国际视野

AI is transforming how science is done. Science education must reflect this change.

There is growing interest in the use of artificial intelligence (AI) in science education. Many issues and questions raised about the role of AI in science education target primarily science learning objectives. They relate to AI’s capacity to generate tools for teaching, learning, and assessment, as well as the advantages and disadvantages of using such tools. But another important discussion receiving far too little attention in science education concerns how AI is transforming the nature of science (NOS) itself and what such transformation implies for the education of young children. For education, it is critical to ask what AI-informed NOS is, what skills it demands of learners, and how schools can aim to achieve them.

With respect to the advantages of AI for teaching, learning, and assessment, the use of simulations, including immersive learning experiences, has been advocated as an important benefit. Similarly, educators have observed that AI presents a powerful means to personalize education by tailoring content and experiences in ways that may not have previously been possible. For example, a student’s engagement with a task may be monitored closely and appropriate feedback can be provided in a specific manner where feedback is most needed. In relation to potential disadvantages, questions such as the following have been raised: What becomes of learning when a student can easily compose text for homework using AI tools? How can students’ understanding be measured in a way that ensures the measurement is about the learning and not about a residual of technology?

Some of the concerns about the impact of AI on learning assume outdated notions of human learning. Traditional science education has promoted transmission of facts and recall of information as indicators of learning. For example, traditionally, students may have been expected to memorize the chemical equation of photosynthesis or be able to recite Ohm’s law. In this depiction of learning, information would be easily retrieved through AI, rendering ambiguity in students’ learning outcomes. By contrast, more contemporary perspectives on learning advocate skills such as critical thinking as important outcomes of learning, which can potentially be copied to an extent but are difficult to mimic through AI. Future-oriented skills, such as scenario thinking, systems thinking, and managing uncertainty and complexity, require more than recall or even management of big datasets. They imply considerable creativity and innovation. Some cognitive psychologists argue that although AI can help to summarize and generalize existing information, it is not designed to fulfill more sophisticated human skills that require innovation, such as theory formation. However, emerging research and development in AI is challenging such views—for example, by exploring the potential for AI systems to highlight blind spots in scientific hypotheses and help generate new questions.

 

In terms of science education and NOS, there is evidence that stakeholders engaged in producing educational policy acknowledge the importance of NOS and draw on research findings about effective teaching and learning of NOS. However, further articulation is needed to unpack the relationship between AI and NOS for educational research and practice as well as policy. Contemporary reflections on NOS in science education research have yet to address AI and its implications for how the scientific endeavor is changing. The conventional gap between professional science and school science can be wide, but it now seems to be increasing even at a faster pace.

AI is already influencing how science is done. Scientists are using AI to generate hypotheses, design experiments, collect and interpret data, and gain insights that might not have been possible by using traditional scientific methods alone. In terms of the reasoning and knowing aspects, scientists often construct models from data to explain and predict phenomena. With the advance of AI, datasets can help scientists make sense of an enormous amount of information. But AI systems can also report misleading information if the data on which the systems are trained and operating are biased or not reliable. The abundance and quality of datasets are known to be biased, often unintentionally. In health data, for instance, AI-based dermatology algorithms have been shown to diagnose skin lesions and rashes less accurately in black people than in white people because the models are trained on data predominantly collected from white populations.

Professional bodies are making recommendations for responsible use of AI in scientific research. These raise issues of transparency, risk, and participatory methods that are noteworthy for how AI-informed science ought to develop. Transparency calls for clear documentation of participants, datasets, models, biases, and uncertainties. Risk implies the management of risks and biases in datasets and algorithms and how they might affect the outcomes including unintended consequences. Participatory methods call for ensuring that research designs are inclusive and engage researchers with communities at risk and include domain expertise. These issues are implying how NOS needs to accommodate cultural norms such as transparency of data and processes, evaluation criteria for scientific knowledge such as evaluation of biases, social values such as managing consequences of risks, and inclusive methodologies to incorporate not only expertise but also appropriation of community knowledge.

In light of the emerging trends in how AI is used in scientific research, the question arises as to how school science can help prepare future scientists to understand NOS in the age of AI. Two questions thus emerge for science education: (i) What does AI-informed NOS mean for school science? (ii) What should be prioritized as aspects of AI-informed NOS at the secondary education level? Some might argue that the use of AI in basic science research is far too sophisticated to be relevant for the purposes of secondary schooling, yet others might claim that young children are not cognitively capable of understanding such advanced means of conducting science research. Such potential positions are open to empirical investigation in which school-based research projects can test the developmental capacity of students and the impact of AI-based interventions on students’ learning of NOS.

