About

Bernard Koch is an Assistant Professor in the Department of Sociology at the University of Chicago. His research explores the mechanisms underlying cultural diversification and collapse across various fields such as AI, music, and science. He employs a blend of computational methods, Bayesian statistics, and qualitative interviews to develop evolutionary theories and detailed historical narratives. His current work focuses on how different evaluation systems, including peer review and benchmarking, influence the trajectories of scientific fields, with attention to their ethical, epistemic, and cultural implications. Through historical case studies like AI's convergence on deep learning and the long-standing entanglement of social psychology with controversial racial hereditarian research, he sheds light on the strengths and weaknesses of these evaluation systems. Beyond evaluation mechanisms, Koch investigates the processes that drive the evolution of cultural ideas, developing theories and Bayesian models to explain these dynamics across domains such as music genres and news cycles. His work has been published in various venues including Sociology, Computer Science, Bioethics, and general interest outlets, and has been featured by TIME, Venture Beat, and Mozilla.

Research topics

• Artificial Intelligence
• Computer Science
• Machine Learning
• Data Mining
• Sociology
• Information Retrieval
• Marketing
• Econometrics
• Medicine
• Mathematics
• Ecology
• World Wide Web
• Data science
• Epistemology
• Cognitive science
• Evolutionary biology
• Psychology
• Statistics
• Anthropology

Selected publications

• The Social Structure of Scientific Evaluation: AI, Benchmarking, and the Deep Learning Monoculture

  2026-01-22 · 2 citations

  preprintOpen access1st authorCorresponding

  Evaluation systems are central organizing institutions in science that coordinate consensus and drive epistemic trajectories. Scientific fields have traditionally relied on "organic" evaluation systems (e.g., peer review, citation) where consensus emerges gradually across multiple epistemic values. This paper highlights artificial intelligence research (AIR) as a potent counterpoint to this model. Drawing on interviews with key actors, computational analyses, and archival materials spanning AIR’s history (1956–2021), we examine how AI evolved from a discipline with weak organic evaluation into a field driven by benchmarking, a “formal” evaluation system that defines progress quantitatively as state-of-the-art accuracy on commercial tasks. We demonstrate that benchmarking came to dominate through an intricate symbiosis with deep learning: benchmarking rewards accuracy, which large-scale deep learning uniquely excelled at, while deep learning’s opacity made organic evaluation increasingly difficult. This symbiosis restructured the field organizationally, epistemically, and materially into a “monoculture” dedicated to scaling. While enabling breakneck progress, monoculture discouraged exploration of alternatives with different epistemic strengths. As AI spreads to other knowledge fields (from science to law to art) benchmarking will accompany it. Our findings thus highlight the risk that formalization of evaluation can lead to monoculture in other creative domains.

  PublisherOA PDFDOI

• The Evolutionary Dynamics of Cultural Change (As Told Through the Birth and Brutal, Blackened Death of Metal Music)

  SocArXiv (OSF Preprints) · 2026-01-13

  preprintOpen accessSenior author

  How does culture change? We unify disconnected explanations of change that focus either on individuals or on public culture under a theory of cultural evolution. By shifting our analytical lens from actors to public cultural ideas and object, our theory can explain change in cultural forms over large and long frames of analysis using formal evolutionary mechanisms. Complementing this theory, the paper introduces a suite of novel methods to explain change in the historical trajectories of populations of cultural ideas/objects (e.g., music groups, hashtags, laws, technologies, and organizations) through diversification rates. We deploy our theory and methods to study the history of Metal Music over more than three decades, using a complete dataset of all bands active between 1968 and 2000. Over the course of its history, we find strong evidence that the genre has been fundamentally shaped by competition between ideas for the cognitive resources actors can invest in learning about and reproducing this cultural form over time. Extensive tutorials for the methods are available at http://www.dysoc.org/cesmodules/diversification_module/tutorials.

