Fear and Awe: Making Sense of Generative AI Through Metaphor

Synthesis: Vallis, Wilson & Casey (2025) design and evaluate a metaphor-analysis workshop for making sense of generative AI in higher education, then thematically analyze the metaphors produced by students, academics, and educational support staff at a large Australian university. They map participant metaphors into four categories — Functions (tool-like, e.g. "Swiss army knife"), Roles (human-like, e.g. "helper," "frenemy"), Qualities (unknowable/unreliable, e.g. "black box," "slippery slope"), and Agency (autonomous/threatening, e.g. "competitor," "sinister robot") — surfacing persistent tensions between human versus machine agency and the known versus the unknowable. The study positions metaphor as a low-tech, embodied, collaborative sensemaking tool for critical AI Literacy, offering a workshop model that moves institutional discussion of GenAI beyond reductive "permit or prohibit" policies.

Key Findings

• A four-category taxonomy of GenAI metaphors. Participant metaphors clustered into Functions (practical, tool-like: Swiss army knife, ideas generator, grammar checker), Roles (human-like relationships: helper, study buddy, personal tutor, frenemy), Qualities (unknowable, unreliable, or magical: black box, tip of the iceberg, slippery slope, genie in a computer), and Agency (volition and threat: competitor, invader, Frankenstein, sinister robot, HAL).
• Two recurring tensions shape sensemaking: human versus machine Learner Agency (is GenAI a tool we use or an autonomous actor that may replace us?), and the known versus the unknowable (opacity/"black box" versus magical-seeming capability). Both map closely onto prevailing literature on GenAI metaphors.
• Metaphors reveal deep ambivalence. The most striking pattern was "friend/foe" dualities — double-edged sword, friendly terminator, frenemy, feeding the beast — reflecting participants' inability to take a definitive stance. Anthropomorphic Roles metaphors risk misplaced trust and over-attributed agency, masking the human, material dimensions of AI.
• Cognitive offloading via tool metaphors is seductive but unsettling. Framing GenAI as a neutral "tool" (hammer, calculator) emphasizes utility and control while downplaying its embedded values and biases — and participants were uneasy about offloading tasks requiring human judgment and Creativity.
• The workshop itself was a validated sensemaking method. Grounded in material, collaborative, no-tech activities (sticky notes, a "gallery" of metaphors), the workshops helped participants surface assumptions, connect across roles and cultures, feel "not alone," and think about the Ethics of GenAI — without requiring technical expertise.

What this means for practice

• Instructors. Run a metaphor workshop before redesigning assessment: having a teaching team surface its functions, roles, qualities, and agency metaphors exposes assumptions that a "permit or prohibit" policy debate hides.
• Instructors. Have learners generate and critique the metaphors themselves — the sticky-note gallery method let participants voice ambivalence (for example "frenemy," "genie in a computer") that more structured discussion formats suppress.
• Instructors. Work directly on the friend/foe dualities: address misplaced trust in anthropomorphic helper and tutor roles, and the unease about offloading tasks that require human judgment and creativity.
• Designers. Use the four metaphor categories to move institutional discussion beyond binary policy, covering both augmenting-device opportunities and black-box concerns, and to design professional development that addresses fears such as AI replacing teaching roles.

Limitations

• Three workshops were held at a single business school in one metropolitan Australian university, so the metaphor categories reflect one organizational and disciplinary setting.
• The workshop groups were uneven — 15 educational support staff and 10 academics but only 3 business students — and the authors state that running more workshops with students would be helpful.
• The authors acknowledge that the workshop-based elicitation, prompts, and facilitation may have shaped which metaphors were produced, so the categories may reflect the workshop design as well as participants' natural conceptualizations.
• Time constraints limited how fully the metaphors could be analyzed and discussed with participants, and no follow-up measured whether the workshops changed later teaching or policy practice.

Citation

Vallis, C., Wilson, S., & Casey, A. (2025). Fear and awe: Making sense of generative AI through metaphor. Journal of Interactive Media in Education, 2025(1), 14.