Some example aspects of AI in scientific research (e.g., scientific methods, cultural context, and professional recommendations) already have substantial implications for school science. Although science curricula around the world include some of the traditional aspects of scientific inquiry such as experimentation and data collection and interpretation, other relevant aspects such as modeling, although advocated for many years in the science education research community, are still underrepresented in curricula. Similarly, although in some educational systems, the themes of objectivity and accuracy of data may be set as learning outcomes, these aspects of science are virtually nonexistent in a way related to the advancements in AI and its potential contribution to propagating biases. Drawing out some educational adaptations of professional guidelines will help educate future scientists in instilling in them understanding and responsibility about the ethics of AI in scientific research.

The impact of AI on NOS is a tall order for science education. It calls for a systemic approach to reform across the entire sector. There are implications for restructuring the science curriculum, teaching and learning, and teacher education, to name a few aspects. As a matter of priority, the science curriculum content will need to capture nuances about AI-informed NOS, including the developments on how AI is influencing scientific methods and hypotheses. Other aspects, such as models and modeling of large datasets in the context of AI, will also be integrated into secondary education along with themes such as bias in data and risks involved in errors. Such aspects relate to what are currently being referred to as the scientific practices in some curriculum standards, which can act as the specific locus for revision. New teaching and learning tools and strategies will need to be designed and tested to identify effective ways of capturing in the classroom the changing face of science. Many secondary teachers and students are already using platforms such as ChatGPT. In fact, the use of ChatGPT can potentially simulate what scientists themselves are doing as they use such tools for generating literature background for academic manuscripts. Pedagogical strategies such as questioning (e.g., How do we know that this text produced by ChatGPT is accurate?) can be considered for specific purposes of AI-infused NOS learning—for example, to encourage students to generate and apply evaluation criteria for accuracy. Such approaches, however, will need to be accompanied by training of teachers not only to use AI tools and data, but also to understand how science is changing more broadly in the AI age.

Although the AI-infused NOS agenda in science education is a tall order, some existing educational interventions can provide guidelines for aligning its objectives within the education ecosystem and highlighting how to tackle conventional blind spots in educational reform. For example, open schooling networks can be established to foster learning communities involving a range of stakeholders, including students, teachers, teacher educators, scientists, and policymakers. If secondary science education is to raise the future generation of scientists and equip them with timely and relevant skills, then it is essential that secondary science education comes on board with the latest developments in AI-informed scientific research. Otherwise, the gap between professional science and school science is likely to grow at such a rate that by the time secondary students enter university, their understanding of NOS will already be outdated.

 

人工智能正在改变科学研究的方式科学教育必须反映这种变化。

人们对人工智能(AI)在科学教育中的应用越来越感兴趣。关于人工智能在科学教育中的作用提出的许多问题主要针对科学学习目标。它们涉及到人工智能生成教学、学习和评估工具的能力,以及使用这些工具的利弊。但另一个在科学教育中很少受到关注的重要讨论是,人工智能如何改变科学的本质(NOS)本身,以及这种转变对幼儿教育意味着什么。对于教育来说,至关重要的是要问什么是人工智能支持的NOS,它对学习者的技能要求是什么,以及学校如何实现这些目标。

关于人工智能在教学、学习和评估方面的优势,包括沉浸式学习体验在内的模拟的使用一直被认为是一个重要的好处。同样,教育工作者也注意到,人工智能提供了一种强大的手段,可以通过定制内容和体验来个性化教育,这在以前是不可能的。例如,可以密切监视学生对任务的参与情况,并在最需要反馈的地方以特定的方式提供适当的反馈。关于潜在的缺点,有人提出了以下问题:当学生可以使用人工智能工具轻松地写作业时,学习变成了什么?如何测量学生的理解力,以确保测量是关于学习,而不是关于残留的技术?

一些关于人工智能对学习影响的担忧假设了人类学习的过时观念。传统的科学教育将事实的传递和信息的回忆作为学习的指标。例如,传统上,学生可能被要求记住光合作用的化学方程式或能够背诵欧姆定律。在这种学习描述中,信息很容易通过人工智能检索,导致学生的学习结果不明确。相比之下,更现代的学习观点主张批判性思维等技能是学习的重要成果,这些技能在一定程度上可能被复制,但很难通过人工智能来模仿。面向未来的技能,如场景思维、系统思维、管理不确定性和复杂性,需要的不仅仅是回忆甚至管理大数据集。它们意味着相当大的创造力和革新。一些认知心理学家认为,尽管人工智能可以帮助总结和概括现有的信息,但它并不是为了实现需要创新的更复杂的人类技能而设计的,比如理论形成。然而,人工智能的新兴研究和发展正在挑战这种观点——例如,通过探索人工智能系统的潜力,以突出科学假设中的盲点,并帮助产生新的问题。

就科学教育和人工智能而言,有证据表明,参与制定教育政策的利益相关者认识到人工智能的重要性,并借鉴了有关人工智能有效教与学的研究成果。然而,需要进一步阐明人工智能与人工智能之间的关系,以促进教育研究、实践和政策。当代科学教育研究中对NOS的反思尚未涉及人工智能及其对科学努力如何变化的影响。专业科学和学校科学之间的传统差距可能很大,但现在似乎正在以更快的速度扩大。

