The business of open-sourcing artificial intelligence: commercial interests and collaboration dynamics in machine learning developer communities

Thesis

Open source is widely celebrated as a means to democratise the development and governance of artificial intelligence (AI) technologies. In the last decade alone, open source AI (OSAI) has evolved from a scientific endeavour to a cornerstone of the AI industry, with over a million open source software (OSS) and open model (OM) repositories on GitHub and Hugging Face Hub respectively. The growth of the OSAI ecosystem is partly credited to commercial activity, from the open-sourcing of AI software and models to collaborations thereon. The pervasiveness of commercial activity raises the question of why and how companies participate in OSAI development, and the implications thereof on development practices, governance norms, and the trajectory of the OSAI ecosystem. This thesis addresses this question through four research papers (RP), which make both theoretical and empirical contributions to the nascent research agenda on the political economy of OSAI. RP1 examines the patterns and practices of "open source co-opetition"–that is, open source collaboration among companies, including market competitors–in three major OSAI projects: PyTorch, TensorFlow, and Transformers. RP2 investigates and taxonomises commercial incentives for "democratising AI" through the lens of 43 AI software donations to the Linux Foundation. RP3 presents a large-scale quantitative analysis of development activity on Hugging Face Hub, shedding light on the uneven patterns of OM development and adoption. RP4 examines the roles of private and public funding in sustaining non-commercial, community-led OSAI projects through a case study on scikit-learn. The thesis concludes with a discussion of its key contributions and future research directions.

Authors: Osborne, C

• DOI: 10.5287/ora-9ekpjgzbz
• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford
• Language: English
• Keywords: open source; online collaboration; political economy; co-opetition; artificial intelligence
• Subjects: Digital sociology; Data science; Computer science; Software engineering research