10 examples per class and no pretraining. This differs from few-shot learning using massive pretraining. In the only-few-shot regime of MNIST, EMNIST, Omniglot, and QuickDraw benchmarks, we outperform both modern neural networks and classical ML. For unsupervised learning, by learning the non-Euclidean, general-appearance similarity space in a k -means style, we achieve multifarious visual realizations of abstract concepts by generating human-intuitive archetypes as cluster centroids.

  PublisherOA PDFDOI

Recent grants

• Tracing Influence & Predicting Impact in Science

  NSF · $507k · 2013–2017

• Collaborative Research: Scaling Insight into Science: Assessing the value and effectiveness of machine assisted classification within a statistical system

  NSF · $196k · 2014–2017

• A Knowledge Base for Clinically Relevant Genes and Variants

  NIH · $9.0M · 2013–2018

• NIH Grant R01LM010132

  NIH · $1.2M · 2012

• Collaborative Research: Understanding Team Success and Failure

  NSF · $343k · 2018–2022

Frequent coauthors

• Andrey Rzhetsky

  University of Chicago

  36 shared

• Jacob G. Foster

  23 shared

• Nati Srebro

  18 shared

• JOE M. PHILLIPS

  Emory University

  16 shared

• Nandana Sengupta

  Indian Institute of Technology Delhi

  16 shared

• Lingfei Wu

  University of Pittsburgh

  14 shared

• Dashun Wang

  Northwestern University

  11 shared

• Ishanu Chattopadhyay

  10 shared