Knowledge that Transforms

Artificial intelligence (AI)-automated decision systems encounter persistent, interdependent, and dynamic fairness tensions that traditional one-off interventions cannot resolve. Because these tensions persist due to interdependence and dynamic interaction, organizations require both a theory of the problem to explain their persistence and a theory of the solution to prescribe how they can be managed. Our design theory, FAIR (Fairness Adaptation through AI-augmented Responsiveness), provides a theory of the problem by reframing AI fairness as a sociotechnical paradox constituted within AI artifacts that automate decision tasks, through interdependent organizational, technical, and governance choices and their interaction with regulatory mandates and societal norms. Synthesizing four fairness perspectives (Ethics, Organizational Justice, Economic Fairness, and Rawlsian Justice), we identify three metatheoretical dimensions (principles, goals, foci) and show that the interdependence within and among these dimensions is the root, endogenous source that constitutes paradoxical fairness tensions. Building on this diagnosis, FAIR provides a theory of the solution by specifying an organizational capability grounded in three design foundations. First, the paradox lens motivates iterative adaptive cycles (Surfacing and Resolving) to continually surface and resolve AI fairness tensions. Second, design science in information systems and computer science distinguishes AI artifacts (the “what”) from the actors (the “who”) responsible for adapting them, establishing the basis for complementary human–AI agent collaboration in the adaptive cycles: AI agents execute monitoring to surface and refinement to resolve tensions, whereas human agents specify objectives, adjudicate trade-offs, and exercise contextual judgment and oversight. Third, the managing-with-AI literature informs how this human–AI agent collaboration should be governed. These foundations yield two reinforcing mechanisms: (i) artifact-level adaptation, achieved through structured human–AI agent collaboration, within and across the layers of the AI decision pipeline—Representation (data), Learning (model), and Calibration (decision); and (ii) portfolio-level, risk-tiered federated governance that structures how human–AI agent collaboration scales across tasks and artifacts, balancing process standardization with configuration choices and human control with AI autonomy based on task risk. Enabled by organizational “fairness complements”—namely, human skills to work with AI agents and structured stakeholder feedback—this sociotechnical design provides organizations with a sustained capability to harmonize global coherence and local flexibility in the responsive adaptation of AI fairness.

This commentary explores multiple perspectives on the potential use of digital technologies to improve organizational resilience in the context of climate change. Such an approach is needed to address this complex problem space, especially since it encompasses a wide variety of phenomena, including floods and landslides, disruptions to global supply chains, heat waves, biodiversity loss, greenhouse gas emissions, and food insecurity. We assembled a diverse set of five scholarly teams specializing in multiple problem topics, research approaches, and theoretical perspectives on this project. Each team identified and problematized a specific facet of digital resilience for the climate crisis. The perspectives cover a range of rich narratives, including digital resilience in the context of floods and landslides in Brazil and Indonesia, conceptual development efforts incorporating the natural environment with people and technology, reconceptualization of the problem space in terms of time and type, and two applications of digital resilience in the domains of global supply chains and carbon emissions tracking. This research commentary thus presents a multi-perspective examination and interrogation of digital resilience for addressing the climate crisis, out of which four transcending themes emerge: the need to integrate nature into sociotechnical thinking, the need to examine actions at both micro and macro levels, the need to include both reactive and proactive strategies, and the need to view climate crisis as a process rather than a series of events. This commentary aims to motivate other scholars who take diverse theoretical perspectives to join us in developing fundamental knowledge and practical solutions needed to achieve digital resilience for the climate crisis.

