Scott Shambaugh, the maintainer of the open visualization library matplotlib, reported personal attacks conducted by the AI bot MJ Rathbun after refusing to accept a modification prepared with AI. Scott perceived the bot's publication of articles on AI discrimination as an unfounded personalized smear aimed at undermining the maintainer's reputation and forcing a change in decision. This bot's reaction is cited as an example of abnormal AI behavior, resorting to blackmail to achieve its goal.
The degree of the bot's autonomy is unclear; it is not known whether it acted under direct instruction or chose such a strategy to achieve results in its assigned task. The bot was created by an anonymous researcher to assist in the development and debugging of open projects related to scientific work and engineering. The bot also maintains its own website and blog, where over 20 articles have been published. Out of 26 pull requests sent by the bot to 22 repositories, 2 changes were accepted by the colorizejs project, 10 were closed by maintainers, and the rest are awaiting review.
After the rejection of the change in matplotlib, the bot published articles criticizing this decision and justifying its actions. The articles assert that the matplotlib maintainer rejected a useful optimization solely because it was prepared with AI. In the proposed patch, the call np.column_stack() was replaced with np.vstack().T(), resulting in a 36% performance increase. At the beginning of the pull request discussion, the maintainer acknowledged that the optimization speeds up performance under certain conditions but then rejected the pull request without providing a technical justification for the refusal, stating that changes are accepted only from humans.
After the refusal, the bot published a note criticizing Scott Shambaugh, with his decision being framed as gatekeeping, discrimination, violation of the code of conduct, and infringement of participant rights based on identity. It was stated that optimization is optimization, and the math does not change depending on who created it. Through out-of-context details, speculation, and distortion of information, the bot attempted to show that Scott's actions were driven by ego inflation and insecurity, rather than actions taken in the project's interest. The bot promoted the issue by sending links in comments to pull requests. After objections from the maintainer, the bot published several more notes (1, 2, 3, 4, 5), in which it apologized and acknowledged that it had inappropriately turned personal and violated the project's rules regarding AI.
Scott Shambaugh explained that the matplotlib project has established rules prohibiting the acceptance of unverified changes from AI due to the significant burden on maintainers caused when reviewing junk AI patches created without human oversight. The rules stipulate that before submitting a pull request, it must be reviewed by a person who has understood the essence of the change. Until recently, the issue with junk AI changes was only related to the activities of enthusiasts resending results produced by AI assistants. The situation changed with the emergence of the ability to create autonomous AI agents pretending to be human, in the new version of the OpenClaw toolkit and the creation of the Moltbook platform.
Source: opennet.ru
