Working Papers

AI Upskilling: Decoupling Skill Demand and Skill Premium

Pedro Masi, Yuanyang Liu, and Chuanren Liu

Status: R&R, Major Revision — Information Systems Research (ISR)

Presented: Workshop on Information Technologies & Systems (WITS), Nashville, TN, Dec 2025; Conference of Information Systems & Technology (CIST), Atlanta, GA, Oct 2025;

Understanding how AI reshapes the market value of skills is essential for guiding effective workforce responses to labor market disruptions. We examine two dimensions of skill market value — skill demand intensity (the frequency with which employers require a skill) and skill wage premium (the salary differential associated with possessing a skill) — across a comprehensive set of over 20,000 skills in 442 skill categories. To this end, we measure occupational AI exposure using dynamic LLM embeddings that capture the textual similarity between AI patent descriptions and O*NET occupational task descriptions. Drawing on 347 million U.S. job postings from 2010 to 2023 and employing a two-stage least squares approach with a breakthrough patent instrumental variable, we find three main results. First, occupations with greater AI exposure exhibit stronger employer demand for both human-centric skills (e.g., critical thinking, communication) and foundational computer science skills. Second, only computer science and system design skills command higher wage premiums in response to AI exposure; human-centric skills, despite rising in demand, do not generate corresponding salary rewards. Third, both dimensions exhibit meaningful heterogeneity across occupations and job zones: management occupations experience broad upskilling across human-centric, technical, and domain skills, while computer occupations face rising demand for critical thinking yet negative salary premiums for those same skills. These findings reconcile conflicting narratives in the literature by showing that AI can simultaneously increase the importance of skills while failing to raise their monetary value, thereby producing both upskilling and deskilling effects depending on the dimension considered. Our empirical framework offers a dynamic tool for tracking AI-driven changes in market value of job skills to inform hiring strategy, education policy, and workforce development.

AI Skill Herding and Capital Investment: When Does Following the Crowd Pay Off?

Pedro Masi, Howard Zhong, Yuanyang Liu, and Chuanren Liu

Status: In preparation — target MIS Quarterly (MISQ)

Presented: Americas’ Conference on Information Systems (AMCIS), Reno, NV, August 2026 (accepted, forthcoming); Computational Methods in Management Research Workshop, University of Notre Dame (AIS SIG TECH), June 2026; DePaul Doctoral/Junior Faculty Research Symposium, Chicago, IL, March 2026;

Firms increasingly respond to competitive pressure in AI adoption by aligning their AI hiring practices with those of their peers, which extends classic herding behavior from individuals to organizations. However, it remains unclear whether capital markets reward these visible signals of AI strategic positioning. Using firm-level data on AI skill demand from job postings linked to startup capital investment outcomes, we examine how alignment with peer AI hiring patterns shapes capital allocation. We find that firms that exhibit stronger AI-skill herding attract greater capital, which indicates that mimicking AI talent can serve as an effective signal to investors. We also show that this association is significantly stronger for firms with a robust STEM workforce, which indicates that a deep base of scientific and engineering talent enhances the credibility of AI hiring signals. In contrast, proximity to external sources of AI innovation does not systematically strengthen this relationship. Finally, we show that the benefits of AI hiring alignment depend on the perceived quality of firms in capital markets, specifically, lower-quality firms experience stronger gains from aligning with peer hiring patterns, while higher-quality firms do not derive additional advantages.

Papers in Progress

Decoding AI Innovation: A Transformer-Based Approach to Patent Landscaping

Pedro Masi, Yuanyang Liu, and Chuanren Liu

Conference Presentations & Invited Talks

  • “Herding Toward AI” — Americas’ Conference on Information Systems (AMCIS), Reno, NV, August 2026 (accepted, forthcoming)
  • “AI and the Market Value of Skills” — DePaul Doctoral/Junior Faculty Research Symposium, Chicago, IL, March 2026
  • “AI Upskills Human-Centric and Fundamental Computer Science Knowledge”
    • Workshop on Information Technologies & Systems (WITS), Nashville, TN, Dec 2025
    • Conference of Information Systems & Technology (CIST), Atlanta, GA, Oct 2025
  • “Tracking the New Frontiers of Work in Tennessee” — AI Tennessee Workshop, Knoxville, TN, Jan 2025
  • “AI Talk: Can Executives Keep Their Promises?” — INFORMS Annual Conference, Seattle, WA, Oct 2024
  • “AI for Business Innovation” — Business Analytics Forum, UTK, Aug 2024