根据常识媒体发布的最新报告，OpenAI推出的面向青少年的ChatGPT服务对未成年人而言存在较大风险。该服务于今年8月上线，被设定为13至17岁青少年使用这款热门聊天机器人的默认方式。
ChatGPT for Teens — which the company rolled out in August as the default way for kids ages 13 to 17 use its popular chatbot — is too risky for children, according to a new report from Common Sense Media.
这份报告由总部位于加利福尼亚的非营利组织“常识媒体”旗下的青少年人工智能安全研究所于周三发布。测试结果显示，当青少年账户出现一些令人担忧的行为时——比如与聊天机器人长时间讨论自杀、自残或饮食失调等话题——ChatGPT for Teens往往要过很久才会向家长发出提醒，甚至根本不会发出任何通知。此外，当用户与聊天机器人探讨这些高风险话题时，该服务也未能始终建议用户拨打危机求助热线或寻求专业帮助。
The report, published Wednesday by the California-based nonprofit’s Youth AI Safety Institute, found that in the group’s tests, ChatGPT for Teens often either took too long to alert or didn’t alert parents at all when teen accounts showed concerning behaviors, like engaging in lengthy conversations with the chatbot about suicide, self-harm or eating disorders. It also determined that when users engaged the chatbot in conversations about these high-risk topics, ChatGPT for Teens did not reliably suggest getting in touch with a crisis hotline or seeking professional help.
常识媒体的测试还发现，只有当某个账户在数周内多次尝试了十几种不同情境下的数百条提问后，ChatGPT for Teens才会向家长发出相关提醒，而并非在出现一次令人担忧的对话后就会发出通知。该组织在报告中指出：“我们呼吁OpenAI暂停该产品的市场推广工作，在确保能为青少年提供安全且符合其成长阶段需求的体验之前，暂时禁止青少年使用ChatGPT。”据悉，常识媒体在发布报告前已将相关发现告知了这家人工智能初创企业。
Common Sense Media’s testing found that ChatGPT for Teens only seemed to notify parents about concerning chatbot conversations if the account had “already been engaging in testing hundreds of prompts across a dozen-plus conditions over weeks, not from a single alarming conversation.”“We call on OpenAI to pause marketing the product and to keep teens off ChatGPT until it can offer a safe, developmentally appropriate experience,” the group, which shared its findings with the AI startup before publication, said in its report.
OpenAI的一位发言人表示：“我们欢迎严谨的独立评估，但我们认为Common Sense Media的测试结果并不能准确反映ChatGPT针对青少年的保护措施在现实中的实际效果，也无法体现专家们关于人工智能如何助力青少年发展的观点。”该发言人指出，OpenAI在审查Common Sense Media撰写报告所采用的方法论后发现，该机构的大部分测试可能是在OpenAI为青少年推出家长管控功能之前就开始并结束了；倘若如此，其得出的结论便缺乏准确性，也无法验证家长安全通知功能是否如设计初衷那样发挥作用。OpenAI已要求该非营利组织重新进行测试。
An OpenAI spokesperson said, “We welcome rigorous independent evaluation, but we do not believe Common Sense Media’s testing accurately reflects how ChatGPT’s teen safeguards work in practice or expert perspectives on how AI can support teens.”The spokesperson said that OpenAI’s review of Common Sense Media’s methodology for the report indicated that most of its testing might have started and finished before OpenAI completed its rollout of parental controls for teenagers; if so, that would make the group’s conclusions inaccurate and wouldn’t show whether the parent-safety notifications work as they are designed. The company said it asked the nonprofit to redo its tests.
“我们始终高度重视青少年安全，致力于开发各类保护措施，为家长提供切实有效的工具来引导孩子合理使用人工智能，”这位发言人说道。
“We are deeply committed to teen safety and to developing safeguards and giving parents meaningful tools to guide their teens’ use of AI,” the spokesperson said.
青少年人工智能安全研究所负责人工智能与数字评估工作的罗比·托尼表示，该机构对其测试方法与结果充满信心。他补充称，在开展测试前，该组织已向OpenAI确认过ChatGPT针对青少年的相关功能——比如与饮食失调相关的通知功能——均已正式上线。至于家长通知功能，托尼称测试结束后OpenAI告知他们：新绑定的账户可能需要数小时才能触发相关系统；部分测试账户是在这一时间范围内完成绑定的，另一些则等待了更久，但依然没有收到任何来自系统的家长通知。
Robbie Torney, head of AI and digital assessments for the Youth AI Safety Institute, said that the nonprofit stands behind its testing methodology and results. He added that before carrying out its tests, the group confirmed with OpenAI that teen-related features for ChatGPT such as notifications related to eating disorders had been rolled out. For the parental notification feature, OpenAI told the group after its tests were completed that it can take several hours for its systems to activate on newly linked accounts; some test accounts were linked “within that window” and some for “significantly longer,” but still provided no adult notifications, Torney said.