  Publisher

• The Evolutionary Dynamics of Cultural Change (As Told Through the Birth and Brutal, Blackened Death of Metal Music)

  2026-01-13

  articleOpen access1st authorCorresponding

  How does culture change? We unify disconnected explanations of change that focus either on individuals or on public culture under a theory of cultural evolution. By shifting our analytical lens from actors to public cultural ideas and object, our theory can explain change in cultural forms over large and long frames of analysis using formal evolutionary mechanisms. Complementing this theory, the paper introduces a suite of novel methods to explain change in the historical trajectories of populations of cultural ideas/objects (e.g., music groups, hashtags, laws, technologies, and organizations) through diversification rates. We deploy our theory and methods to study the history of Metal Music over more than three decades, using a complete dataset of all bands active between 1968 and 2000. Over the course of its history, we find strong evidence that the genre has been fundamentally shaped by competition between ideas for the cognitive resources actors can invest in learning about and reproducing this cultural form over time. Extensive tutorials for the methods are available at http://www.dysoc.org/cesmodules/diversification_module/tutorials.

  PublisherDOI

• Genetics and Scientific Values: Aaron Panofsky, Kushan Dasgupta, Nicole Iturriaga, and Bernard Koch Reply

  The Hastings Center Report · 2025-05-01 · 1 citations

  letterSenior author

  This letter responds to the letter by Jan te Nijenhuis, Bryan J. Pesta, and John G. R. Fuerst in the May-June 2025 issue of the Hastings Center Report.

  PublisherDOI

• LLM Social Simulations Are a Promising Research Method

  ArXiv.org · 2025-04-03 · 4 citations

  preprintOpen access

  Accurate and verifiable large language model (LLM) simulations of human research subjects promise an accessible data source for understanding human behavior and training new AI systems. However, results to date have been limited, and few social scientists have adopted this method. In this position paper, we argue that the promise of LLM social simulations can be achieved by addressing five tractable challenges. We ground our argument in a review of empirical comparisons between LLMs and human research subjects, commentaries on the topic, and related work. We identify promising directions, including context-rich prompting and fine-tuning with social science datasets. We believe that LLM social simulations can already be used for pilot and exploratory studies, and more widespread use may soon be possible with rapidly advancing LLM capabilities. Researchers should prioritize developing conceptual models and iterative evaluations to make the best use of new AI systems.

  PublisherOA PDFDOI

• Confronting the “Weaponization” of Genetics by Racists Online and Elsewhere

  The Hastings Center Report · 2024-12-01 · 9 citations

  articleOpen accessSenior author

  Genomics research is regularly appropriated in social and political contexts to publicly legitimize unjust and malicious political views, policies, and actions. In recent years, there have been high-profile cases of mass shooters, public intellectuals, and political insiders using genomics findings to convince audiences that deadly force and coercive policies against racial minorities are warranted. To create a just genomics, geneticists must consider what makes their research so attractive and adaptable for the legitimization of unjust ends and what they can do to counter such appropriations. We offer insights and recommendations drawing from our research into the many ways online white nationalist and far-right political movements mobilize genetics research to promote their racist, sexist, antisemitic, and homophobic views. First, geneticists should identify and change routine research practices that feed eugenic thinking. Second, geneticists should adopt creative extra-scholarly communication efforts to counter the use of their field's research that occurs in nonscholarly spaces. Third, we identify permissive epistemological and professional practices within the genetics field that have enabled such unjust appropriations to thrive, and we recommend strategies for institutional reform.

  PublisherOA PDFDOI

• A Primer on Deep Learning for Causal Inference

  Sociological Methods & Research · 2024-08-16 · 7 citations

  article1st authorCorresponding

  This primer systematizes the emerging literature on causal inference using deep neural networks under the potential outcomes framework. It provides an intuitive introduction to building and optimizing custom deep learning models and shows how to adapt them to estimate/predict heterogeneous treatment effects. It also discusses ongoing work to extend causal inference to settings where confounding is nonlinear, time-varying, or encoded in text, networks, and images. To maximize accessibility, we also introduce prerequisite concepts from causal inference and deep learning. The primer differs from other treatments of deep learning and causal inference in its sharp focus on observational causal estimation, its extended exposition of key algorithms, and its detailed tutorials for implementing, training, and selecting among deep estimators in TensorFlow 2 and PyTorch.

