From OSS to Open Source AI: an Exploratory Study of Collaborative Development Paradigm Divergence

A new study from researchers Hengzhi Ye and Minghui Zhou reveals that open-source AI development operates under a fundamentally different paradigm than traditional open-source software. Analyzing 1.4 million GitHub repositories (OSS) and 1.4 million Hugging Face repositories (OSM), the researchers found that AI projects exhibit significantly lower collaboration intensity, with 40% fewer contributors on average. The study also shows lower collaboration openness regarding direct code contributions, though knowledge exchange remains relatively open through forums and discussions.

The research identifies a key divergence: while traditional OSS focuses on collaborative improvement through pull requests and issue tracking, open-source AI development leans toward adaptive utilization. Users are more likely to fine-tune existing models like Llama 3 or Stable Diffusion for specific tasks rather than contribute directly to the core model. Through semi-structured interviews, the researchers attribute these differences to socio-technical factors including the black-box nature of AI models, high computational resource requirements, and different skill sets needed for model development versus application.

These findings have significant implications for platform designers and community managers. Understanding that AI development follows different collaborative patterns suggests that platforms like Hugging Face might need different tools than GitHub to foster effective collaboration. The study provides empirical evidence for what many in the field have observed anecdotally: building AI models requires different community structures and incentives than building traditional software.