Abstract:

[Purposes] This paper aims to point out the publishing ethical issues of articles partly written by machines by investigating the current situations, and then propose identification and preventive countermeasures. This study is expected to arouse the attention and thoughts of the academic publishing industry. [Methods] A web survey was conducted on the tools for writing papers, and retracted papers by Retraction Watch tagged with “Randomly Generated Content” were investigated. Through case studies and interpretation methods, the situations, characteristics, and ethical controversies of articles partly written by machines were summarized. [Findings] There are many ways and tools for machines to participate in article writing. Manuscripts directly generated by machines have a high risk of academic misconduct, and the trend of articles submitted by Chinese authors retracted for using machine-generated content has increased obviously. The number of articles partly written by machines is growing rapidly, accompanied by the problems of data ethics, academic innovation, algorithm opacity and bias, citation, and authorship. [Conclusions] Machine-generated articles are identified with characteristics of academic misconduct, such as reducing repetition rate, fabrication, improper citation, and interference in peer review. For articles partly written by machines, editors need to judge the degree of autonomy of the machine through keywords and other methods. It is necessary to prevent the publishing ethical risk of machine participation in article writing by perfecting the receiving rules, improving detection technology, opening peer review, reviewing algorithms, enhancing machine transparency, standardizing authorship rules, standardizing data citation and code software citation, and perfecting internal copyright agreement and external collaborative governance.

Key words: Artificial intelligence, Machine writing, Academic misconduct, Academic ethics, Preventive countermeasure

摘要：

【目的】 调查机器参与论文写作的现状,指出其中的出版伦理问题,并提出识别和防范对策,以期引起科技期刊界对机器参与论文写作现象的关注和思考。【方法】 对部分论文写作工具展开实验,分析Retraction Watch中因“随机生成内容”而被撤销的稿件情况。通过案例及阐释方法总结机器参与论文写作的发表现状、特征及伦理争议,提出相应对策。【结果】 机器参与论文写作的方式和工具较多。机器直接生成的稿件涉及学术不端风险的概率较高,国内作者因使用机器生成内容被撤稿的趋势明显上升。机器间接参与论文写作的研究数量快速增长,但存在数据伦理问题、学术创新争议、算法黑箱与偏见问题、引用问题、署名问题。【结论】 机器生成的论文有降重、伪造、不当引用、干扰同行评审等学术不端特征可供识别,其他类机器参与写作的论文需要编辑通过关键词等方法判断机器的自主程度。应从完善收稿规则、改进检测技术、开放同行评审、审查算法、提高机器透明度、规范署名、规范数据引用和代码软件引用、完善内部版权协议和外部协同治理等方面防范机器参与论文写作的出版伦理风险。

关键词: 人工智能, 机器写作, 学术不端, 学术伦理, 防范对策