m5k1umn 发表于 2024-9-28 17:14:09

你用AI写的论文，已然被“盯”上了！


    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;">采写丨科技日报实习记者 吴叶凡 记者 付丽丽</span></strong></span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;">编辑丨刘恕 李坤</span></strong></span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">“太不像话了！学生用人工智能生成的期末论文糊弄我。”<span style="color: black;">近期</span>，上海某高校教师在社交<span style="color: black;">媒介</span>上“吐槽”自己遇到的新<span style="color: black;">困难</span>——<span style="color: black;">有些</span>想偷懒的学生<span style="color: black;">起始</span>用人工智能技术完成论文。</span></span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">以ChatGPT为<span style="color: black;">表率</span>的生成式人工智能技术（AIGC）横空出世，似乎为人们写论文<span style="color: black;">供给</span>了新<span style="color: black;">助手</span>。从<span style="color: black;">供给</span>选题到文稿润色、从统计分析到图表制作……其功能之强大，几乎覆盖了学术论文写作过程的方方面面。</span></span></p>
    <div style="color: black; text-align: left; margin-bottom: 10px;"><img src="https://p3-sign.toutiaoimg.com/tos-cn-i-axegupay5k/df271d5ceed643c080a20742b56b2ce2~noop.image?_iz=58558&amp;from=article.pc_detail&amp;lk3s=953192f4&amp;x-expires=1727620016&amp;x-signature=sSP%2B4mxPQPi3Jf5GkrSXmRgeMR4%3D" style="width: 50%; margin-bottom: 20px;"></div>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">面对ChatGPT等工具的潜在<span style="color: black;">危害</span>，争议随之而来。不少人质疑，人工智能到底能<span style="color: black;">不可</span>用于辅助学术论文写作。有人认为，它只是<span style="color: black;">加强</span><span style="color: black;">研究</span>效率的工具。有人则对此持审慎态度，认为容易<span style="color: black;">诱发</span>大规模的学术诚信问题。</span></span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;">人工智能技术在论文写作中的应用程度<span style="color: black;">怎样</span>？技术应用的边界在哪里？<span style="color: black;">怎样</span>对这一技术进行有效治理？</span></strong>科技日报记者对此进行了深入采访。</span></p>
    <h1 style="color: black; text-align: left; margin-bottom: 10px;">AI生成的文本“非常水”</h1>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">有多少人尝试过用人工智能技术写论文？去年《自然》杂志对<span style="color: black;">全世界</span>博士后的一项调查<span style="color: black;">发掘</span>，</span><strong style="color: blue;"><span style="color: black;"><span style="color: black;">大概</span>三分之一的受访者<span style="color: black;">运用</span>人工智能聊天<span style="color: black;">设备</span>人来优化文本、生成或编辑代码、整理文献。</span></strong></span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">当记者尝试在社交<span style="color: black;">媒介</span>上搜索“AI”“论文”“写作”等关键词，五花八门的AI论文写作<span style="color: black;">指点</span>教程映入眼帘。其中大部分宣<span style="color: black;">叫作</span>能够教会用户在几分钟内<span style="color: black;">经过</span>几个简单的<span style="color: black;">过程</span>，生成一篇几万字的“<span style="color: black;">优秀</span>”论文。这些教程的浏览量最高已达数百万。</span></span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">AI真的能生成一篇完整的“<span style="color: black;">优秀</span>”论文吗？记者<span style="color: black;">根据</span>教程<span style="color: black;">起始</span>了尝试：“请提出与民族志纪录片<span style="color: black;">相关</span>的论文选题。”几乎无需等待，几个看起来很“<span style="color: black;">可靠</span>”的选题就出<span style="color: black;">此刻</span>对话框里。</span></span></p>
    <div style="color: black; text-align: left; margin-bottom: 10px;"><img src="https://p3-sign.toutiaoimg.com/tos-cn-i-6w9my0ksvp/0e1d41cb16dd4c24bd3e3c468a83b1d6~noop.image?_iz=58558&amp;from=article.pc_detail&amp;lk3s=953192f4&amp;x-expires=1727620016&amp;x-signature=O%2BZ%2FGj%2FL8THGlNS3kWvwzXw7%2Fi8%3D" style="width: 50%; margin-bottom: 20px;">
      <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">某大语言模型生成的民族志纪录片论文选题。网页截图</p>