人工智能已经在影响科学研究的方式。科学家们正在使用人工智能来产生假设、设计实验、收集和解释数据,并获得仅使用传统科学方法可能无法获得的见解。在推理和认知方面,科学家经常从数据中构建模型来解释和预测现象。随着人工智能的发展,数据集可以帮助科学家理解大量的信息。但如果人工智能系统训练和运行的数据有偏见或不可靠,人工智能系统也可能报告误导性信息。众所周知,数据集的丰富程度和质量是有偏差的,通常是无意的。例如,在健康数据方面,基于人工智能的皮肤病学算法在诊断黑人皮肤病变和皮疹方面的准确性低于白人,因为这些模型主要是根据从白人人群收集的数据进行训练的。

专业机构正在就在科学研究中负责任地使用人工智能提出建议。这些提出了透明度、风险和参与性方法等问题,这些问题对于人工智能科学应该如何发展值得注意。透明度要求明确记录参与者、数据集、模型、偏差和不确定性。风险意味着对数据集和算法中的风险和偏差的管理,以及它们如何影响结果,包括意想不到的后果。参与式方法要求确保研究设计具有包容性,并使研究人员与面临风险的社区接触,并包括领域专业知识。这些问题意味着NOS需要如何适应文化规范,如数据和流程的透明度,科学知识的评估标准,如偏见评估,社会价值观,如风险后果管理,以及包容性方法,不仅包括专业知识,还包括对社区知识的挪用。

 

鉴于人工智能在科学研究中应用的新趋势,出现了一个问题,即学校科学如何帮助未来的科学家在人工智能时代理解NOS。因此,科学教育出现了两个问题:(i)人工智能支持的NOS对学校科学意味着什么?(ii)在中学教育层面,哪些方面应优先考虑人工智能知识?一些人可能会认为,在基础科学研究中使用人工智能过于复杂,与中学教育的目的无关,而另一些人可能会声称,幼儿的认知能力不足以理解进行科学研究的这种先进手段。这些潜在的职位可以进行实证研究,以学校为基础的研究项目可以测试学生的发展能力以及基于人工智能的干预对学生学习NOS的影响。

 

人工智能在科学研究中的一些例子(例如,科学方法、文化背景和专业建议)已经对学校科学产生了重大影响。虽然世界各地的科学课程包括一些科学探究的传统方面,如实验和数据收集和解释,但其他相关方面,如建模,虽然在科学教育研究界提倡多年,但在课程中仍然代表性不足。同样,尽管在某些教育系统中,数据的客观性和准确性的主题可能被设定为学习成果,但在与人工智能的进步及其对传播偏见的潜在贡献相关的某种程度上,科学的这些方面实际上不存在。制定一些专业指导方针的教育调整将有助于教育未来的科学家,向他们灌输对科学研究中人工智能伦理的理解和责任。

 

人工智能对NOS的影响对科学教育来说是一个艰巨的任务。它呼吁对整个行业进行系统性改革。这对重构科学课程、教与学、教师教育等方面都有影响。作为优先事项,科学课程内容需要捕捉到人工智能相关NOS的细微差别,包括人工智能如何影响科学方法和假设的发展。其他方面,如人工智能背景下的大型数据集的模型和建模,也将与数据偏差和错误风险等主题一起纳入中学教育。这些方面与目前在某些课程标准中被称为科学实践的内容有关,可以作为修订的具体场所。需要设计和测试新的教学工具和策略,以确定在课堂上捕捉科学不断变化的面貌的有效方法。许多中学教师和学生已经在使用ChatGPT等平台。事实上,ChatGPT的使用可以潜在地模拟科学家自己正在做的事情,因为他们使用这些工具为学术手稿生成文献背景。提问等教学策略(例如,我们如何知道ChatGPT生成的文本是准确的?)可以考虑用于人工智能注入的NOS学习的特定目的,例如,鼓励学生生成并应用准确性评估标准。然而,这些方法将需要对教师进行培训,不仅要使用人工智能工具和数据,还要了解科学在人工智能时代如何更广泛地发生变化。

 

尽管在科学教育中引入人工智能的NOS议程是一项艰巨的任务,但一些现有的教育干预措施可以为在教育生态系统中协调其目标提供指导,并突出如何解决教育改革中的传统盲点。例如,可以建立开放的学校网络,以促进包括学生、教师、教师教育者、科学家和政策制定者在内的一系列利益攸关方参与的学习社区。如果中等科学教育是为了培养下一代科学家并使他们具备及时和相关的技能,那么中等科学教育就必须与人工智能相关的科学研究的最新发展相结合。否则,专业科学和学校科学之间的差距可能会以这样的速度增长,到中学生进入大学时,他们对NOS的理解已经过时了。