DigitalEcon.org

@digitalecon.org

814 Followers 8 Following

We organize the Digital Economy Workshop and the Virtual Digital Economy Seminar http://digitalecon.org #VIDEseminar

ABSTRACT Researchers and industry practitioners increasingly propose and deploy large language models (LLMs) as human surrogates, replacing human surveys with synthetic LLM counterparts in behavioral experiments, focus group studies, political polling, and more. These approaches share a common premise: collect limited human data, then fine-tune or prompt LLMs to impute unobserved preferences or behaviors across tasks and populations, attempting to leverage the vast pretrained knowledge encoded within LLMs. However, LLM intelligence is a fundamentally different form of intelligence from that of humans, relying on probabilistic patterns from vast, noisy, and epistemically contaminated data and lacking the embodied experience and survival-driven objective functions that shape human cognition and behavior. Human behavior is heterogeneous, idiosyncratic, and often unstable, making it fundamentally different from the types of systems that traditional digital twins are designed to model. This talk critically examines the pitfalls of using LLMs as human surrogates, highlighting key failure modes, foundational theoretical limitations, and core empirical dilemmas, and situates these concerns within the emerging body of evidence drawn from a handful of studies.
Does Generative AI Crowd Out Human Creators? Evidence from Pixiv Sueyoul Kim, Ginger Zhe Jin, and Eungik Lee Abstract: Using a comprehensive dataset of posts from a major platform for anime- and manga-style artwork, we study the impact of the launch of a prominent text-to-image generative AI. Focusing on the majority of incumbent creators who do not adopt AI as a primary tool, we show that the AI launch led to a significant decline in post uploads by illustrators, whereas comic artists were less affected, reflecting the need for tight stylistic alignment across sequential images in comics. We present empirical evidence for two underlying mechanisms. First, illustration posts experience a loss of viewer attention, measured by bookmarks, following the AI launch, which can significantly harm creators’ business models. Second, direct competition from AI-generated content plays an important role: illustrators working on intellectual properties (IPs, such as Pokémon) that are more heavily invaded by AI reduce their uploads disproportionately more. We further examine creators’ responses and show that illustrators with greater exposure to AI avoid using tags favored by AI-generated content after the AI launch and broaden the range of IPs they work on, consistent with a risk-hedging response to AI invasion.
Paper title: Long or Short? Personalizing Ad Length and Frequency Authors: Ali Goli (University of Rochester), David Reiley (UC Berkeley), Wenfeng Qiu (Sirius XM Holdings Inc.), Jonas Tungodden (Norwegian School of Economics) Abstract We study the economics of advertising length using a large-scale randomized experiment on 18.3 million Pandora listeners. The experiment promoted a listening feature and randomly assigned users to advertising campaigns that varied in ad format (10 vs. 30 seconds) and frequency. For both users unfamiliar with the feature and those already familiar, we document that both ad formats generate incremental effects, with the relative advantage of long over short ads substantially larger among unfamiliar users. We then build a machine learning model to capture heterogeneity in both responsiveness to the campaign and realized exposure levels, and conduct off-policy evaluation of alternative campaign designs. We find that a personalized policy that optimally assigns users across the available experimental conditions, choosing the ad format and frequency for each user, increases incremental ad effects by about 30 percent. Two-thirds of this efficiency gain comes from leveraging heterogeneity in exposure rather than heterogeneity in responsiveness to the campaign.
Schedule - Spring 2026 Wednesday, February 4 - David Reiley (UC Berkeley) Long or Short? Personalizing Ad Length and Frequency (with Ali Goli, Jonas Tungodden and Wenfeng Qiu) Wednesday, March 4 - Ginger Zhe Jin (U of Maryland) TBD Wednesday, April 8 - Joshua Gans (U of Toronto Rotman) TBD Wednesday, May 6 - Dokyun Lee (Boston U) Take caution in using LLMs as human surrogates: Scylla Ex Machina
Screen shot of front page of paper Paper title: How Have Smartphones Changed the Structure of Online Content Consumption? Authors: Hong Lee, Jeffrey Prince, and Shane Greenstein Date: July 2025 Abstract: Smartphones have become the primary medium for consuming online content globally. Despite this significant change in how content is consumed, little is known about whether and how the introduction of mobile devices as a means of online content consumption, in addition to home computers, has impacted the way content is consumed. In this paper, we empirically analyze the distribution of session lengths and the variety of online consumption for both mobile devices and home computers. The analyses utilize data on home computer consumption in 2008 and 2019, as well as data on mobile consumption from 2019. We find little change in the variety of consumption on home computers between 2008 and 2019. It also shows that mobile platforms are associated with shorter, more frequent sessions and more concentrated consumption patterns compared to home devices. As users spend more time online via mobile devices, they tend to focus on a limited set of high-utility apps or websites, resulting in reduced variety compared to home computer usage. Taken together, the results suggest that the introduction of smartphones had little effect on the allocation patterns of home computer consumption, while encouraging a more concentrated pattern of use on mobile devices.
ABSTRACT We examine how the shift of streaming platforms to an advertiser-supported revenue model affects their content offerings and consumers. This shift forces platforms to adjust their content strategies, though the direction is unclear. Platforms may aim to increase ad impressions by offering content that appeals to a broad audience, thereby driving up viewership. Alternatively, they may seek to increase the value of impressions by targeting niche audiences, who are typically more valuable to advertisers. These changes also impact consumers: while they generally dislike ads, consumers might benefit from adjusted content offerings if aligning with their preferences. We plan to answer this question using data on content offerings of platforms and consumer viewership.
Schedule - Fall 2025 - Wednesday, September 3 - Jin-Hyuk Kim (Colorado Boulder) Platform Screening, Investor Learning, and Default Risk in Marketplace Lending (with Peter Newberry and Nick Vikander) - Wednesday, October 8 - Pinar Yildirim (U Penn Wharton) TBD - Wednesday, November 5 - Julie Holland Mortimer (Wash U) The Effect of Ad-Supported Plans on Content Offerings of Streaming Platforms (with Sylvia Hristakeva, Ashwin Nair, and Yihao Yuan) - Wednesday, December 3 - Jeff Prince (Indiana Business School) TBD
Schedule - Spring 2025 - Wednesday, February 5 - Joel Waldfogel (University of Minnesota) Information and the welfare benefits from differentiated products (with Imke Reimers and Christoph Riedl) - Wednesday, March 5 - Shane Greenstein (Harvard Business School) New Economic Forces Behind the Value Distribution of Innovation (with Timothy Bresnahan and Pai-Ling Yin) - Wednesday, April 2 - Alessandro Acquisti (Carnegie Mellon University) An Experimental Infrastructure to Investigate the Impact of Online Tracking, Targeting, and Advertising on Consumer Behavior and Consumer Welfare - Wednesday, May 7 - Garrett Johnson (Boston University) Privacy-Enhanced versus Traditional Retargeting: Ad Effectiveness in an Industry-Wide Field Experiment (with Shunto Kobayashi and Zhengroung Gu)