  PublisherDOI

• The Social Structure of Scientific Evaluation: AI, Benchmarking, and the Deep Learning Monoculture

  2024-04-12 · 1 citations

  preprintOpen access1st authorCorresponding

  Evaluation systems are central organizing institutions in science that coordinate consensus and driveepistemic trajectories. Scientific fields have traditionally relied on ”organic” evaluation systems (e.g.,peer review, citation) where consensus emerges gradually across multiple epistemic values. This pa-per highlights artificial intelligence research (AIR) as a potent counterpoint to this model. Drawingon interviews with key actors, computational analyses, and archival materials spanning AIR’s history(1956–2021), we examine how AI evolved from a discipline with weak organic evaluation into a fielddriven by benchmarking, a “formal” evaluation system that defines progress quantitatively as state-of-the-art accuracy on commercial tasks. We demonstrate that benchmarking came to dominate through anintricate symbiosis with deep learning: benchmarking rewards accuracy, which large-scale deep learninguniquely excelled at, while deep learning’s opacity made organic evaluation increasingly difficult. Thissymbiosis restructured the field organizationally, epistemically, and materially into a “monoculture” dedi-cated to scaling. While enabling breakneck progress, monoculture discouraged exploration of alternativeswith different epistemic strengths. As AI spreads to other knowledge fields (from science to law to art)benchmarking will accompany it. Our findings thus highlight the risk that formalization of evaluationcan lead to monoculture in other creative domains.

  PublisherOA PDFDOI

• Deep Learning of Potential Outcomes

  arXiv (Cornell University) · 2021-10-09 · 7 citations

  preprintOpen access1st authorCorresponding

  This review systematizes the emerging literature for causal inference using deep neural networks under the potential outcomes framework. It provides an intuitive introduction on how deep learning can be used to estimate/predict heterogeneous treatment effects and extend causal inference to settings where confounding is non-linear, time varying, or encoded in text, networks, and images. To maximize accessibility, we also introduce prerequisite concepts from causal inference and deep learning. The survey differs from other treatments of deep learning and causal inference in its sharp focus on observational causal estimation, its extended exposition of key algorithms, and its detailed tutorials for implementing, training, and selecting among deep estimators in Tensorflow 2 available at github.com/kochbj/Deep-Learning-for-Causal-Inference.

  Publisher

• Uncovering Sociological Effect Heterogeneity Using Tree-Based Machine Learning

  Sociological Methodology · 2021 · 47 citations

  • Machine Learning
  • Computer Science
  • Artificial Intelligence

  Individuals do not respond uniformly to treatments, such as events or interventions. Sociologists routinely partition samples into subgroups to explore how the effects of treatments vary by selected covariates, such as race and gender, on the basis of theoretical priors. Data-driven discoveries are also routine, yet the analyses by which sociologists typically go about them are often problematic and seldom move us beyond our biases to explore new meaningful subgroups. Emerging machine learning methods based on decision trees allow researchers to explore sources of variation that they may not have previously considered or envisaged. In this article, the authors use tree-based machine learning, that is, causal trees, to recursively partition the sample to uncover sources of effect heterogeneity. Assessing a central topic in social inequality, college effects on wages, the authors compare what is learned from covariate and propensity score-based partitioning approaches with recursive partitioning based on causal trees. Decision trees, although superseded by forests for estimation, can be used to uncover subpopulations responsive to treatments. Using observational data, the authors expand on the existing causal tree literature by applying leaf-specific effect estimation strategies to adjust for observed confounding, including inverse propensity weighting, nearest neighbor matching, and doubly robust causal forests. We also assess localized balance metrics and sensitivity analyses to address the possibility of differential imbalance and unobserved confounding. The authors encourage researchers to follow similar data exploration practices in their work on variation in sociological effects and offer a straightforward framework by which to do so.

  PublisherOA PDFDOI

Frequent coauthors

• Christine E. Schnitzler

  13 shared

• Jacob G. Foster

  12 shared

• Joseph F. Ryan

  Whitney Museum of American Art

  10 shared

• Andreas D. Baxevanis

  National Human Genome Research Institute

  10 shared

• Mark Q. Martindale

  Whitney Museum of American Art

  8 shared

• Stephen A. Smith

  University of Michigan–Ann Arbor

  6 shared

• Tyra G. Wolfsberg

  National Human Genome Research Institute

  5 shared

• Yizhou Sun

  5 shared