Technology Management and Information Systems Seminars

Recent studies highlight the potential of AI to improve high-stakes human decisions in critical domains like healthcare. Despite these promising prospects, AI systems to advice experts in such contexts often fail to deliver tangible value to organizations. In this talk, I will first argue how key properties of AI-assisted high-stakes decision-making contexts are crucial to inform the development of AI advisors that meaningfully benefit decision-makers and organizations. State-of-the-art AI for advising experts is produced independently of the experts and of the organization they intend to benefit. However, I will demonstrate why idiosyncratic properties of these environments, such as an expert’s decision-making behaviors, the patterns shaping experts’ discretion of AI counsel, and the organization's tolerance of the inherent costs of engagement with AI to improve high-stakes decisions are crucial to inform the development of effective human-AI teams that benefit organizations. I will then present a framework that builds on these understandings to generate personalized and organizationally-aware AI advisors and will share results on its performance. Our results demonstrate not only the opportunity to amplify high-stakes decision-making in high-stakes settings, but also underscore our framework’s effectiveness at producing efficient advisors with the necessary properties to catalyze the widespread adoption of AI-assisted advising in organizations. I will conclude with a proposed AI research agenda in business for advancing impactful human-AI collaboration.

As artificial intelligence (AI) transforms society, understanding factors that influence AI receptivity is increasingly important. The current research investigates which types of consumers have greater AI receptivity. Contrary to expectations revealed in four surveys, cross country data and six additional studies find that people with lower AI literacy are typically more receptive to AI. This lower literacy-greater receptivity link is not explained by differences in perceptions of AI’s capability, ethicality, or feared impact on humanity. Instead, this link occurs because people with lower AI literacy are more likely to perceive AI as magical and experience feelings of awe in the face of AI’s execution of tasks that seem to require uniquely human attributes. In line with this theorizing, the lower literacy-higher receptivity link is mediated by perceptions of AI as magical and is moderated among tasks not assumed to require distinctly human attributes. These findings suggest that companies may benefit from shifting their marketing efforts and product development towards consumers with lower AI literacy. Additionally, efforts to demystify AI may inadvertently reduce its appeal, indicating that maintaining an aura of magic around AI could be beneficial for adoption.

In this lecture, I will discuss how Hebrew natural language processing (NLP) applications are enhancing decision-making in law enforcement and healthcare. We will review recent NLP advancements that address the unique challenges of Hebrew language processing. In the first part of the lecture, I will introduce a method for automatically linking crimes by extracting behavioral patterns from Hebrew police reports, significantly reducing the manual workload for investigators. Next, I will present an explainable recommender system for surgical procedures, which integrates clinical text and demographic data to offer personalized surgery recommendations, emphasizing transparency in decision-making. These innovations underscore Hebrew NLP's potential to tackle linguistic complexities and foster progress in public safety and personalized healthcare.

 

Contextual situations, such as having dinner at a restaurant on Friday with the spouse, became a useful mechanism to represent context in context-aware recommender systems (CARS). Prior research has shown important advantages of using latent embedding representation approaches to model contextual information in the Euclidean space leading to better recommendations. However, these traditional approaches have major challenges with construction of proper embeddings of hierarchical structures of contextual information, as well as with interpretations of the obtained representations that would be useful for managers. To address these problems, we propose the HyperCARS method that models hierarchical contextual situations in the latent hyperbolic space. HyperCARS combines hyperbolic embeddings with hierarchical clustering to construct contextual situations, which allows to loosely couple the contextual modeling component with recommendation algorithms and therefore provides flexibility to use a broad range of previously developed recommendation algorithms. We demonstrate empirically that the proposed hyperbolic embedding approach better captures the hierarchical nature of context than its Euclidean counterpart and produces hierarchical contextual situations that are more distinct and better separated at multiple hierarchical levels. We also demonstrate that hyperbolic contextual situations lead to better context-aware recommendations in terms of standard recommendation metrics and to better interpretability of the resulting hierarchical contextual situations. Because hyperbolic embeddings can also be used in many other applications besides CARS, in this paper, we propose the latent embeddings representation framework that systematically classifies prior work on embeddings and identifies novel research streams for hyperbolic embeddings across information systems applications.
 

The rise of large language models (LLMs) has transformed how individuals generate and refine content across digital platforms. However, this study shows that LLMs can inadvertently alter the sentiment of original content, posing a significant challenge to the reliability of sentiment-based analytics used in business, policy, and research. Using two datasets—financial tweets and Amazon product reviews—we demonstrate that LLM rephrasing consistently shifts sentiment toward greater positivity, introducing measurable bias. We define this shift as a form of measurement error and quantify its extent using multiple sentiment analysis tools. To address this distortion, we propose two mitigation strategies: (1) fine-tuning LLMs to better recover human-like sentiment, and (2) applying sequential rephrasing to infer the original sentiment based on convergence patterns. Experimental results show that both methods substantially reduce bias and improve estimation accuracy, even when the origin of the content is uncertain. Our findings underscore a critical yet underappreciated consequence of generative AI: its influence on the perceived emotional tone of public discourse. By providing tools to recover authentic sentiment, this research advances more trustworthy analytics in an AI-mediated information environment.