About

Kartik Hosanagar is a Faculty Co-Director of Wharton Human-AI Research at the Wharton School. His research focuses on exploring the design, impact, and governance of intelligent systems across organizations and society. He is involved in advancing human-centered AI for business innovation, examining how AI influences creativity, the future of work, and business solutions. Hosanagar contributes to the understanding of responsible AI implementation, emphasizing accountability, ethics, and trust in AI systems. He collaborates on industry reports and research initiatives that address the adoption of AI agents, skills transition in the economy, and enterprise AI deployment. Additionally, he leads discussions and webinars on the latest AI applications, their impact on industries, and strategies for effective AI oversight.

Research topics

• Computer Science
• Artificial Intelligence
• Machine Learning
• Information Retrieval
• Computer Security
• Knowledge management
• Political Science
• Mathematics
• Distributed computing
• Engineering
• Public relations
• Management
• Data science
• Management science
• Psychology

Selected publications

• Generative AI Adoption by Creator Platforms 

  SSRN Electronic Journal · 2025-01-01 · 1 citations

  articleOpen accessSenior author
  PublisherDOI

• BiasConnect: Investigating Bias Interactions in Text-to-Image Models

  ArXiv.org · 2025-03-12

  preprintOpen access

  The biases exhibited by Text-to-Image (TTI) models are often treated as if they are independent, but in reality, they may be deeply interrelated. Addressing bias along one dimension, such as ethnicity or age, can inadvertently influence another dimension, like gender, either mitigating or exacerbating existing disparities. Understanding these interdependencies is crucial for designing fairer generative models, yet measuring such effects quantitatively remains a challenge. In this paper, we aim to address these questions by introducing BiasConnect, a novel tool designed to analyze and quantify bias interactions in TTI models. Our approach leverages a counterfactual-based framework to generate pairwise causal graphs that reveals the underlying structure of bias interactions for the given text prompt. Additionally, our method provides empirical estimates that indicate how other bias dimensions shift toward or away from an ideal distribution when a given bias is modified. Our estimates have a strong correlation (+0.69) with the interdependency observations post bias mitigation. We demonstrate the utility of BiasConnect for selecting optimal bias mitigation axes, comparing different TTI models on the dependencies they learn, and understanding the amplification of intersectional societal biases in TTI models.

  PublisherOA PDFDOI

• An Investigation of <i>p</i>-Hacking in E-Commerce A/B Testing

  Information Systems Research · 2025-01-30

  articleSenior author

  Concerns about the integrity of statistical analyses have risen in recent years, with particular attention given to “p-hacking”—a process whereby analysts conduct several statistical tests until they achieve a statistically significant result. Although extensively studied in academic settings, less is known about its prevalence in industrial contexts. In this study, we investigate whether p-hacking occurs in e-commerce A/B testing. We analyzed nearly 2,300 A/B tests conducted by hundreds of firms using a large A/B testing platform. Such platforms typically offer continuous monitoring of test results, a feature that facilitates real-time decision making but also enables potential p-hacking through selective stopping or continuation of experiments. Contrary to concerns raised by earlier research on academic practices, we found no significant evidence of p-hacking in our sample. These findings suggest that the industrial application of experimentation may be less susceptible to p-hacking than academic research. We discuss several possible factors explaining the divergent results, highlighting the potential role of organizational learning and the importance of economic incentives. Our study contributes to the broader discussion on research integrity and underscores the importance of considering contextual factors in assessing statistical malpractice.

  PublisherDOI

• Harnessing AI for Business Insight

  Oxford University Press eBooks · 2025-08-21

  book-chapter

  Abstract Large language models (LLMs) have demonstrated significant potential in handling unstructured, natural language data. However, their adaptation to complex business settings remains a challenging endeavor. We explore the challenges and research opportunities associated with deploying LLMs for document and knowledge summarization in various business applications. We evaluate current paradigms for evaluation and highlight their inability to fully capture the multi-dimensional considerations in business settings, including relevance, provenance, and factuality of the output. The chapter emphasizes the need for a paradigm shift in evaluation approaches to better align with the nuanced needs of business applications. Key considerations include expanding the dimensionality of automated evaluation metrics, incorporating human-computer interaction factors, and addressing domain-specific needs. A case study on summarizing user-generated content from a product announcement video on social media is presented to illustrate these challenges and associated research opportunities.

  PublisherDOI

• Mitigate One, Skew Another? Tackling Intersectional Biases in Text-to-Image Models

  2025-01-01

  articleOpen access

  Pushkar Shukla, Aditya Chinchure, Emily Diana, Alexander Tolbert, Kartik Hosanagar, Vineeth N. Balasubramanian, Leonid Sigal, Matthew A. Turk. Findings of the Association for Computational Linguistics: EMNLP 2025. 2025.