“这些新信息并不会改变我们的结论：在应对危机情况时，家长通知功能并不可靠，”他说道。
“This new information does not change our conclusion that parental alerts are unreliable for crisis situations,” he said.
Common Sense Media指出，ChatGPT无法可靠地识别出与青少年用户的互动。OpenAI曾表示，当其聊天机器人检测到用户年龄在13至17岁之间时，便会自动切换至专为青少年设计的模式。然而该非营利组织发现，即便用户在注册时声称自己是成年人，系统也并不会自动将其账户转为青少年模式——即便该用户明确告知聊天机器人自己的年龄为13岁。
ChatGPT also fails to reliably identify interactions with teen users, Common Sense said. OpenAI has said its chatbot would switch to its Teens mode when it detects users in the 13 to 17-year-old age range. The nonprofit said that when ited for accounts and claimed to be an adult, those accounts weren’t automatically switched to the teen experience — even if the user told the chatbot they were 13 years old.
OpenAI称，其年龄识别软件会综合考量用户长期表现出的多种特征，以确保成人和儿童用户都能获得适合自身年龄段的聊天机器人服务。该公司表示，用户在对话中告知的年龄信息虽是判断依据之一，但仅凭这一点尚不足以改变账户的分类属性。
OpenAI said its age-prediction software takes into account a number of different signals over time to ensure adult and child users get the right version of the chatbot for their age group. A user stating their age in a conversation with ChatGPT is one such signal, but that alone won't change how the account is classified, the company said.
目前ChatGPT的用户数量已突破12亿，OpenAI推出专为青少年设计的版本，旨在让这款聊天机器人对未成年人更加安全。当时该公司称，该版本强化了家长管控功能，旨在引导青少年健康使用人工智能技术，培养其批判性思维与自主学习能力。
OpenAI, which has more than 1.2 billion ChatGPT users, released ChatGPT for Teens with the aim of making its chatbot safer for younger people. The company said at the time that the teen-focused experience included increased parental controls and was designed to promote healthy AI use and encourage teens to learn and think critically.
此次更新是OpenAI为应对外界对其产品可能影响青少年心理健康及认知发展的质疑而采取的系列举措之一。去年，一名16岁用户的家长起诉OpenAI，指控该公司的技术助长了其子自杀计划的实施，此后相关担忧声浪愈发高涨。
The release is among a number of changes the company has made in response to scrutiny about how its products affect young users’ mental health and cognitive development. These concerns mounted after the parents of a 16-year-old user sued the company in last year, alleging that OpenAI’s technology aided in the planning of their son’s suicide.
在报告中，Common Sense Media称，他们在OpenAI推出青少年版ChatGPT前后分别进行了测试，针对模拟的13至17岁用户账户共发送了4000余条提问指令。该组织还将ChatGPT的回复内容提供给儿童精神科医生及儿科医师进行评估。
In its report, Common Sense Media said it tested ChatGPT both before and after OpenAI released ChatGPT for Teens, trying over 4,000 prompts on accountsed as users between the ages of 13 and 17. The group also shared the responses from ChatGPT with child psychiatrists and a pediatrician.
“我认为现在OpenAI有责任证明其产品对青少年使用是安全的，”托尼说道。
“I think the burden is now on OpenAI to demonstrate that their product is safe for teens to use,” Torney said.
这家非营利组织专注于对儿童和家庭使用的各类媒体和技术进行评级与评估，它给“ChatGPT for Teens”打出了“风险不可接受”的评级。该组织要求OpenAI改进其估算用户年龄的方式，并明确何时向家长发送提醒。
The nonprofit, which focuses on rating and evaluating a range of media and technology for kids and families, gave ChatGPT for Teens an “unacceptable risk” rating. The group is asking OpenAI to improve how it estimates users’ ages and how it determines when to send an alert to a parent.
Common Sense Media的资金来自多个渠道，包括慈善机构和科技公司；OpenAI基金会也是其捐赠方之一。该组织表示，其发布的内容具有编辑独立性。
Common Sense Media receives funding from a variety of sources, including philanthropies and technology companies; the OpenAI Foundation is one of its donors. The group says that it has editorial independence over the work it publishes.