January 3, 2025, by Brigitte Nerlich
Food for thought: AI and culinary metaphors
Just before Christmas 2024 I read an article in Nature about AI running out of data. It said that “The Internet is a vast ocean of human knowledge, but it isn’t infinite. And artificial intelligence (AI) researchers have nearly sucked it dry.” Then, on Christmas Eve Aparna Nair said on Bluesky: “I know academics who use AI regularly, for work and play, and I’d really like them, and universities, to tell me how they contend with AI’s bottomless appetites for power and water. Is asking ChatGPT to interpret a Xmas song or making it write your syllabus worth this to you?”
AI and food metaphors
These two remarks made me think again about AI and food metaphors, something I also touched upon last year. In my first post of the new year, I want to be a bit more systematic and see what types or categories of food metaphors we are dealing with.
I collected various sentences and expressions which portray generative AI or AIs (LLMs, chatbots etc.) metaphorically as food and eating in the widest sense and I then extracted several ‘conceptual metaphors’, which I’ll use as headings in this post. Some of them are a bit contrived and if anybody has a better idea, let me know.
Conceptual metaphors are overall mappings across at least two conceptual domains: a conceptual source domain (e.g. food) and the conceptual target domain (e.g. thought), where the conceptual source domain is normally something concrete and experiential and the target domain something more abstract. In this case the conceptual metaphor would be THOUGHT IS FOOD. That conceptual metaphor underlies utterances like: ”This is food for thought”, “I’ll have to digest this” or “Chewing on half-baked ideas”. These mappings are not arbitrary. Rather, they are grounded in our everyday experience of the body and the world we live in.
It’s not surprising that food, which can be used as a source domain for thinking and talking about thought, ideas and intelligence, can also become the source domain for (generative) artificial intelligence, as in the overarching conceptual metaphor AI IS FOOD. That general metaphor has several sub-categories depending on the aspect of food (the stuff you eat, the acts of eating, drinking, feeding, digesting, making etc.) that one wants to focus on. We’ll examine these below.
Let’s start with exploring the various facets of how AIs ingest food, that is, with their acts of culinary consumption, and go from there to what they consume and so on.
AI ingests stuff
We often hear that AI ‘eats’ data, water, power, and even more metaphorically ‘the world’. Some link this metaphor back to the venture capitalist Marc Andreessen saying in 2011 that “Software is eating the world”. Indeed, AI is said to ‘feast’ on user data information and we are told to beware of an increasing digital feast.
AI also ‘drinks’. As we heard at the beginning, AI is ‘sucking’ the ocean of human knowledge dry. It’s also said to have un “unquenchable thirst” and as ‘guzzling’ billions of litres of water.
Worst of all, there is a fear that AI will ‘devour’, art, film, “all of the internet’s written knowledge”, “human culture” even, “product teams whole”. In the end it might devour us, humanity, and, again, as in the case of ‘eating’, it might devour ‘worlds’. But one group is safe: insurance brokers! As one insurance blog post proclaims: “AI devours data but likely won’t eat insurance brokers.”
But AI not only devours data, information, culture, knowledge and us, it also devours energy, electricity and water. The word ‘devour’ associates our musings about AI with our cultural knowledge of all sorts of monsters. AI thus becomes not only a voracious eater with an insatiable appetite, but a potential monster.
That monster has an Achilles heel, however. It is beginning to devour or ‘eat’ itself. Some call this autophagy and some fear that this leads to model collapse disorder like mad cow disease.
Data is food
Let’s now shift the focus from the act of consumption to what is consumed, which, in the case of AI, is data or information, as well as water and energy. As one blog post said: “Data is to AI as food is to humans”. And as food is fuel for humans, so data fuels AI. Another post says: “Data is food for AI” and, in order to benefit from AI “we need to feed it with high quality data”.
These high culinary or nutritional standards are threatened when we ‘feed’ AI processed food in the sense of synthetic data, and things get even worse when AI feeds on its own synthetic data or what one may call ultra processed food, as we have seen above.
One project based on the AI-food analogy even asked: “Is there such a thing as toxic data? Can an artificial intelligence be [food] intolerant?” But what about the stuff that AI produces rather than consumes?
