Ongoing Research Projects

Gen(X)AI: Boosting Explainable Artificial Intelligence for All

Authors: Francis Joseph Costello, Jooyoung Park, Keongtae Kim, Jason Bennett Thatcher

In Preparation Management Science

As AI becomes embedded in organizational decision-making, the demand for explainability has intensified due to regulatory mandates and the need to promote organizational trust. Yet current Explainable AI (XAI) approaches often prioritize technical transparency over cognitive interpretability, leaving end-data users unable to translate algorithmic outputs into actionable understanding. To address this gap, we introduce GenXAI: a generative AI–mediated explanation layer conceptualized as a boosting intervention designed to enhance user competencies rather than simplifying information content. Drawing on Cognitive Fit and Construal Level theories, we theorize that representational alignment—specifically linguistic concreteness—is critical for effective explanation design. Across two randomized experiments using a loan default prediction task, we find that GenXAI significantly improves task understanding and confidence compared to traditional static dashboards. Crucially, concrete, analytically structured explanations significantly outperform abstract narratives. Furthermore, interaction telemetry reveals a behavioral signature of active cognitive engagement—efficient information processing coupled with extended deliberation—consistent with competency development rather than passive reliance. These findings advance XAI scholarship by reframing explainability as a capability-building infrastructure and establishing linguistic style as a primary design parameter via generative AI within human-AI collaboration. Practically, we demonstrate that organizations should transition from static transparency reporting to interactive interfaces that allow for real-time reasoning. This approach democratizes XAI systems and oversight by empowering end-data users to understand and govern algorithmic decisions more efficaciously, transforming explainability into a strategic mechanism for organizational capability development.

Keywords: Explainable AI, Generative AI, Cognitive Fit, Construal Level Theory, Human-AI Collaboration

Assessing Open Source Software Developer's Opportunity Costs in a Time of Generative AI

Posted: March 6, 2025; Last revised: 12 Jan 2026

Authors: Xinyu Li, Francis Joseph Costello, Keongtae Kim, Xiaoquan (Michael) Zhang

Under Review 1st Round R&R at Management Science

Recent advancements in generative artificial intelligence (AI) have paved the way for tools that can automate knowledge work, with open source software (OSS) development seeing an influx of AI-based tools for code automation and feedback. These AI tools have shown promise in increasing productivity. However, less is known about how developers' technical activities have changed and whether AI-augmented OSS development yields personal career benefits. We leverage the March 2023 release of GitHub Copilot X as a natural experiment, analyzing a 25-week panel dataset of 1,373 GitHub developers through difference-in-differences with coarsened exact matching. Furthermore, we augment our GitHub data with individual LinkedIn career trajectories over 6–12 months post-treatment. We find that developers engage in “cognitive arbitrage,” experiencing lowered cognitive and time opportunity costs in development activities, enabling increased sustained and exploratory development of more varied coding activities. Additionally, heterogeneous analysis provided further insights by suggesting that casual developers utilized AI more to reduce opportunity costs. Interestingly, we found evidence that this enhanced development activity translated into increased short to medium-term career benefits for all developers. Overall, our research suggests that generative AI fundamentally alters the open innovation equation by removing human cognitive barriers, allowing developers to assume more creative roles while simultaneously accelerating their career advancement opportunities.

Keywords: Generative AI, Open-source Software Communities, Code Autocompletion, Opportunity Costs

From Bytes to Bites: Examining the Online and Offline Spillover Impacts of Online Food Delivery Platform Subscriptions

Posted: March 19, 2025

Authors: JaeHo Myeong, Yongjin Park, Francis Joseph Costello, Keongtae Kim, Jae Hyeon Ahn

Under Review 2nd Round R&R at Management Science

With the rapid growth of online food delivery (OFD) platforms and intensifying competition, platform subscriptions have emerged as a key strategy to retain consumers and influence their behavior. This study explores how OFD subscriptions affect consumer ordering patterns and offline dining behavior in a developed Asian market. Using a Difference-in-Differences approach with a large-scale individual-level transaction dataset, we find online spillovers between the focal and competing platform after the release of the subscription together with lock-in effects on the focal platform depending on the subscription upfront costs incurred. However, alongside this shift, we did not find cannibalization, rather the overall OFD market share increased, leading to offline spillovers that enhanced local dining and consumption of proximally close offline non-food products and services. These findings show how platform subscriptions in location-based services (LBS) like OFD reshape consumer mobility patterns and stimulate not only the focal sector but also non-competing sectors within the local economy. The results provide actionable implications for platform operators to refine subscription strategies, for policymakers and urban planners to manage and embrace LBS platform creation and proliferation thanks to local compensatory offline spillovers leading to greater vitality, and for local restaurant owners to capitalize on increased patronage opportunities via platform exposure.

Keywords: Platform subscriptions, Online food delivery, Spillovers, Location-based Services, Consumer Mobility

An Exploration of Behavioral and Neurophysiological Responses to Dark Patterns: A NeuroIS Investigation with fNIRS

Authors: Francis Joseph Costello, Min Gyeong Kim, Cheong Kim

In Preparation Decision Support Systems

This study investigates consumer behavioral and neurophysiological responses to dark patterns through dual-process theory, examining how manipulative digital interfaces designed to exploit automatic System 1 processing can trigger deliberative System 2 resistance. To specify the mechanisms underlying this transition, we draw on False Tagging Theory (FTT), which proposes that prefrontal doubt-generation processes mark automatically accepted representations as suspect, enabling cognitive override. Using a dual-study design, we conducted an online behavioral experiment (N=120) testing five dark pattern categories and a neuroimaging study (N=40) employing functional near-infrared spectroscopy (fNIRS) to measure prefrontal cortex activation during dark pattern exposure. Results demonstrate a layered dual-process response: dark patterns initially engage System 1 acceptance, evidenced by higher information satisfaction reflecting an illusion of benefit, but subsequently activate System 2 doubt mechanisms that significantly reduce purchase intentions, recommendation likelihood, and emotional valence. These effects varied systematically by category—sneaking, hidden subscription, and urgency produced reliable resistance, while confirmshaming and hard-to-cancel manipulations did not, suggesting that resistance depends on the salience and immediacy of perceived financial threat. Neuroimaging findings reveal coordinated activation in left frontopolar cortex (BA10) and left ventrolateral prefrontal cortex (BA47), supported by complementary oxygenated and deoxygenated hemoglobin changes, providing direct neurophysiological evidence consistent with FTT's proposed false tagging mechanisms. These findings challenge assumptions about consumer passivity by demonstrating measurable neural resistance to manipulation and reframe dark patterns as a decision support failure mode. The research contributes to IS theory by establishing a neurophysiologically-grounded, mechanism-oriented account of consumer resistance to dark patterns, extending dual-process frameworks beyond outcome-focused descriptions, and providing evidence-based foundations for ethical interface design principles.

Keywords: Dark Patterns, Digital Nudging, False Tagging Theory, NeuroIS, fNIRS