    </div>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">“请就某一选题生成写作大纲。”几秒后，7个像模像样的章节<span style="color: black;">所有</span>生成完毕。“请就提纲中某项内容，<span style="color: black;">仔细</span>描述2000字。”重复几次操作后，一篇几万字的“论文”<span style="color: black;">火速</span>就完<span style="color: black;">成为了</span>。但记者浏览后<span style="color: black;">发掘</span>，</span><strong style="color: blue;"><span style="color: black;">其生成的段落中，存在大部分重复且言之无物的内容。</span></strong></span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">除了说“车轱辘”话，某985高校人工智能专业硕士<span style="color: black;">科研</span>生温睿还<span style="color: black;">发掘</span>了此类论文的行文特点：“<span style="color: black;">通常</span>是先写一句话，<span style="color: black;">而后</span>进行分条论述。当老师看到<span style="color: black;">这般</span><span style="color: black;">招数</span>化的内容就会猜测，这类<span style="color: black;">文案</span>很大程度上是人工智能写的。”</span></span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;"><span style="color: black;">文案</span>开头那位教师的经历印证了温睿的<span style="color: black;">发掘</span>。</span><strong style="color: blue;"><span style="color: black;">“<span style="color: black;">这般</span>的论文看似条理清晰、层次丰富，但<span style="color: black;">实质</span>上<span style="color: black;">每一个</span>层面的内容都很少，<span style="color: black;">况且</span>非常空洞。我马上就<span style="color: black;">可疑</span>是AI生成的。”</span></strong>该老师说。</span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">不少期刊编辑、审稿人<span style="color: black;">亦</span><span style="color: black;">发掘</span>了<span style="color: black;">一样</span>的问题。</span></span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">某人文社科期刊审稿人徐彬向记者透露，用AI写论文的关键在于提示词。<span style="color: black;">倘若</span>提示词<span style="color: black;">选择</span>的不恰当，就极有可能得到一篇<span style="color: black;">招数</span>化的<span style="color: black;">文案</span>。他<span style="color: black;">日前</span><span style="color: black;">已然</span>收到过五六篇“一眼就能看出来”用AI写的稿子。</span></span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">“这些<span style="color: black;">文案</span>的<span style="color: black;">一起</span>特点<span style="color: black;">便是</span>非常水。虽然它生成的语言连贯性不错，<span style="color: black;">然则</span>缺乏深度，创新性<span style="color: black;">亦</span>不强。”对此，徐彬略显无奈，</span><strong style="color: blue;"><span style="color: black;">“综述类<span style="color: black;">文案</span>是<span style="color: black;">运用</span>AI的重灾区，但<span style="color: black;">日前</span>期刊还缺乏<span style="color: black;">关联</span>的<span style="color: black;">评估</span>标准和处理机制。”</span></strong></span></p>
    <h1 style="color: black; text-align: left; margin-bottom: 10px;">伪造数据集更具隐蔽性</h1>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">在清华大学人工智能国际治理<span style="color: black;">科研</span>院副院长梁正看来，论文核心<span style="color: black;">评估</span>标准<span style="color: black;">包含</span>作者发挥的创造性、对论文的贡献程度。</span><strong style="color: blue;"><span style="color: black;">一篇大部分由AI生成且隐瞒<span style="color: black;">运用</span><span style="color: black;">状况</span>的<span style="color: black;">文案</span>，既<span style="color: black;">无</span>作者智力的贡献，<span style="color: black;">亦</span>不符合<span style="color: black;">研究</span>诚信的<span style="color: black;">需求</span>，属于学术造假。</span></strong></span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">AIGC<span style="color: black;">导致</span>的学术造假还<span style="color: black;">出现</span>在数据<span style="color: black;">行业</span>。记者在采访过程中，多位业内专家<span style="color: black;">说到</span>了伪造数据集问题。相比直接的文本生成，这一方式更<span style="color: black;">拥有</span>隐蔽性。</span></span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">GPT-4的ADA功能是一种结合了计算机编程语言Python的模型，<span style="color: black;">能够</span>执行统计分析和创建数据可视化。