Table 1
Criteria for Evaluating Appropriacy of AI Metaphors.
| CRITERIA FOR METAPHOR EVALUATION | RATIONALE |
|---|---|
| Accessibility | An appropriate metaphor will be familiar to students and accessible to a general audience. This means that the metaphor will not require specialised knowledge in order to understand it. For example, a metaphor which focuses on describing AI systems as a form of quantum entanglement would require a knowledge of what ‘quantum entanglement’ means, thus is relatively inaccessible. |
| Explanatory Power | An appropriate metaphor has the potential to illuminate key aspects of GenAI systems and develop understanding. This means that the metaphor can bring to light an aspect of AI systems in a creative and meaningful way. |
| Critical AI Literacy Potential | An appropriate metaphor may encourage critique of AI systems’ limitations and capabilities or may draw attention to a key aspect of CAIL, such as algorithmic bias potential, or other social, environmental, or ethical impacts. |
| Pedagogical Utility | An appropriate metaphor will support the creation of varied learning activities to support the required learning outcomes from the lesson. Such a metaphor will have the potential to lead learners towards these specific learning objectives. |
Table 2
Initial List of Potential Metaphors to Guide AI Literacy.
| METAPHOR | DESCRIPTION |
|---|---|
| Stochastic Parrot (Bender et al. 2021) | Coined in a highly cited paper to refer to the probabilistic, non-understanding nature of AI language models. |
| Black Box | An opaque system with observable input and outputs but no access to inner processes. |
| Iceberg (Furze 2024) | Generative AI is powered by an enormous, largely unknowable dataset ‘below the waterline’. Consumers and users only interact with a small portion of the model. |
| Funhouse Mirror | Provides a reflected version of reality but in distorted and warped ways. |
| Assistant | An aide that can assist with simple tasks but requires supervision. |
| Loudspeaker (Gupta et al. 2024) | Amplifies and broadcasts existing patterns. |
| Double Edged Sword | A tool or weapon with both beneficial and harmful aspects. |
| Calculator for Words (Willison 2023) | A mathematical calculation of language. |
| Natural Disaster | A powerful force that can be prepared for, but not avoided. |
| Collaborative Artist | A creative partner that can contribute to an artistic process. |
| Map | A representative map of society and culture based on the training data. |
| Pattern Matching Machine (Furze 2024) | A system that identifies and reproduces patterns from data. |
| Echo Chamber | A system that reflects and reinforces existing patterns. |
Table 3
Selected Metaphors, Selection Criteria and Alignment to UNESCO AI Competency Framework.
| SELECTED METAPHOR | RELATION TO SELECTION CRITERIA | ALIGNMENT TO UNESCO AI CURRICULUM GOALS |
|---|---|---|
| AI is a Funhouse Mirror | Accessibility: The concept of a funhouse mirror is familiar across age groups. Explanatory Power: Illustrates how AI systems may distort reality. Critical AI Literacy Potential: May lead to discussion about bias and representation. Pedagogical Utility: Lends itself to physical activities with realia (e.g. a distorted mirror) and links to activities exploring data and algorithmic bias. | CG4.1.3.2 Develop conceptual knowledge on how AI is trained based on data and algorithms. |
| AI is a Map | Accessibility: Draws on familiar concepts of maps as a way of representing the world. Explanatory Power: Demonstrates how AI is a representation but not a true reflection of the world. Critical AI Literacy Potential: Encourages the examination of power structures and colonialism. Pedagogical Utility: Allows for debate and critical thinking regarding power and representation. | CG2.1.1 Surface ethical controversies through a critical examination of use cases of AI tools in education. |
| AI is an Echo Chamber | Accessibility: Echoes a universal physical reality across cultures. Explanatory Power: Demonstrates how AI systems may reinforce ideas, biases, or concepts in the training data. Critical AI Literacy Potential: Leads to the critical discussion of how to mitigate feedback loops and filter bubbles. Pedagogical Utility: Supports numerous practical and personal activities, for example exploring algorithmic advert selection on social media. | CG4.1.4.1 Scaffold critical thinking skills on when AI should not be used. |
| AI is a Black Box | Accessibility: The concept of a black box is a long-standing metaphor with high familiarity explaining technology and systems where inner workings are hidden. Explanatory Power: Helps illustrate the challenge of understanding AI systems. Critical AI Literacy Potential: May promote discussion of transparency and explainability. Pedagogical Utility: May be demonstrable by encouraging learners to generate unexplainable outputs. | CG4.1.2.1Illustrate dilemmas around AI and identify the main reasons behind ethical conflicts. |