The AI Pascal’s Wager: Why Open-Source Projects Should Reject AI-Generated Contributions for Now

Summary

Ploum examines whether open-source projects should accept contributions generated by large language models. The article identifies three possible positions: broadly accepting AI-assisted contributions, adopting an ambiguous compromise such as a Debian-style policy, or strongly rejecting generated code. It argues that the middle position is unstable because AI-generated contributions will gradually enter projects despite safeguards such as mandatory human review, effectively making the project pro-AI without admitting it. Accepting AI may increase contributions and reduce work on tedious tasks, but it can alienate community members and make development dependent on expensive proprietary tools. Rejecting AI may lose contributors who insist on using LLMs, yet the author argues that it can strengthen support among anti-AI users and avoid accepting code that contributors cannot independently understand or test. The article also raises longer-term risks involving degraded human understanding of the codebase, dependence on assistants that could become unaffordable, and possible license conflicts from copied generated code. Using a Pascal’s-wager-style argument, Ploum concludes that an AI-agnostic project should reject AI-generated contributions for now because that decision can be reversed later, while allowing them in may create damage that is difficult to undo.