围绕OpenAI声称其AI代理在88小时内解决纳维-斯托克斯方程相关问题的争议，数学家Tristan Buckmaster质疑未发表的研究草稿可能通过Codex工具泄露给模型 1。这一事件揭示了数据使用界限的模糊性，凸显了人工智能在加速科学研究的同时，正引发对科学完整性的深层担忧 1。
多项数据显示，AI技术已广泛渗透至科研各个环节，但也带来了新的风险 1。2024年，根特大学教授Dmitry Kobak发现近90%的生物医学论文带有AI写作指纹 1。斯坦福大学团队分析指出，计算机科学会议中高达17%的审稿报告文本由语言模型大幅修改 1。此外，康奈尔大学团队发现去年约有147,000篇虚构的“幻影”研究被作为引用列入学术论文中 1。这些现象包括AI生成的“垃圾”论文、虚假引用以及同行评审被篡改等，威胁到科学界赖以生存的信任基石 1。
尽管存在上述风险，业界正在探索通过建立“研究级AI”标准和人类监督机制来平衡技术红利与伦理挑战 1。Google DeepMind的AlphaFold预测了2亿个蛋白质折叠结构，该项目负责人及首席科学家因此获得诺贝尔化学奖，展示了AI带来的突破 1。与此同时，出版机构正采取应对措施：Elsevier开发了约100种信号标记系统用于筛查论文，Nature期刊禁止AI生成图像并严禁审稿人将手稿上传至AI系统 1。
The debate over the role of artificial intelligence in research has intensified following a dispute regarding OpenAI’s claim to solve problems related to the Navier-Stokes equations. Tristan Buckmaster questioned whether unpublished drafts of his work might have been leaked to OpenAI models via the Codex tool, after the company stated its AI agent had addressed these issues within 88 hours 1. This incident serves as a focal point for broader anxieties that while AI accelerates discovery, it simultaneously threatens the foundational trust of the scientific community through unclear data boundaries and compromised review processes 1.
Beyond individual disputes, systemic issues with AI-generated content are widespread. In 2024, University of Ghent professor Dmitry Kobak identified AI writing signatures in nearly 90% of biomedical papers 1. A Stanford University team discovered that up to 17% of reviewer texts at computer science conferences were significantly altered by language models, while researchers at Cornell University found approximately 147,000 fictitious "phantom" studies cited in academic papers last year 1. Despite these risks, AI continues to drive breakthroughs, such as Google DeepMind’s AlphaFold, which predicted 200 million protein folding structures and earned its lead scientists the Nobel Prize in Chemistry 1. In response to integrity threats, publishers like Elsevier have developed around 100 signal markers for paper screening, and Nature prohibits AI-generated images while forbidding reviewers from uploading manuscripts to AI systems 1.
评论
还没有评论，欢迎留下第一条。