Reflecting Reality, Amplifying Bias? Using Metaphors to Teach Critical AI Literacy

Conceptual Metaphor Theory

Teaching CAIL presents unique pedagogical challenges, as educators must help learners grasp the complex technological systems and their societal implications. Metaphors may offer a valuable solution by serving as a powerful learning resource for sparking discussions and creative thinking about AI systems. Although the concept of metaphor is familiar to most people through formal schooling, informal conversation, and popular media, its scholarly study is distinct and has evolved into a rich field of enquiry. The study of metaphor is embedded in multiple frameworks that aim to describe topics such as language, thought, and communication (Gibbs Jr. 2008), with multiple handbooks, volumes, and academic journals dedicated to investigating the relationship between metaphors and the social world.

In this paper, we frame our understanding of metaphor through Conceptual Metaphor Theory (CMT), developed in Lakoff and Johnson’s seminal work Metaphors We Live By (Lakoff & Johnson 2003). Following this theoretical framing, we contend that metaphors are not simply turns of phrase or rhetorical flourishes, but are powerful, pervasive conceptual tools for structuring, restructuring, and even creating reality (Batten 2012). In essence, this form of metaphor is operationalised through understanding and experiencing one kind of thing in terms of another (Lakoff & Johnson 2003). Furthermore, metaphors can fulfil multiple social and cognitive functions, ranging from strategic persuasive devices (Ferreira, Lemgruber & Cabrera 2023) to acting as constructs to help us organise our knowledge of the world (Saban 2006) and understand abstract concepts (Niemeier 2017). In explaining the relationship between metaphor and thought, Kövecses (2020) provides an example of the ‘journey of life’. Such usage of the ‘journey’ metaphor does not necessarily only refer to how we speak about life, but also how we think about life, and subsequently, how we may go on to act in this belief that life is a journey. Given that fostering CAIL requires critical thinking and that metaphors can help us restructure our reality (Batten 2012), there is a natural alignment between these two concepts for pedagogical use.

Metaphor in Education

Historically, metaphors have been used to facilitate education to expand students’ minds and promote critical thinking (Low 2008). Instruction often requires the description and discussion of complex meanings to progress understanding, and metaphors are appropriate tools for achieving this objective (Carter & Pitcher 2010). As instruction requires moving from known to unknown and concrete to abstract, metaphors can use concrete examples to explain abstract principles (Clarken 1997) and can be used for a variety of learning activities, such as finding memorable labels for complex concepts or helping learners understand challenging learning materials (Low 2008). Metaphors may be employed as a powerful learning mechanism even in early childhood, with studies showing that three- and four-year-olds can use metaphors to make inferences about the functional features of objects (Zhu & Gopnik 2023). Metaphors have also been studied in other disciplines which require close social relations and dialogue, such as psychotherapy, with authors noting that the use of appropriate metaphor can lead to positive client outcomes and cognitive engagement (McMullen & Tay 2023). Indeed, there is a rich body of literature describing how metaphors are used to refer to the teaching and learning process itself (Alger 2009; Batten 2012; Clarken 1997; Hager 2008; Saban 2006), with Hager (2008) claiming that humans are unable to think about learning without employing some form of metaphor.

Recent research has empirically validated the value of metaphors in various classroom contexts in higher education. For example, Pager-McClymont & Papathanasiou (2023) demonstrated the value of using CMT to teach English for Academic Purposes (EAP), by using the ‘A is B’ metaphor structure to teach composition, describing arguments (A) as buildings (B) and writing (A) as cooking, eating, and digesting (B). In another related example, Haidet et al. (2017) explored the use of jazz music as a metaphor for teaching effective communication strategies in patient-doctor interaction, noting positive results in using this metaphor to engage students in medical training. Jin et al. (2025) have explored the use of metaphors by students when describing their experiences with GenAI, including representative metaphors such as “high heels” for technical support, a “compass” for text development, and a “drug” for the potential threats and addictive nature of overreliance on GenAI applications.

At the same time, it is important to be aware of the limitations of metaphor use in an educational environment. Inappropriate uses, opaque uses, or metaphors which do not have a high degree of similarity between the two topics may create difficulties for learners or impede the educational process. The question of culture also arises when selecting metaphors, especially when teaching a foreign language. To illustrate, Low (2008) notes that as metaphors may be culturally specific, if teaching learners the skill of language proficiency in a foreign language, there is an argument as to whether it is more appropriate to teach cultural concepts first rather than teach the metaphor. Furthermore, the limits of metaphors are an important consideration when teachers are engaged in the process of selecting one to use. Carter and Pitcher (2010) explain this by drawing attention to the ubiquitous use of metaphors in teaching electricity. The authors explain that when teaching the concept of electricity, the metaphor of electricity as water is commonly applied. The reasoning behind this is that there is a perceived similarity in the ‘flow’ between electrons in a wire and water in a pipe. However, they highlight that extending this metaphor results in a conceptual break, as there are significant differences in some areas of flow; electron flow cannot be seen, whereas water can, and a broken pipe results in water continuing to flow, but electricity in a wire will stop (Carter & Pitcher 2010). Consequently, while there is significant evidence that metaphors are appropriate and beneficial concepts to use in educational processes, their usefulness is limited; metaphors must be appropriate and understandable, and the limits of comparison should be considered by the educator. We draw on these principles to develop a set of criteria that can be applied when choosing to use metaphors to illustrate information about AI technologies.