梁正向记者讲述了一则真实的案例：国外某<span style="color: black;">公司</span><span style="color: black;">科研</span>人员先是<span style="color: black;">需求</span>GPT-4 ADA创建一个关于圆锥角膜<span style="color: black;">病人</span>的数据集，后又<span style="color: black;">需求</span>它编造临床数据，用以支持深板层角膜移植术比穿透性角膜移植术效果更好的结论。但真实的临床数据证明，两种手术效果并无<span style="color: black;">显著</span>差别。</span></span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">“针对某个问题，提出<span style="color: black;">办法</span>来<span style="color: black;">处理</span>，并<span style="color: black;">经过</span>实验来证明<span style="color: black;">办法</span>的可行性——这是专业论文的常用模式。</span><strong style="color: blue;"><span style="color: black;">人工智能<span style="color: black;">不可</span>做实验，哪怕它给的实验数据再理想，<span style="color: black;">亦</span>都是虚假的。”</span></strong>温睿认为，虚假的数据背离了科学<span style="color: black;">科研</span>的真正<span style="color: black;">道理</span>。</span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">除了数据处理，<span style="color: black;">更加多</span>人<span style="color: black;">运用</span>AIGC来解释概念。温睿<span style="color: black;">发掘</span>AIGC生成的概念简洁明了，查重率<span style="color: black;">亦</span>非常低。但当记者询问这些概念<span style="color: black;">是不是</span>正确时，温睿<span style="color: black;">显出</span>有些迟疑：“我<span style="color: black;">亦</span><span style="color: black;">无</span>把握，<span style="color: black;">一般</span>默认它是对的。”</span></span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">为了验证AIGC给出答案的准确性，</span><strong style="color: blue;"><span style="color: black;">记者就<span style="color: black;">有些</span>新兴概念提问，但它给出的答案<span style="color: black;">常常</span>和真正概念毫不沾边。</span></strong>当记者让AI生成5篇某<span style="color: black;">行业</span>的重点参考文献，它又胡编乱造了5个不存在的作者和不存在的文献。</span></p>
    <div style="color: black; text-align: left; margin-bottom: 10px;"><img src="https://p3-sign.toutiaoimg.com/tos-cn-i-6w9my0ksvp/85066ec1bd85419e836e1d85b7b680a3~noop.image?_iz=58558&amp;from=article.pc_detail&amp;lk3s=953192f4&amp;x-expires=1727620016&amp;x-signature=jM%2ByJYFvqSDkllQW89rPEJxEYgI%3D" style="width: 50%; margin-bottom: 20px;">
      <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;">以上作者和论文并不存在。网页截图</p>
    </div>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">在人工智能<span style="color: black;">行业</span>，描述AI“一本正经地胡说八道”的专业名词是“AI幻觉”。哈尔滨工业大学（深圳）特聘校长助理、教授张民解释，AI幻觉<span style="color: black;">指的是</span>AI会生成貌似<span style="color: black;">恰当</span>连贯，但与输入问题意图不一致、与现实或已知数据不符合或<span style="color: black;">没法</span>验证的内容。这<span style="color: black;">大都是</span><span style="color: black;">因为</span>AI对知识的记忆不足、理解能力<span style="color: black;">不足</span>、训练方式固有的<span style="color: black;">坏处</span>端及模型本身技术的局限性所<span style="color: black;">引起</span>。</span></span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;">“<span style="color: black;">倘若</span>不<span style="color: black;">警觉</span>AI幻觉，<span style="color: black;">特别有</span>可能损害科学<span style="color: black;">科研</span>的真实性和客观性。”</span></strong>梁正<span style="color: black;">暗示</span>，AI生成的错误信息一旦被广泛传播，不仅会<span style="color: black;">导致</span>“学术垃圾”泛滥，还将影响学术生态的良性发展。</span></p>