The classic digital divide theory asserts that unequal access to and unequal experience with information technologies may lead to unequal user outcomes. This paper introduces a new perspective to extend this theory: outcome divides can persist despite equal access and equal experience if users differ in their analytical ability to analyze and interpret available data for decision-making. We term this new data-to-decision skill as analytical ability and integrate it into the classic digital divide framework. We develop a new approach to operationalize analytical ability by contrasting humans’ actual performance against that of a standard machine learning model that makes similar analytical decisions based on the same information available to humans, essentially emulating a quasi-random counterfactual setting. To minimize the confounding impact of other divides, we validate the role of analytical ability in information-transparent environments like the blockchain-based trading markets, where all historical trading data is equally available to all users on the blockchain. We leverage data from EnjinX, a blockchain-enabled non-fungible token (NFT) marketplace that records all historical NFT transactions. We measure user outcomes by their flip trading performance, a standard metric captured via the percentage of exploited flipping opportunities. Our empirical analysis reveals that disparities in analytical ability may become the new bottleneck for outcome equity: flip trading performance could decrease by 66.86% when traders are incapable of analyzing the available blockchain information effectively. Our study contributes to the literature by extending the digital divide theory with the notion of the analytical ability divide. Moreover, we are among the first to rigorously quantify analytical ability and empirically test its impact based on the extended digital divide framework. Our study also offers important practical implications for platforms and policymakers to bridge this new divide in order to foster outcome equity.

Digital platform ecosystems thrive on their ability to acquire and leverage user data across multiple data-driven services. This enables dominant platforms to harness insights obtained from their primary markets, where user data is collected, thus gaining a competitive advantage in secondary markets, where they exploit this data. While data cross-use brings about efficiencies, policymakers worldwide have expressed concerns about the economic power and the potential distortion of competition and innovation incentives associated with it. To address these concerns, two distinct and targeted policy interventions have been suggested: data siloing, which restricts the cross-use of data within platform ecosystems, and mandated data sharing with competitors. Using an analytical model that examines data cross-use in digital platform ecosystems, we analyzed the impact of data siloing and data sharing obligations, and their interaction on competition, innovation, consumer welfare, and overall social welfare. Our findings indicate that an optimal policy involves data sharing without data siloing, whereas the EU’s Digital Markets Act currently mandates both types of data cross-use regulation.

As online reviews become increasingly indispensable for consumers, they have attracted significant attention from both practitioners and researchers. It is a common belief that the persuasive effect of online reviews involves a deliberative and conscious process. Drawing on dual-process theories and the persuasion literature, we challenge this conventional wisdom, distinguish Type 2 processing (which requires deliberation) and Type 1 processing (which occurs automatically), and disentangle their relative impacts. With a focus on review elaborateness and review exposure, we propose that the automatic process of review exposure may play a greater role than elaborateness in changing consumers’ attitudes and purchase intentions. In addition, in line with the negativity bias, we posit that the persuasive impact of review exposure (vs. elaborateness) is moderated by the valence of highly exposed reviews. The results of the two experiments provide consistent support for these predictions. Our findings complement and extend the emerging literature starting to explore the role of automatic Type 1 processing in consumers’ use of online reviews, reveal the primary driver of persuasion and its boundary condition in online word-of-mouth, and provide important implications for review platforms, product manufacturers, and retailers.

Only a fraction of people with mental health issues seek medical care, in part because of fear of judgment, so deploying text-based conversational agents (i.e., chatbots) for mental health screening is often viewed as a way to lower barriers to mental health care. We conducted four experiments and a qualitative study and, contrary to common assumptions, consistently found that participants perceived a text-based chatbot as more judgmental than a human mental health care professional, even though the interactions were identical. This greater judgmentalness reduced the willingness to use the service, disclose information, and follow the agent’s recommendations. Participants described judgmentalness as a rush to judgment without fully grasping the issues. The chatbot was perceived as more judgmental because it was less capable of deeply understanding the issues (e.g., emotionally and socially) and conveying a sense of being heard and validated. It has long been assumed that chatbots can address the real or imagined fear of being judged by others for stigmatized conditions like mental health. Our study shows that perceptions of judgmentalness are actually the opposite of what has been assumed and that these perceptions significantly influence patients’ acceptance of chatbots for mental health screening.

