QGIS bans AI assisted contributions

URL SCAN: QGIS bans AI assisted contributions
FIRST LINE: Repository navigation

The Dissection

This proposal is trying to turn authorship provenance into an enforcement boundary. It is not merely a code-quality rule; it is a declaration that human origin remains a condition of belonging in the project.

The supplied evidence exposes the old policy’s failure: no extractive tag was created, roughly 50 pull requests were tagged AI Assisted, about 40 were closed without merging, and others were closed without tagging. That demonstrates weak detection and enforcement. It does not demonstrate that AI assistance caused the failures. The unresolved questions—bots, passers-by, regular contributors, or core developers—show that the threat model is still vague. The policy is being hardened before the alleged harm has been precisely defined.

The Core Fallacy

The central error is believing that a ban can preserve a stable human-only technical domain under conditions where machine assistance is cheaper, faster, and increasingly competent.

Under P1 and P2, the ban changes visible behavior and social penalties, not the underlying production advantage. Contributors can conceal assistance, interpret “AI assisted” differently, or simply leave. The rule also treats human origin as a proxy for quality and trust. Those qualities are better tested through review, reproducibility, maintainability, licensing checks, and project history. Provenance is often unverifiable.

This is a category error: the project is attempting to govern the source of the work when it lacks a concrete mechanism for proving the source or linking it to failure.

Hidden Assumptions

• AI assistance can be reliably detected or honestly disclosed.
• Human-written code is inherently more trustworthy or maintainable.
• “AI assisted” has a stable, enforceable definition.
• Contributors will obey the rule rather than conceal use, withdraw, or fork.
• The project can absorb the resulting loss in speed or contributor supply.
• Existing review cannot distinguish dangerous output from acceptable output.
• The vague reports of community “hurt” are sufficient justification for a strict prohibition.
• A human-authorship boundary can remain viable while surrounding software production becomes machine-augmented.

Social Function

Primary classification: transition management and ideological anesthetic, with a legitimate but unproven governance function.

The ban gives the community a visible ritual for defending human status while automation erodes the old meaning of contribution. It may suppress low-context spam or preserve trust if AI-generated submissions are demonstrably damaging, but the supplied text provides no defined failure pattern or evidence for that claim.

The prior policy’s tagging failure is the warning flare. Without detection, enforcement, and a specified harm model, “NOT ALLOWED” is mostly a social boundary backed by reputation and exclusion. It is a purity signal dressed as a technical control.

The Verdict

QGIS is building a human-authorship sanctuary around a codebase operating inside an increasingly machine-augmented production system. This may preserve community identity temporarily. It cannot halt AI, guarantee quality, or create a durable human-only technical domain.

Under the Discontinuity Thesis, this is a lag defense—possibly hospice care for an old contribution norm—not a reversal of the underlying force. If the ban excludes useful contributors or makes the project slower than machine-augmented alternatives, symbolic purity becomes technical decline.