    <h1 style="color: black; text-align: left; margin-bottom: 10px;">一场你追我逃的“猫鼠游戏”</h1>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">一项新技术的<span style="color: black;">显现</span>，<span style="color: black;">针对</span>社会的发展<span style="color: black;">常常</span>是把双刃剑。虽然人工智能技术存在种种隐患，但其在图文创作、数据处理等方面的强大能力已被大<span style="color: black;">都数</span>人认可。</span><strong style="color: blue;"><span style="color: black;">“归根结底，<span style="color: black;">咱们</span>认为AI将<span style="color: black;">增多</span>人类的智慧，而非取代人类。其<span style="color: black;">运用</span>应在人类监督之下，并将道德<span style="color: black;">原因</span><span style="color: black;">思虑</span>在内。”</span></strong>施普林格·自然集团发言人说。</span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">推动AI向善发展，需要借助行之有效的技术手段。值得<span style="color: black;">重视</span>的是，AI生成的论文并<span style="color: black;">不可</span>被查重工具检测出来。<span style="color: black;">因此呢</span>，</span><strong style="color: blue;"><span style="color: black;">国内外都在探索<span style="color: black;">开发</span>专门针对AIGC的检测工具。</span></strong></span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">从原理看，AIGC检测技术是在“用AI打败AI”。同方知网数字出版技术股份有限<span style="color: black;">机构</span>副总经理柯春晓介绍：“人类的创作<span style="color: black;">常常</span>是随机且富有灵感的，而接受过<span style="color: black;">海量</span>文本训练的AI<span style="color: black;">已然</span>形<span style="color: black;">成为了</span>生产文本的‘固有’范式，倾向于<span style="color: black;">运用</span>‘一致’的结构和规则，<span style="color: black;">因此呢</span><span style="color: black;">拥有</span>更高的可预测性。”</span><strong style="color: blue;"><span style="color: black;">AIGC检测的核心<span style="color: black;">便是</span>依托海量的文本和数据样本，识别出人类和AIGC工具在平均句子长度、词汇多样性和文本长度等方面的<span style="color: black;">区别</span>点，从而揪出AI论文“枪手”。</span></strong></span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;"><span style="color: black;">有些</span>期刊出版<span style="color: black;">公司</span><span style="color: black;">经过</span>检测工具<span style="color: black;">发掘</span>了AIGC代写论文的痕迹。“从去年7月底到<span style="color: black;">此刻</span>，<span style="color: black;">咱们</span><span style="color: black;">发掘</span>涉嫌AI写作的论文数据<span style="color: black;">每一个</span>月都在<span style="color: black;">提升</span>，大<span style="color: black;">大概</span>六七十篇的<span style="color: black;">文案</span>疑似<span style="color: black;">运用</span>AI的程度超过了50%。”《中华医学杂志》社有限责任<span style="color: black;">机构</span>新<span style="color: black;">媒介</span>部<span style="color: black;">专家</span>沈锡宾介绍。</span></span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">沈锡宾向记者展示了检测过程：一篇论文经过检测系统后，会<span style="color: black;">表示</span>疑似AI生成占全文比重，<span style="color: black;">关联</span>疑似段落<span style="color: black;">亦</span>会被标红。但记者<span style="color: black;">重视</span>到，和传统的查重报告单<span style="color: black;">知道</span>标注重复痕迹<span style="color: black;">区别</span>，AIGC检测报告单只<span style="color: black;">指的是</span>出某些文本AIGC的“置信度”，并<span style="color: black;">不可</span>回答<span style="color: black;">为何</span>是这个值。</span></span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">“这使得报告单<span style="color: black;">常常</span>只起到参考和警示<span style="color: black;">功效</span>。”柯春晓说。</span></span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;"><span style="color: black;">日前</span>，人工智能大模型正在以“周”为单位进行迭代升级。<span style="color: black;">怎样</span>适应<span style="color: black;">持续</span>升级的技术，是摆在AIGC检测工具面前的一道必答题。</span></span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;"><span style="color: black;">做为</span><span style="color: black;">运用</span>者的人类本身<span style="color: black;">亦</span>在<span style="color: black;">持续</span>“进化”。“类似人们逃避查重的方式，<span style="color: black;">倘若</span>人们<span style="color: black;">认识</span>到AI检测的方式，<span style="color: black;">亦</span><span style="color: black;">能够</span>重新组织<span style="color: black;">关联</span>内容，对AI生成的文本进行人工润色。<span style="color: black;">这般</span>很可能就检测不出来了。”沈锡宾说。</span></span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">作<span style="color: black;">坏处</span>与反作<span style="color: black;">坏处</span>的过程，实质上是场“猫鼠游戏”。