Over the last few weeks I have written a trilogy of blog posts about metaphors for AI, trying to survey emerging metaphors as well as those studying those metaphors, and calling for a metaphor observatory. Three posts is a lot to read. For those who want to have a quick overview, here is one. I initially intended this to be a ‘thread’ on Bluesky but that turned out to be a bit fiddly.
***
Since the launch of ChatGBT in late 2022 metaphors for AI have exploded: from stochastic parrots, mirrors, co-pilots, digital plastic, to lawnmowers of thought, and more. These images don’t just decorate our language; they shape how we think, feel, and regulate AI.
In the first post of this trilogy, I argue for the creation of an AI metaphor observatory: a way of systematically watching how people make sense (and nonsense) of AI through metaphors across research, media, policy, and everyday life. The idea is to track not just the metaphors themselves, but the power relations and imaginaries they carry.
Early on, my own metaphor collecting was just haphazard—like picking up seashells on a beach, bringing them home and sorting them into little boxes. Now, an entire academic mini-industry is emerging that studies these shell boxes of metaphors for AI more systematically, across disciplines and domains. It’s time to take stock.
Making the case for an AI observatory
Part 1 of the trilogy briefly maps both this emerging landscape of academic work on metaphors for AI (including a list of references) and some trends in emerging metaphors for AI.
In terms of academic, work we can see analyses of metaphors for AI from AI itself and for itself to taxonomies of metaphor clusters, media and policy studies, social representations, and work in critical AI literacy. Education researchers, in particular, are using metaphors both as objects of study and as tools to teach about AI.
In terms of metaphors themselves, they seem to be becoming more mature, more diverse, more critical, and more methodologically ‘meta’.
More mature means that some core metaphors have settled in: the stochastic parrot, AI mirror, blurry JPG, or lossy compression of the internet all foreground limits, distortions, and risks as much as capabilities. At the same time, more mundane framings like assistant or co-pilot are now widely used but increasingly scrutinised for what they obscure.
Metaphors are also becoming more diverse and context-specific, grounded in lived experience. Think of GenAI as a high-heeled shoe that makes your writing look elegant but might make you fall flat on your face, or AI as autotune for knowledge, smoothing away imperfections at the cost of human texture.
Some of the most vivid metaphors are explicitly critical: ‘eating plastic for your cognition’, ‘spicy autocomplete’, a ‘lawnmower of thought’ that regresses everything to the mean, or a ‘colonising loudspeaker’ that amplifies dominant languages and silences others. These images crystallise concerns about synthetic text, homogenisation, and epistemic injustice.
We have moved from metaphors of wonder and magic, framing AI metaphorically as oracle, crystal ball, or knowledge engine, to more grounded, critical, and pedagogical framings. The question is no longer just “What is this thing?” but “What does this thing do to our practices, students, and knowledge systems?”.
This short survey also exposed many gaps in research that need filling from critically studying metaphors for AI safety or AI infrastructures to studying the history of metaphors for AI over time and tracing this history back in time to salient periods before 2022.
Surveying recent academic literature
Part 2 of the trilogy digs a bit deeper into this emerging academic work listed in part 1 and shows how AI metaphor studies have themselves evolved (and are they are still evolving!). The field is shifting from cataloguing metaphors to critically analysing their social functions, and increasingly to reflexive, pedagogical uses of metaphor in teaching and research.
Many studies now focus on collecting and categorising metaphors: crowdsourced metaphors, speculative metaphors, and those drawn from specific communities and contexts, even LLMs themselves. Others trace how iconic metaphors, such as the ‘stochastic parrot’ or ‘AI mirror’ travel between academia, policy, classrooms, and public discourse.
Education-focused work reveals intriguing differences: for instance, STEM students often reach for support metaphors, while humanities students more frequently use threat metaphors. Metaphors here double as diagnostic tools, revealing perceptions, attitudes, anxieties, and expectations around AI.
In education research, metaphors have become methods. Scholars use them as research instruments to survey stakeholders’ perceptions and as pedagogical tools to teach critical AI literacy, inviting students to interrogate and rework the images through which AI is made thinkable. A whole new field of Critical AI Literacy studies is developing which includes the critical study of metaphors.
Surveying shifts in metaphor usage
Part 3 examines in more detail how metaphors for AI are changing and why that matters. Metaphors for AI have shifted from magical hype to critical reckoning, and now to more nuanced frames.
Since ChatGPT’s launch in November 2022, people have reached for metaphors to make GenAI ‘thinkable’ – from crystal balls and wizards to assistants and co‑pilots. Early on, the tone was wonder, utility, and optimism, with only faint hints of risk. Now the metaphor landscape is dominated by critical and often visceral images. Think ‘theft‑tech’, ‘doomsday machine’, ‘eating plastic for your cognition’, ‘fast food’, and ‘opium’.
Metaphors have also become political and reflexive: ‘colonising loudspeaker’, ‘technology of power’, ‘Western museum’, ‘registry of power’, and “I am a stochastic parrot, and so r u.” They shift focus from what AI does for individuals to what it does to societies, and to how AI talk reshapes how we imagine human cognition, agency, and creativity.
Researchers now use metaphor frameworks – like tool / transformer / threat or the 4T pyramid – as teaching devices for critical AI literacy. Metaphors are no longer just descriptions of AI; they are tools for intervening in debates about power, harm, and responsibility.
Across the time, several patterns stand out: from benefits to concrete harms, from individual use to social systems, from magical to material, and along an anthropomorphising gradient from calculator to overlord. Older images have faded and newer ones are emerging as attention moves from adoption to consequences and from hype to embedding.
The metaphor‑makers have changed too: less tech marketing, more critical scholars, educators, and students reflecting on lived impacts.
Conclusion
Across the trilogy, one question crops up again and again: who gets to coin and stabilise the metaphors that make AI thinkable and debatable and through which AI becomes socially meaningful? Metaphors invite action ‘as if’ the world were configured in a certain way, so they are also tools of power and governance.
An AI metaphor observatory would track not just the metaphors, but the power dynamics behind them. It would use metaphors as social sensors.
The case for the AI metaphor observatory that I made in this trilogy is, ultimately, a call to watch the watchers: to monitor AI metaphors themselves, but also the growing academic and policy apparatus around them. If metaphors make GenAI and LLMs thinkable and debatable, we need collective and participatory spaces that keep those debates open, plural, and critically aware of who is (and is not) doing the framing.
What metaphors are you using for AI? This observatory is a collective project – share your observations in the comments below or tag me on Bluesky using the hashtag: #AIMetaphorObservatory
Image: Seashells on a beach near Valencia, Wikipedia Commons
Making Science Public 
Leave a comment