Strategic Behavior and AI Training Data
Authors:
Christian Peukert,
Florian Abeillon,
Jérémie Haese,
Franziska Kaiser,
Alexander Staub
Abstract:
Human-created works represent critical data inputs to artificial intelligence (AI). Strategic behavior can play a major role for AI training datasets, be it in limiting access to existing works or in deciding which types of new works to create or whether to create new works at all. We examine creators' behavioral change when their works become training data for AI. Specifically, we focus on contri…
▽ More
Human-created works represent critical data inputs to artificial intelligence (AI). Strategic behavior can play a major role for AI training datasets, be it in limiting access to existing works or in deciding which types of new works to create or whether to create new works at all. We examine creators' behavioral change when their works become training data for AI. Specifically, we focus on contributors on Unsplash, a popular stock image platform with about 6 million high-quality photos and illustrations. In the summer of 2020, Unsplash launched an AI research program by releasing a dataset of 25,000 images for commercial use. We study contributors' reactions, comparing contributors whose works were included in this dataset to contributors whose works were not included. Our results suggest that treated contributors left the platform at a higher-than-usual rate and substantially slowed down the rate of new uploads. Professional and more successful photographers react stronger than amateurs and less successful photographers. We also show that affected users changed the variety and novelty of contributions to the platform, with long-run implications for the stock of works potentially available for AI training. Taken together, our findings highlight the trade-off between interests of rightsholders and promoting innovation at the technological frontier. We discuss implications for copyright and AI policy.
△ Less
Submitted 29 April, 2024;
originally announced April 2024.