只要技术<span style="color: black;">持续</span>升级，两者间的博弈就不会停止。</span><strong style="color: blue;"><span style="color: black;"><span style="color: black;">日前</span>，AIGC检测技术仍处在萌芽期。<span style="color: black;">怎样</span>对AI生成的虚假<span style="color: black;">照片</span>、虚假数据进行识别仍是难点。<span style="color: black;">因此呢</span>，人们引入智能检测技术的<span style="color: black;">同期</span>，<span style="color: black;">亦</span>要<span style="color: black;">创立</span>人工审查机制。</span></strong></span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">“审稿人要当好‘守门人’，发挥同行评议的<span style="color: black;">功效</span>，仔细甄别判断论文的数据<span style="color: black;">是不是</span>和认知存在偏差。出版<span style="color: black;">公司</span><span style="color: black;">亦</span><span style="color: black;">能够</span><span style="color: black;">需求</span>作者<span style="color: black;">供给</span>原始数据，多管齐下，<span style="color: black;">保证</span><span style="color: black;">研究</span>诚信。”沈锡宾说。</span></span></p>
    <div style="color: black; text-align: left; margin-bottom: 10px;"><img src="https://p3-sign.toutiaoimg.com/tos-cn-i-6w9my0ksvp/4f13dc854a6e48afbc7e57b7db45f0db~noop.image?_iz=58558&amp;from=article.pc_detail&amp;lk3s=953192f4&amp;x-expires=1727620016&amp;x-signature=ReY9Zd09zzN%2BZWvs3Qqa69ONYPg%3D" style="width: 50%; margin-bottom: 20px;"></div>
    <h1 style="color: black; text-align: left; margin-bottom: 10px;">技术向善要他律更要自律</h1>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">加强技术治理的<span style="color: black;">同期</span>，各方都在翘首以盼，期待达成某些共识以及<span style="color: black;">关联</span>政策尽快出台。“教育、<span style="color: black;">研究</span>、出版各方都很关注AIGC<span style="color: black;">运用</span>的边界，期待对<span style="color: black;">恰当</span><span style="color: black;">运用</span>AIGC形成一个共识性规范。”知网技术专家呼吁。</span></span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">其实，早在去年初，中国科学技术信息<span style="color: black;">科研</span>所（以下简<span style="color: black;">叫作</span>中信所）就牵头爱思唯尔、施普林格·自然、约翰威立等国际知名出版集团和<span style="color: black;">研究</span>信息分析<span style="color: black;">公司</span>，在广泛调研并梳理业内<span style="color: black;">关联</span><span style="color: black;">科研</span>和探索工作的<span style="color: black;">基本</span>上，完<span style="color: black;">成为了</span>中英文版的《学术出版中AIGC<span style="color: black;">运用</span>边界指南》（以下简<span style="color: black;">叫作</span>《指南》），并于去年9月20日在国内外同步发布。</span></span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">去年12月21日，科技部发布的《负责任<span style="color: black;">科研</span><span style="color: black;">行径</span>规范指引（2023）》（以下简<span style="color: black;">叫作</span>《指引》）<span style="color: black;">更加是</span>受到了业内的广泛关注。</span></span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;">《指引》和《指南》就<span style="color: black;">怎样</span>负责任地<span style="color: black;">运用</span>AIGC，解答了令<span style="color: black;">研究</span>工作者、期刊编辑、审稿人困惑的<span style="color: black;">有些</span>问题。</span></strong></span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;"><span style="color: black;">首要</span>是披露问题。《指引》提出，<span style="color: black;">运用</span>生成式人工智能生成的内容应<span style="color: black;">知道</span>标注并说明其生成过程，<span style="color: black;">保证</span>真实准确和尊重他人知识产权。《指南》中<span style="color: black;">更加是</span><span style="color: black;">供给</span>了声明的模板，供<span style="color: black;">研究</span>人员参考。</span></span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;"><span style="color: black;">针对</span><span style="color: black;">有些</span>人想用AIGC投机取巧的<span style="color: black;">行径</span>，《指引》<span style="color: black;">知道</span>提出，不得<span style="color: black;">运用</span>AIGC直接生成申报材料；《指南》规定，AIGC<span style="color: black;">不该</span>该用来产生<span style="color: black;">科研</span>假设、直接撰写整篇论文文本、解释数据、得出<span style="color: black;">科研</span>结论。