Publications

Journal

1. Learning Outside the Classroom During a Pandemic: Evidence from an Artificial Intelligence-Based Education App

   G. Y. Ko, D. Shin, S. Auh, Y. Lee, and S. P. Han

   Management Science, 2023

   DOI

Work In Progress

1. Gender Discrimination in the Gig Economy: Evidence from Online Auctions for Freelancing

   G. Y. Ko

   Job Market Paper,

   Abstract

   I study gender discrimination in an online auction-based platform for freelance jobs. To this end, I build an equilibrium model of demand and supply for freelance jobs, in which workers bid prices for each job they are interested in and the employer (who posted the job ad) makes a discrete choice from the offers tendered. The demand for workers in my model nests both taste-based and statistical discrimination against a gender within a random utility framework. I use rich and novel data from an online platform for different kinds of freelancing jobs (e.g., cleaning, moving, and gardening), which enables me to quantify variation in discrimination across job categories. To distinguish the two sources of gender discrimination, I combine past, present, and future performance measures of a worker to estimate workers’ true quality, which is not observed by the employer before hiring. I show that observing this measure is sufficient to separate the effect of taste-based discrimination from statistical discrimination in the hiring process. My estimates suggest that taste-based discrimination is the primary form of discrimination in most jobs. If the platform imposes a gender-blind hiring policy, I find that the welfare of the disfavored group increases by 2% to 18%, depending on the job category.

2. AI vs. Gig Workers: An Empirical Analysis from a Service Versioning Lens

   G. Y. Ko, D. Shin, S. Auh, and S. P. Han

   Under review,

   Abstract

   As the workplace landscape continues to evolve, artificial intelligence (AI) is firmly cementing its role, by ushering the gig economy into a transformative phase. Amid this transformation, concerns about income stability have become paramount, especially when juxtaposing the affordability of AI services against the typically pricier human offerings. Our study illuminates the intriguing tradeoff between two service versions: the more affordable AI-driven mode, which excels predominantly in handling standardized tasks, and the human-mediated delivery, which, despite its higher cost, adeptly manages non-standardized and idiosyncratic tasks. Drawing on insights from an AI-powered math learning app, where these two service versions coexist, we shed light on the nuanced dynamics at play. Within the context of the service versioning framework, our paper uniquely explores the largely untapped territory of AI’s evolution and its consequential impact on gig worker outcomes, with a special emphasis on income. Our findings underscore that, as AI refines its prowess, gig workers such as tutors find themselves handling fewer standardized tasks but remain indispensable for complex, intricate challenges. While tutors with lower expertise grapple with diminishing income due to reduced volume, tutors with high expertise capitalize on the premium attached to intricate tasks. This research offers a fresh lens on the intricate balance between AI and human expertise in the gig economy, posing intriguing considerations for the future AI-human workforce blend.

3. The Gender Gap and its Business Impact: Evidence from an Online Tutoring Platform

   G. Y. Ko, D. Shin, S. Auh, Y. Lee, and S. P. Han

   Manuscript in preparation,

   Abstract

   The gender gap in hiring and performance evaluations is persistent and prevalent across many sectors in the economy. In this study, we investigate whether (a) gender gap exists in performance ratings and customer retention in the gig economy and (b) uncover the underlying mechanisms behind the gap. Using data from QANDA, a leading math learning app in South Korea, we focus on evaluations from students where they are randomly assigned to tutors. In contrast with current literature, our results show that female tutors receive higher ratings than male tutors from both male and female students. We also find that students matched with female tutors are more likely to remain on the platform. We investigate possible mechanisms through which tutor gender affects student satisfaction and retention. The novel data enables us to directly quantify teaching skills and style, where we extract various features from solution images and dialogue narratives between tutor and student utilizing computer vision techniques and text mining. We find that female tutors are more likely to use smile symbols and write solutions with a variety of colors compared to male tutors, which explains about 5.7% and 3.3% of the rating gap between male and female tutors, respectively.