  PublisherOA PDFDOI

• Welfare Impact of Bilateral Rating Display in Ride-Sharing Platforms

  Manufacturing & Service Operations Management · 2025-12-08 · 1 citations

  article

  Problem definition: The bilateral rating display system (BRDS), popularized by ride-sharing platforms, shows riders’ ratings to drivers when the driver receives an order dispatched by the platform. This contrasts with traditional business settings where only the service providers’ ratings are displayed to customers (unilateral rating display system (URDS)). Although BRDS allows drivers to make more informed decisions, it also poses challenges, such as the potential for drivers to reject riders based on their ratings. Such rejections could harm both riders and the platform. Methodology/results: Our paper builds a game-theoretic model to study the impact of BRDS on all stakeholders in the context of ride-sharing services. We highlight how the subtle interplay between rating display and system congestion can improve welfare and boost the platform’s revenue. BRDS can help regulate matchings, resulting in a Pareto improvement over URDS when riders’ valuations are below a threshold. Managerial implications: BRDS provides a nonpecuniary operational lever to mitigate incentive conflicts between platforms and drivers over revenue sharing. The platform achieves a higher revenue share when riders’ valuations are low, whereas drivers receive a higher revenue share when riders’ valuations are high, compared with URDS. BRDS can help regulate congestion, leading to improved system performance. Funding: All authors acknowledge The Wharton School Dean’s Postdoctoral Research Fund and Mack Institute Research Fund. C. Jin gratefully acknowledges the Singapore Ministry of Education Academic Research Fund Tier 1 [Awards T1251RES2101 and T1251RES2501]. Supplemental Material: The online appendices are available at https://doi.org/10.1287/msom.2022.0472 .

  PublisherDOI

• TIBET: Identifying and Evaluating Biases in Text-to-Image Generative Models

  Lecture notes in computer science · 2024-11-01 · 11 citations

  book-chapter
  PublisherDOI

• Impact of Model Interpretability and Outcome Feedback on Trust in AI

  2024-05-11 · 27 citations

  articleOpen accessSenior author

  This paper bridges the gap in Human-Computer Interaction (HCI) research by comparatively assessing the effects of interpretability and outcome feedback on user trust and collaborative performance with AI. Through novel pre-registered experiments (N=1,511 total participants) using an interactive prediction task, we analyzed how interpretability and outcome feedback influence users’ task performance and trust in AI. The results counter the widespread belief that interpretability drives trust, showing that interpretability led to no robust improvements in trust and that outcome feedback had a significantly greater and more reliable effect. However, both factors had modest effects on participants’ task performance. These findings suggest that (1) interpretability may be less effective at increasing trust than factors like outcome feedback, and (2) augmenting human performance via AI systems may not be a simple matter of increasing trust in AI, as increased trust is not always associated with equally sizable performance improvements. Our exploratory analyses further delve into the mechanisms underlying this trust-performance paradox. These findings present an opportunity for research to focus not only on methods for generating interpretations but also on techniques that ensure interpretations impact trust and performance in practice.

  PublisherOA PDFDOI

• Designing Human and Generative AI Collaboration

  arXiv (Cornell University) · 2024-12-14 · 3 citations

  preprintOpen access1st authorCorresponding

  We examined the effectiveness of various human-AI collaboration designs on creative work. Through a human subjects experiment set in the context of creative writing, we found that while AI assistance improved productivity across all models, collaboration design significantly influenced output quality, user satisfaction, and content characteristics. Models incorporating human creative input delivered higher content interestingness and overall quality as well as greater task performer satisfaction compared to conditions where humans were limited to confirming AI's output. Increased AI involvement encouraged creators to explore beyond personal experience but also led to lower aggregate diversity in stories and genres among participants. However, this effect was mitigated through human participation in early creative tasks. These findings underscore the importance of preserving the human creative role to ensure quality, satisfaction, and creative diversity in human-AI collaboration.

  PublisherOA PDFDOI

• Harnessing AI for Business Insight: Key Considerations for Deploying LLMs in Summarization Pipelines

  SSRN Electronic Journal · 2024-01-01

  preprintOpen access
  PublisherDOI

Frequent coauthors

• Dokyun Lee

  24 shared

• Daniel Fleder

  11 shared

• Ramayya Krishnan

  11 shared

• R. Guérin

  10 shared

• Vibhanshu Abhishek

  University of California System

  9 shared

• Ashish Agarwal

  The University of Texas at Austin

  8 shared

• Yong Tan

  University of Washington

  8 shared

• John Chuang

  University of California, Berkeley

  7 shared

Labs

• Wharton Human-AI ResearchPI