<span style="color: black;">科研</span>人员<span style="color: black;">运用</span>的数据必须是<span style="color: black;">科研</span>人员进行实验并收集所得，如<span style="color: black;">运用</span>AIGC<span style="color: black;">供给</span>的统计分析结果需进行验证。</span></span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">随着AIGC的<span style="color: black;">运用</span>边界<span style="color: black;">持续</span>清晰，越来越多的出版<span style="color: black;">公司</span>达成共识，制定了<span style="color: black;">运用</span>规范。施普林格·自然集团发言人介绍说，<span style="color: black;">她们</span><span style="color: black;">日前</span><span style="color: black;">已然</span><span style="color: black;">知道</span>了<span style="color: black;">相关</span>作者身份和图像方面的规定。例如，人工智能不能担任作者，真正作者如<span style="color: black;">运用</span>大语言模型须加以透明描述，AI生成的图像<span style="color: black;">一般</span><span style="color: black;">不可</span>用于<span style="color: black;">发布</span>等。</span></span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">“《科学》杂志在去年1月份发布的政策是禁止<span style="color: black;">运用</span>任何AIGC工具。而11月16日<span style="color: black;">她们</span>更新了投稿规则、放宽了限制，<span style="color: black;">暗示</span>只要进行了适当披露，<span style="color: black;">运用</span>工具是<span style="color: black;">能够</span>接受的。”中信所博士郑雯雯说道。</span></span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><strong style="color: blue;"><span style="color: black;">“《指引》覆盖较为全面，对AIGC的<span style="color: black;">运用</span>总体呈现出平衡<span style="color: black;">包涵</span>、敏捷治理的态度，而非一味禁止。这<span style="color: black;">亦</span>说明治理的目的并不是阻止<span style="color: black;">研究</span>工作者<span style="color: black;">运用</span>新一代人工智能技术，而是让<span style="color: black;">研究</span>工作者能够负责任地去<span style="color: black;">运用</span>。”</span></strong>梁正<span style="color: black;">说到</span>，在政策制定的<span style="color: black;">行径</span>框架之下，还要关注学科差异问题。“<span style="color: black;">运用</span>AIGC可能因学科的<span style="color: black;">区别</span>而有所差异，其伦理问题<span style="color: black;">亦</span>要<span style="color: black;">按照</span>学科特点细化。”</span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">例如，在自然科学<span style="color: black;">行业</span>，AIGC的强大功能<span style="color: black;">更加多</span><span style="color: black;">表现</span>在数据处理<span style="color: black;">行业</span>，<span style="color: black;">倘若</span>失范<span style="color: black;">运用</span>，<span style="color: black;">常常</span>难以<span style="color: black;">发掘</span>。而<span style="color: black;">针对</span>人文社科<span style="color: black;">行业</span>，直接<span style="color: black;">运用</span>AIGC生成内容的痕迹非常容易被<span style="color: black;">发掘</span>，尤其是在高水平的<span style="color: black;">科研</span><span style="color: black;">其中</span>，优劣之分更为<span style="color: black;">显著</span>。</span></span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">“<span style="color: black;">因此呢</span>，<span style="color: black;">针对</span>更加注重文字表达、数据资料支持的学科，<span style="color: black;">例如</span>企业管理、理工科、医学等，需要防范产生虚假的数据集或论证材料。”梁正说，“对AIGC<span style="color: black;">运用</span>的披露程度、疑似度的数据指标等，都需要学术<span style="color: black;">一起</span>体进一步探索，来推动形成广泛共识。”</span></span></p>