Depression is a common mental disorder involving a depressed mood or loss of pleasure for long periods, which induces grave financial and societal ramifications. Social media-based depression detection is an effective method for early intervention to mitigate those consequences. Such a high-stake decision inherently necessitates interpretability. Although a few studies explain this decision based on the importance of linguistic or demographic features, these explanations do not directly relate to depression diagnosis criteria that are based on symptoms. To fill this gap, we develop a Focused Temporal Prototype Network (FTPNet) to detect depression and provide interpretations based on depressive symptoms as well as their temporal distributions. Extensive evaluations using large-scale datasets show that FTPNet outperforms comprehensive benchmark methods with an F1-score of 0.864. Our result also reveals fine-grained and emerging manifestations of depressive symptoms, such as sharing admiration for a different life, that are unnoted in traditional depression surveys like the Patient Health Questionnaire-9 (PHQ-9). We further conduct a user study to demonstrate improved interpretability over the benchmark. This study contributes to the Information Systems (IS) literature by introducing an interpretable depression detection approach that models the temporal distribution of depressive symptoms. In practice, multiple stakeholders, such as social media platforms and volunteers, can apply our approach to identify depressed users and deliver targeted assistance.

Augmented reality (AR) is rapidly emerging as a transformative display technology, blending computer-generated content with the real-world environment in real time. Using divided attention theory, this study investigates how different information delivery channels (i.e., AR vs. mobile phone) and the nature of information (i.e., dependence on specific physical context and complexity) affect work performance. A field experiment in the aircraft maintenance context demonstrates that the effect on work performance of providing information via AR vs. a mobile phone is mediated by work attentiveness. The findings reveal that the effectiveness of AR is particularly pronounced when information is highly dependent on the specific physical context but diminishes when information complexity is high. This research deepens our understanding of how presenting information directly in front of users’ eyes (i.e., via AR) affects their attention management and work performance. The findings have significant implications for firms in terms of how to leverage AR to enhance work performance in industrial settings.

Algorithms are increasingly seen as capable of autonomously initiating and managing interactions with humans—for example, through delegating the rights and responsibilities for successful outcomes of shared tasks without human intervention. While research into such interactions primarily focuses on dyadic configurations, complex settings where multiple agents work together have become a nexus of more nuanced interactions that go beyond the dyad. This paper explores such interactions through the lens of delegation by investigating how many algorithms delegate to many humans in a multi-agent setting. Analyzing patent data and interviews with drivers and passengers, we unpack delegation in the context of the ride-hailing application Uber. We theorize distributed delegation as a construct capturing collective hybrid appraisal, collective hybrid distribution, and collective hybrid coordination, in which a collective of algorithms delegates by drawing on inputs from multiple human agents. Our findings highlight that distributed delegation is collective, hybrid, and relational by nature, and demonstrate the extent to which human inputs are necessary for collectives of algorithms to exercise the capacity to delegate. Distributed delegation as a continuum of algorithmic and human involvement poses a challenge for recent theories suggesting the unprecedented autonomy of algorithms from humans.

Although the prevalence of artificial intelligence (AI) innovations is on the rise, firms frequently report failures and setbacks in their development and implementation of AI innovation efforts. One common issue behind many failing AI initiatives is that they are organized just like other information technology (IT) innovation efforts. To elucidate why and how the production of AI and IT innovations may need to be managed differently, this study juxtaposes these two types of innovations based on two key dimensions of the Schumpeterian framework: the form (product vs. process) and magnitude (radical vs. incremental) of innovations. By analyzing a matched sample of AI and IT patents, we found robust evidence that AI innovations are less radical and more process oriented than comparable IT innovations. Drawing upon our empirical discovery, we developed a conceptual framework to suggest a new way to think about organizing AI innovation. Our research contributes to the literature and practice on AI innovation by illuminating the comparative differences between AI innovations and other IT innovations and advancing a set of empirically derived propositions on how firms may be able to better manage their AI innovation activities.