TLDR: This research establishes a baseline for how human-governed open-source software projects evolve their governance structures. Analyzing 710 projects, it finds that communities define more roles and actions over time, distributing them more evenly, while the prescriptive force of rules remains stable. This “maturation by accretion” provides a crucial reference point for evaluating how future AI-mediated workflows might impact authority and participation in software development.
The world of software development is rapidly evolving, with Artificial Intelligence (AI) systems increasingly taking on roles in project management. From drafting pull requests to triaging issues and enforcing release gates, AI is becoming an integral part of how software projects are run. This shift raises crucial questions about how authority and decision-making are redistributed when algorithms work alongside human maintainers.
A recent research paper titled “A Human Behavioral Baseline for Collective Governance in Software Projects” by Mobina Noori, Mahasweta Chakraborti, Amy X. Zhang, and Seth Frey, delves into these questions by establishing a foundational understanding of how human-governed software projects evolve their governance structures before the widespread adoption of AI tools. This baseline is critical for evaluating whether future AI-mediated workflows concentrate or redistribute authority.
The researchers focused on open-source software (OSS) communities, which serve as an excellent transparent testbed for studying governance. Unlike traditional organizational settings, OSS projects often externalize their governance rules in version-controlled files like GOVERNANCE.md. This makes changes to rules explicit, textual, and historically archived, allowing for systematic observation of how authority changes over time.
To conduct their study, the team analyzed a large corpus of 710 open-source projects spanning from 2013 to 2022. They developed a sophisticated text-based analytical framework, parsing governance documents into fundamental components: actors (roles), rules (deontics), actions, and objects. For instance, in a rule like “The technical committee must ratify the development roadmap,” the “technical committee” is the role, “ratify” is the action, “roadmap” is the object, and “must” indicates the deontic force.
Using advanced Natural Language Processing (NLP) techniques, including Sentence-BERT and BERTOPIC, they clustered these institutional statements and measured changes over time using metrics like Shannon entropy (for evenness), richness (for diversity), and Jensen-Shannon divergence (for drift). This allowed them to quantify how roles, responsibilities, and decision rights evolved within these communities.
Key Findings on Governance Evolution
The study revealed several significant patterns in human-governed open-source projects:
- Projects tend to define a wider array of roles and govern more actions over time. This means communities expand the “who acts” and “what is governed” aspects of their projects.
- Attention across these roles and actions becomes more evenly distributed. This suggests a rebalancing of participation and responsibilities within the community.
- The prescriptive force of rules, expressed through modal verbs like “may,” “should,” or “must” (deontics), remains broadly stable. This indicates that while the scope of participation expands, the fundamental nature of how rules are enforced doesn’t drastically change.
These findings suggest that governance in open-source projects matures by expanding and balancing categories of participation without major shifts in prescriptive power. It’s a process of accretion, where catalogs of actors and activities broaden and rebalance.
Also Read:
- Rethinking AI Software Agent Evaluation: The Benchmark Mutation Approach
- SEER: Integrating Sustainability into Software Requirements with AI
Implications for AI in Software Management
This research provides a crucial benchmark for understanding the impact of AI systems on software project governance. By establishing how human-authored governance evolves, researchers can now evaluate whether AI-mediated platforms concentrate power, narrow participation opportunities, or enable more inclusive governance.
The authors suggest that AI assistance tools should prioritize helping communities define and rebalance roles and actions, highlighting uneven coverage, and proposing governance-ready statements. Furthermore, given the stability of prescriptive force, AI systems should require explicit acknowledgment before intensifying restrictive language and make such shifts transparent during reviews.
While the study primarily analyzed governance text rather than actual behavior and focused on English-language open-source projects, it lays a fundamental step for participatory AI research. Understanding these organic governance trajectories can inform the design of AI systems that allow collective human input in decision processes and ensure that AI infrastructure serves the common good. For more details, you can refer to the full research paper here.