Metaphors and Artificial Intelligence

The field of AI is no stranger to metaphors, and in the current era of developing AI technologies, metaphors are increasingly common, although some have been described as problematic and anthropomorphising (Furze 2024). At the same time, despite the burgeoning research on AI and GenAI, our initial literature search using the Scopus and Web of Science (WoS) databases resulted in few recent studies on the relationship between AI and metaphor. Despite not being a highly active area of research, explorations of metaphors and AI have however been taking place for a significant period. Gozzi (1994), for example, used a dataset of articles in the press at the earliest stages of development in the field of AI, from 1966 to 1970, and found that common metaphors for computers were ‘brains’ and that this also extended to the description of thought as a computational process, resonating with the common metaphors for AI that we encounter in popular media today.

More recent work on the intersection of AI and metaphor has explored the use of AI in science fiction as a metaphor for other aspects of human existence (Hermann 2023), and studies have been undertaken to ascertain whether Large Language Models (LLMs), such as GPT-4, can interpret literary metaphors (Ichien, Stamenković & Holyoak 2024). Carbonell, Sánchez-Esguevillas and Carro (2016) describe the connections between computational metaphors for the brain and vice versa, arguing that this reinforces the relationship between the brain and computers, and that this subsequently has an effect that shapes society. In the realm of explainable AI, ways of explaining AI chatbots using the metaphor of fermentation and bread-making have also been explored (Nicenboim et al. 2023). Van Es and Nguyen (2024) conducted a semiotic analysis of ChatGPT’s self-representation, noting that the language model seemed to imply a metaphorical representation of itself, and Hunger (2023) describes the historical process of anthropomorphising metaphors related to AI.

Rehak (2021: 155) argues that powerful metaphors of AI are used to perpetuate the deterministic myths of AI technology. Even the term “Artificial Intelligence” itself is problematic, since much of the technology is neither artificial nor intelligent (Crawford 2021). Machine learning algorithms and datasets which underpin these systems also serve to distance the output from human responsibility, complicating earlier ethical debates (Campolo & Crawford 2020).

The language surrounding the processes of AI are also commonly anthropomorphised. Rehak (2021) points to words such as “‘recognition’, ‘learning’, ‘acting’, ‘deciding’, ‘remembering’, ‘understanding’” in processes carried out by machine learning algorithms, comparing the anthropomorphic words used in the field of AI to the more abstract language typically used in mathematics. The Royal Society (2018) are critical of this kind of language as part of the AI narrative, claiming that it causes a disconnect from the technology itself which may contribute to a “hype bubble”, public fears about technology and subsequent reluctance to adopt beneficial technologies, and a distortion of the discourse surrounding the future of digital technology.

One recent work by Anderson (2023) focuses on the metaphorical framing of Large Language Models such as ChatGPT as either a tool or a collaborator, calling for activities which enable us to understand the biases, inaccuracies, and faults of such tools and foster the development of students’ digital literacy. At the same time, the author points out that there is a need to study how metaphorical language is used when discussing ChatGPT, as such uses could complicate, rather than aid, how we understand these technologies. Furthermore, conveying the functions and malfunctions of LLMs using precise and accurate metaphors can help foster an understanding of how these tools work among the public and academics (Smith, Greaves & Panch 2023). Ye and Li (2024) used a corpus linguistics approach to investigate conceptual metaphor in the European Union AI Act (EU AIA), finding that metaphors used included those related to the concepts of ‘Journey’, ‘Human’, ‘War’, and ‘Object’.

Specific research related to education has explored how learners in educational contexts perceive AI technologies through metaphors. Yan, Sun and Zhao (2024) investigated Chinese EFL learners’ metaphorical conceptualisations of GenAI and found that participants viewed GenAI through multiple metaphorical categories such as humans, tools or machines, brains, resources, food and drink, and medicine. Given this diversity of conceptualisation, the authors call for further work to promote digital literacy for students using GenAI. A similar study was undertaken with student teachers in Germany (N = 100) (Şentürk & Akol Göktaş 2024), which found that participants conceptualised GenAI as a library, student, talking pool, warehouse, world, star, game, human, and brain, among others. In relation to the findings of Yan, Sun and Zhao (2024), corresponding findings of categories such as ‘humans’ offer insight into potential cross-cultural similarities in how we use metaphor to structure understanding of these technologies.