And food/data has to be harvested of course and sowed and reaped – so there is a whole adjacent conceptual field to be examined for metaphors as well…..(what about leaving fields fallow etc etc.)
AI is junk food
Some people ask whether using AIs like ChatGPT is like “eating plastic for your cognition”? While somebody else says on Bluesky: “LLM articles are like diet coke, they are the appearance of the thing but not the thing itself.” One article said “’It looks like cat food’: The grey goop dreamed up by artificial intelligence”. And so, we come to ‘slop’, a term now widely used to describe low-quality media, including writing and images, made using generative AI. Apparently, 2024 has been the year of AI slop.
To counteract this awful diet, a blog post, using and extended food analogy, recommends a varied diet when it comes to data as AI’s culinary “base ingredients”.
AI is a digestive system
People that consume food, don’t just stop, once they have eaten. After a while they build up an appetite again, they become hungry, and they need feeding again. So do AIs. AI is “hungry for power” and for data (and as we have seen it’s also thirsty). As one blog post asks: “Generative AI Is Data-Hungry: Is the All-You-Can-Eat Buffet Over?” AI is often said to have a big or insatiable appetite for power, energy, water, data and information.
Various scholars have talked about the fact that AI or AIs need to be constantly fed. In a recent book, James Muldoon and Mark Graham, ask: “Who is feeding the machine?” This metaphor is extended in a conversation they have with another expert in that field, namely Antonio Casilli entitled: “In the belly of AI: Feeding the machine”. These scholars make clear that AI feeds off the work or labour of human beings, human beings who, in fact, often work in inhumane conditions. “A.I. is an extraction machine that feeds off humanity’s collective effort and intelligence, churning through ever-larger datasets to power its algorithms.”
Others ask what happens if AI ‘feeds off’ its own answers and others caution users to take care when feeding AI models.
And of course, a digestive system not only needs input, it also has outputs and so AI is said to feed on garbage and to “spit out” garbage.
AI is a food technology
Going beyond consumption and digestion, some food related metaphors map food technologies or appliances onto aspects of AI. AI can be seen as a toaster, for example. In a conversation with Philip Ball, Shannon Valor, the author of book that uses a different AI metaphor “The AI Mirror”, says “that’s what LLMs are—they are mindless toasters that do a lot of cognitive labor without thinking” (focusing on the fact that these machines don’t have the capacity to make meaning). Others have compared AI to a blender rather than a toaster: “AI is a blender that makes human prejudice smoothies.” And I assume there are more examples to be found.
AI is a kitchen
One blogger goes further and uses a whole kitchen analogy to talk (with ChatGPT) about AI. It’s very good and covers all aspects of AI, but you’ll have to read the blog post! (e.g. there is The Kitchen (AI System), The Chefs (Machine Learning and Deep Learning Models): The Prep Team (Data Preprocessing and Vectorization): The Pantry (Vector Store): and much more!)
AI is baking or cooking
AI can be seen quite straightforwardly as baking or cooking you something, such as a text or a picture, but as this piece on Substack about ‘AI as a cake’ makes clear, the metaphor is not that simple. The post invites us to think critically about the baking metaphor. Others have used the cooking metaphor to say that AI can also be compared to a “trained circus bear that can make you porridge in your kitchen” but might also “ransack your kitchen”.
Conclusion
I was just finishing this post when I heard my husband on the phone talking to a friend about AI and saying something like “AI will soon be in everything, I suppose, but the problem is that one doesn’t know whether it has a finger in the pie or not”. Another food metaphor. I bet there are hundreds more used in discussions about the risks and benefits, the promises and pitfalls of generative AI.
Thinking and eating, thought and food, have been conceptually linked for a very long time. As Shakespeare wrote in Sonnet 75: “So are you to my thoughts as food to life”. It’s no surprise then that artificial thought is also framed through the lens of food and eating.
Image: Nature morte avec compotier/Still Life with Compotier (Fruit Dish) by Paul Cézanne, 1879 (ok, I had searched Google Images for ‘still life with computer’ and this came up and I couldn’t resist)
No comments yet, fill out a comment to be the first
Leave a Reply