Research Papers
Is Turing Transformation Happening? A Study with a New Measure of Occupational AI Exposure
Authors:Liu, Liu, Huang
Published:SSRN
Date:2025-08
A central concern regarding artificial intelligence (AI) is its potential to replace human workers. However, recent research suggests that AI may improve the workers' job prospects through a Turing Transformation process: AI simplifies work, lowers barriers to job entry, and thus broadens job opportunities for more workers. In this paper, we present a conceptual framework that considers the two dimensions of AI's impact on job content and job opportunity, placing the Turing Transformation as a special case. We develop a novel occupational AI exposure measure using a sentence transformer model to compare the semantic similarity between the occupation descriptions and AI patents. We find that, on average, occupations with higher AI exposure experience a decrease in the importance of a wide range of work activities, coupled with an increase in job opportunities. This provides the first empirical evidence for the existence of the Turing Transformation process. In addition, we observe important heterogeneity in AI's work activity and job opportunity effects across occupations with different education requirements. These findings contribute to the literature by offering a comprehensive occupational view of the “where,” “what,” and “how” of AI's impact on work.
Download
Artificial Intelligence Upskills Human-Centric and Fundamental Computer Science Knowledge
Authors:Masi, Liu, Liu
Published:SSRN
Date:2025-08
What skills should workers acquire to remain valuable in a labor market being transformed by artificial intelligence (AI)? This is a crucial question for workers to adapt to the workforce disruptions driven by AI adoption. We address this question by identifying the job skills that are in increasing demand as AI applications expand across occupations. Specifically, we measure occupational AI exposure by considering the textual similarity between occupational descriptions and AI patents. Then, based on 347 million job postings in the US, we consider 442 distinct skills and examine how the demand for skills changes with occupational AI exposure from 2010 to 2023. We find that occupations with high AI exposure show rising demand for two types of skills: (1) human-centric capabilities such as critical thinking, problem solving, and communication; and (2) foundational computer science knowledge. These AI upskilling effects appear not only in high-skilled occupations, but also for those that do not require a college degree. Our findings uncover the human-machine complementarity at the fine-grained skill level. They also offer detailed guidance for workforce reskilling in the age of AI.
Download
Artificial Intelligence and Inventory Performance: An Exploratory Investigation
Authors:Xia, Liu, Huang, Liu
Published:SSRN
Date:2025-07
Problem definition: Artificial Intelligence (AI) has transformed many facets of modern business. At its core, machine learning promises more accurate demand forecasting under uncertainty. However, it remains unclear whether AI actually leads to better inventory performance in practice. In this paper, we explore the impact of AI from an operations management perspective: Does AI really improve inventory performance? Methodology/Results: We adopt a data-driven approach by integrating firm-level AI job posting data with Compustat. We develop a novel human-capital-based metric that quantifies AI investment in supply chain specifically and examine the impact of AI investment on inventory performance in U.S. retail and wholesale sectors. Our regression results with fixed effects and Difference-in Differences show a significant negative relationship between recruitment of supply chain human capital equipped with AI skills and relative inventory levels, with effects intensifying over time. Interestingly, we observe heterogeneous effects: pure online players benefit less from AI investment than traditional and multichannel retailers and wholesalers. In addition, firms with higher operational complexity experience larger decreases in inventory levels. Conversely, high labor intensity hinders the effectiveness of AI investment and adversely impacts inventory performance. Managerial Implications: We provide empirical evidence that AI investment improves inventory performance, especially when combined with supply chain expertise. Managers should consider: prioritize "algorithm bilinguals"-professionals blending AI skills with domain knowledge-over overall AI human capital, account for a payoff lag, tailor AI investment to their specific operational contexts to optimize AI-driven reductions in inventory levels.
Download
Does Machine Learning Shift Job Requirements? Impacts on Entry-Level Opportunities
Authors:Liu, Liu, Huang
Published:SSRN
Date:2025-05
We study the impact of machine learning (ML) utilization on the job requirements for workers. We posit that, by automating common tasks and processes, ML technologies shift human workers toward handling more complex and novel scenarios that demand higher levels of professional expertise. Our empirical analyses employ over 51 million job postings of S&P 500 companies from 2011 to 2023 with a Shift-Share IV estimation strategy, leveraging texts from occupational descriptions and AI patents. We find that firms utilizing ML technologies also raise their job requirements for prior work experience and skills, especially those related to decision-making. These effects are evident not only for knowledge workers but also for roles that typically do not require a college education. Furthermore, these effects are especially pronounced in occupations characterized by high skill turnover and non-routine work. These findings demonstrate how ML utilization within the firm may have changed the skill composition of workers and hence reshaping the nature of work at scale.
Download
Uncovering IT Career Path Patterns with Job Embedding-based Sequence Clustering
Authors:Zhong, Liu, Wu
Published:ACM Transactions on Management Information Systems
Date:2025-03
Extracting typical career paths from large-scale and unstructured talent profiles has recently attracted increasing research attention. However, various challenges arise in effectively analyzing self-reported career records. Inspired by recent advancements in neural networks and embedding models, we develop a novel career path clustering approach and apply it to uncover information technology (IT) career path patterns. Specifically, we construct employment profiles of over 60,000 IT professionals, and form their career path sequences by chaining the job records in each profile. Then we simultaneously learn cluster-wise job embeddings and construct career path clusters. The resultant cluster-wise likelihoods of career paths can quantify their soft bonding with different clusters, and the job embeddings can reveal connections among job titles within each cluster. With both real and simulated data, we conduct extensive experiments with our framework to establish the modeling performance and great improvement over the traditional optimal matching analysis methods. The empirical results from analyzing real data on career paths show that our approach can discover distinct IT career path patterns and reveal valuable insights.
Download