Gupta et al. (2024) explored how using discussions related to metaphors for AI can assist in developing awareness of how AI systems work, as a method of fostering Critical AI Literacy. By engaging in a digital collaborative autoethnographic study, the authors developed a set of metaphors derived from popular online media, scholarly literature, social media, and research participants. The authors contend that such an approach is pedagogically powerful as it can stimulate the affective domain, thus benefiting learning. Crucially, these authors offer advice for teaching CAIL through metaphors, which we foreground in our work, suggesting that educators could ask students to share metaphors that they have discovered, and then challenge them to come up with additional input to extend or offer a new metaphor to deepen their understanding of these complex concepts.

Framework Development Process

To demonstrate the ways in which CAIL can be fostered by using metaphor, we opted to engage in a rigorous, reflexive, and creative investigation to develop a potential approach. We aimed firstly to identify appropriate metaphors and then create sample learning activities which are firmly embedded in existing frameworks for guiding AI competency development in education. Consequently, we selected the UNESCO AI competency framework for students (Miao et al. 2024) to underpin our learning activity development. Our first task was to generate a list of popular metaphors used to describe AI systems. As one of the benefits of metaphor is their ubiquity in everyday discourse, we aimed to choose metaphors which have widespread appeal and potential familiarity to both teachers and students. To do this, we adopted a similar approach to Gupta et al. (2024), drawing on existing literature on traditional and online media, social media, and scholarly work to arrive at a list of commonly used metaphors. Through collaborative discussion among the authors, we developed a set of criteria against which to measure the suitability of these metaphors for classroom activities and assessed their ability to foster AI literacy by evaluating their alignment with curricular goals for developing AI literacy in UNESCO’s AI competency framework for students (Miao et al. 2024).

The UNESCO AI competency framework for students outlines three levels of competency: Understanding, Applying, and Creating. Among these, the most suitable area for CAIL to be developed using metaphors is Understanding. Miao et al. (2024) explain that at this level, students are required to develop an awareness of what AI is and be able to interpret different ethical issues, technical knowledge, and underlying processes that power AI. The authors also suggest that real-life practices should be integrated to help support understanding. Following these principles and through discussion and consensus building, we created four criteria for evaluating the appropriateness of metaphors to guide the understanding of these concepts: Accessibility, Explanatory Power, Critical AI Literacy Potential, and Pedagogical Utility. Descriptions of these criteria are presented in Table 1.

We opted to use these criteria holistically and interpretively, rather than assigning each criteria a weighting or quantitative score for relevance. Instead, we relied on the criteria as a guiding framework for consensus-building discussions. After deciding on these criteria for appropriate metaphor selection, we aimed to develop an exhaustive list of potential metaphors for AI that we could appraise using these criteria. This aspect of the research was intentionally reflexive and required the research team to draw on our personal experiences. Our research team consisted of three individuals, who each have expertise as investigators, educators, and consultants on issues relating to AI and education. As a result, we collaboratively discussed and shared our existing knowledge of common metaphors, including those that we had referred to in our own work previously, those we had come across in established literature, encountered in the classroom, or found in other environments.

We arrived at a final list of 13 metaphors based on these consensus-building discussions in light of our appraisal criteria, and these are displayed in Table 2. Where there is a clear source that we drew on in selecting our metaphor, we have included a citation. Others are not traceable to any singular published source material, and thus no source is supplied.

Following the development of this list, through collaborative discussions, our research team holistically re-assessed each metaphor against our established criteria (Accessibility, Explanatory Power, Critical AI Literacy Potential, and Pedagogical Utility) and their alignment with UNESCO’s AI competency goals. This qualitative analysis was mindful of Low’s (2008) emphasis on cultural accessibility, and Carter and Pitcher’s (2010) caution about how metaphors can break down when extended too far. Batten’s (2012) work on how metaphors can reconstruct understanding in educational contexts guided our analysis of how each metaphor functions pedagogically. Recent empirical studies by Yan, Sun and Zhao (2024) and Şentürk and Akol Göktaş (2024) validated our approach, demonstrating that learners naturally conceptualise AI through multiple metaphorical categories; therefore, we sought broadness across the chosen metaphors, making a final selection based on a combination of alignment with our selected criteria, as well as the UNESCO AI competency curriculum goals and elements of understanding in the AI competency framework that they most closely matched (Miao et al. 2024). We put these into an ‘A is B’ structure, as shown in Table 3.

Activity 1: AI is a Funhouse Mirror

Activity 2: AI is an Echo Chamber

Activity 3: AI as a Map – Representation, Power and Bias

Activity 4: AI is a Black Box

AI Usage Disclaimer

This study used GenAI tools for revision and editorial purposes throughout the production of the manuscript. Models used were ChatGPT (GPT-4o) and Claude 3.5 Sonnet. The authors reviewed, edited and take responsibility for all outputs of the tools used in this study.