    <div style="color: black; text-align: left; margin-bottom: 10px;"><img src="https://p3-sign.toutiaoimg.com/tos-cn-i-6w9my0ksvp/25e16f72a7b448c1ad039b9c9e97fe18~noop.image?_iz=58558&amp;from=article.pc_detail&amp;lk3s=953192f4&amp;x-expires=1727620016&amp;x-signature=1NypSmJebciB8LnBX8Qb0Q9zpl0%3D" style="width: 50%; margin-bottom: 20px;"></div>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;"><span style="color: black;">另外</span>，尽管国家出台了相应的规则，但从<span style="color: black;">外边</span>监督到行业自治还需要一个过程。</span><strong style="color: blue;"><span style="color: black;">AIGC的<span style="color: black;">运用</span><span style="color: black;">触及</span><span style="color: black;">包括</span><span style="color: black;">科研</span>人员、出版<span style="color: black;">公司</span>、<span style="color: black;">关联</span>行业组织、政府等方方面面。<span style="color: black;">怎样</span>厘清各方关系，各司其职是关键。“</span></strong>简单说，<span style="color: black;">便是</span>出了问题，谁来查？有<span style="color: black;">无</span>能力查？”郑雯雯强调。</span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">记者<span style="color: black;">认识</span>到，中华医学会杂志社在今年1月9日<span style="color: black;">颁布</span>了其<span style="color: black;">针对</span>AIGC技术<span style="color: black;">运用</span>的<span style="color: black;">相关</span>规定。其中不仅<span style="color: black;">触及</span>了作者要遵守的细则，还提出了查处方式——经编辑部研判的违反AIGC<span style="color: black;">运用</span>的情形，将直接退稿或撤稿；情节严重者，将<span style="color: black;">加入</span>作者学术失信名单。</span></span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">“<span style="color: black;">咱们</span>下一步的<span style="color: black;">目的</span>是把存在问题的<span style="color: black;">文案</span>作一个归纳总结，进一步摸清AIGC<span style="color: black;">运用</span>的规律，为科学治理<span style="color: black;">累积</span>经验。”沈锡宾说。</span></span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">“尽管新兴技术有着潜在<span style="color: black;">危害</span>，但<span style="color: black;">亦</span>有着无可比拟的<span style="color: black;">优良</span>，不宜一味封堵，而是要做好引导、<span style="color: black;">恰当</span>合规地<span style="color: black;">运用</span>新技术。”郑雯雯<span style="color: black;">暗示</span>，</span><strong style="color: blue;"><span style="color: black;">归根到底，科学<span style="color: black;">科研</span>的主体是人。<span style="color: black;">倘若</span>心中的那杆“秤”倾斜了，即使再完善的监管政策、再高端的检测技术，<span style="color: black;">亦</span>难以抵挡学术不端的侵袭。</span></strong></span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">梁正<span style="color: black;">亦</span>强调，<span style="color: black;">做为</span><span style="color: black;">研究</span>诚信的<span style="color: black;">第1</span>责任人，<span style="color: black;">研究</span>人员<span style="color: black;">必定</span>要保持严谨的学术态度，关注<span style="color: black;">科研</span><span style="color: black;">行业</span>的真问题，坚守学术<span style="color: black;">科研</span>的基本原则，如原创性和透明性；<span style="color: black;">知道</span>认识到ChatGPT等工具的潜在<span style="color: black;">危害</span>，避免<span style="color: black;">运用</span><span style="color: black;">欠妥</span>而<span style="color: black;">导致</span>学术不端。</span></span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">“<span style="color: black;">研究</span>诚信和伦理是<span style="color: black;">研究</span>的生命线，<span style="color: black;">研究</span>人员<span style="color: black;">必定</span>要存敬畏、有底线。一旦在这方面有瑕疵，职业生涯或将葬送。”梁正提醒。</span></span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;"><span style="color: black;">（文中温睿、徐彬均为化名）</span></span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">文中<span style="color: black;">照片</span>除标注外均来自视觉中国</span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">编辑：张琦琪</span></p>
    <p style="font-size: 16px; color: black; line-height: 40px; text-align: left; margin-bottom: 15px;"><span style="color: black;">审核：张爽</span></p>




j8typz 发表于 2024-10-26 06:28:53

期待更新、坐等、迫不及待等。
页: [1]
查看完整版本: 你用AI写的论文，已然被“盯”上了！