GenAI vs. Human Creators: Procurement Mechanism Design in Two-/Three-Layer Markets

Rui Ai [email protected] · 2025

Read GenAI vs. Human Creators: Procurement Mechanism Design in Two-/Three-Layer Markets by Rui Ai [email protected] (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-06559.txt.

Introduction

Rui Ai, David Simchi-Levi, and Haifeng Xu design markets where platforms buy content from human creators and generative models side by side. Content is not an ordinary good here, because material from one domain can train and improve generation in another. Video models, for instance, lean heavily on image data to gain quality.

The authors build optimal nonlinear procurement mechanisms that maximize platform revenue or social welfare, and isolate which traits of generative systems keep the high-dimensional design solvable. Some domains face fiercer competition and others overproduction. Adding data intermediaries as a third market layer can leave everyone worse off, a finding the authors use to argue for regulation, backed by simulations.

Catalog record

CREATORRui Ai [email protected]
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE91 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2511-06559

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

GenAI vs. Human Creators: Procurement Mechanism Design in Two-/Three-Layer Markets Rui Ai [email protected] Abstract With the rapid advancement of generative AI (GenAI), mechanism design adapted to its unique characteristics poses new theoretical and practical challenges. Unlike traditional goods, content from one domain can enhance the training and performance of GenAI models in other domains. For example, OpenAI’s video generation model Sora ( 57 ) relies heavily on image data to improve video generation quality. In this work, we study nonlinear procurement mechanism design under data transferability , where online platforms employ both human creators and GenAI to satisfy cross-domain content demand. We propose optimal mechanisms that maximize either platform revenue or social welfare and identify the specific properties of GenAI that make such high-dimensional design problems tractable. Our analysis further reveals which domains face stronger competitive pressure and which tend to experience overproduction. Moreover, the growing role of data intermediaries, including labeling companies such as Scale AI and creator organizations such as The Wall Street Journal, introduces a third layer into the traditional platform–creator structure. We show that this three-layer market can result in a lose-lose outcome, reducing both platform revenue and social welfare, as large pre-signed contracts distort creators’ incentives and lead to inefficiencies in the data market. These findings suggest a need for government regulation of the GenAI data ecosystem, and our theoretical insights are further supported by numerical simulations. 1 Introduction Large models have entered the era of multimodality ( 87 ) . Modern commercial systems such as GPT-4 ( 2 ) , Gemini ( 79 ) , and Claude ( 8 ) are trained on diverse types of data, including text, images, audio, and video. These modalities are not isolated; transferability among them is well established both theoretically and empirically through transfer learning ( 40 ) . Just as film directors may draw inspiration from Shakespeare’s works, image datasets can enhance the performance of video generation models. This interdependence links the valuation of data across domains into a unified economic problem and naturally raises the question: How do we price content across different domains? The answer depends on the degree of transferability, a feature absent from traditional goods. Bricks, for instance, can only build houses but not chips, while data can improve the performance of models across tasks. Data with high transferability should therefore command a higher price, whereas specialized data that benefits only a single task should be valued lower. In this paper, we incorporate parameters capturing data transferability into a formal production model to analyze how this property shapes market outcomes. The transferability of data also raises important ethical concerns. Recent debates have centered on whether GenAI will cause certain professions to vanish, alongside growing discussions on the implications of artificial general intelligence ( 27 ) . We want to ask: From an economic perspective, will any domains disappear? In Section 2.2 , we show that although GenAI may reduce employment in some domains, it cannot completely replace them. This result follows from the principles of diminishing marginal returns and increasing marginal costs. Moreover, in domains with high data transferability, we observe a tendency toward overproduction, which distorts the allocation of creative effort. Such imbalances highlight the need for regulatory oversight and further research on how GenAI reshapes data generation markets. The growing demand for data to train large models has driven many AI companies to purchase high-quality datasets. For example, OpenAI reportedly spent about $250 million acquiring data from The Wall Street Journal ( 84 ) . Industry forecasts further predict that the global generative AI market for content creation will expand from USD 15.2 billion in 2024 to USD 175.3 billion by 2033 ( 58 ) . This surge in data demand can be traced back to the creation of ImageNet ( 19 ) , which catalyzed the modern data economy. The emergence of specialized data annotation companies such as Scale AI, Lionbridge, Aurora AI, and Amazon Mechanical Turk has since transformed the landscape, pushing the market from a two-layer structure toward a three-layer structure that includes professional data brokers. This evolution raises a key question: How does the three-layer market impact revenue and social welfare? In Section 3 , we examine two types of platforms—those that maximize revenue and those that maximize social welfare—and show that in the three-layer market, both platform revenue and social welfare decline relative to the two-layer setting, resulting in a clear lose-lose outcome. The presence of data brokers distorts creators’ incentives and reduces overall efficiency, highlighting the need for regulatory oversight to address inefficiencies in the data market. 1.1 Our Contributions We summarize our contributions in three main aspects and elaborate on each below. Mechanism design with competition between humans and generative AI. Recent studies have begun examining market equilibria that arise from competition between AI content generators and human creators ( 86 ) . Our work is the first to approach this problem from the buyer’s perspective. We study online sharing platforms that rely on both human creators and GenAI for content production, situating the analysis within the broader framework of procurement mechanism design. Going beyond the classical single-dimensional setting ( 63 ) , we derive optimal mechanisms for five of six multi-dimensional environments and establish tight upper and lower bounds for revenue and social welfare in the remaining case. Our results show that although GenAI reduces the overall demand for human-created content, human creators remain indispensable from an economic standpoint. Their outputs not only satisfy subscriber preferences but also provide essential data for GenAI training. Furthermore, in domains with high data transferability, we find that overproduction is more likely to occur, reflecting the complex interactions between human and algorithmic production incentives. Distinguishing between two/three-layer markets regarding data brokers. Finding an efficient market structure has been a long-standing goal in digital economics ( 53 ; 3 ) . For online platforms, a central question is whether to buy content directly from human creators as crowdsourcing, or through data companies. Our analysis shows that the latter, a three-layer market where the platform purchases data from a broker who sources it from humans, leads to a lose-lose outcome: both platform revenue and social welfare decline. The inefficiency stems from asymmetric information. Large, transparent contracts reveal the platform’s and broker’s valuations to creators, making price discrimination impossible and weakening incentives for efficient production. As a result, both welfare and revenue fall, underscoring the need for policy intervention. Numerical experiments validate the conclusions about market dynamics. Finally, we validate our theoretical results using simulated data. In the two-layer market, the experiments reveal content overproduction in specific domains. We also compare platform revenue and social welfare across the two- and three-layer settings, illustrating the “lose-lose” outcome in the latter. These numerical experiments offer valuable inspiration and insights for future real-world mechanism design and implementation using real data. 1.2 Related Works Our work is closely related to the literature on procurement mechanism design ( 63 ; 60 ; 51 ) , data pricing ( 46 ; 29 ; 75 ; 5 ) , human vs. GenAI competition ( 21 ; 86 ; 23 ; 20 ) and multi-layer markets ( 22 ; 85 ) . We generalize physical goods procurement to

Read the complete text