What magic can teach us about AI
A few months ago I gave a version of this post as a 15 minute talk at The Conference in Malmö, Sweden. You can watch the talk here. A couple of people have asked me for a text version, so I figured I’d write it up.
Metaphors of AI as magic
Over the last few years, we’ve gotten used to the idea that adding AI to software creates something magical.
The little sparkle emoji, that was mainly used in manga, now signals that AI is present.
Beyond the magical emoji, AI demos *look* magical. Sprinklings of AI fairy dust changing everyday software into something new and better. Be it generating full blog posts from typing (to the lamentations of a thousand content marketers)…
…or like this Google Gemini ad, where wielding your phone like a magic wand around your office can help you find lost things. Magical and useful.
Metaphors of AI as “magic” are also part of our lives as practitioners.
When we do UX research on AI services, we often use an approach called Wizard of Oz-ing. Testing something that doesn’t exist (yet), trying to capture insights about how people react to what it *might* look like, or behave, in future.
That hints, I think, to our relationships with AI systems being subtly different those we have with existing digital technologies.Let me give you an example. At the moment, I’m working on something called a Retrieval-Augmented Generation system: a large language model combined with original sources. It’s not publicly available yet, but we use an interface similar to this one from Perplexity AI:
From a UX writing perspective, I’ve been looking at microcopy for users trying to find out what they need to know from this RAG system. Why not just use content design patterns we’ve been using for decades?
Well, technically, the machine learning underpinning a RAG system is quite different to what’s behind conventional search. I don’t believe in overwhelming users with technical detail, but gently signalling this difference in microcopy helps manage user expectations. “Ask” for information, not “Search”. A dialogue, not a command. As practitioners, we’re now designing interfaces with changing, sometimes capricious, outcomes.
What’s going on with the magic metaphor?
Whenever you give a talk like this, you have to ask: So what’s going on here?What is this association between AI and magic? Part of what’s going on is, of course, novelty. As the cliché says: “any sufficiently new technology is indistinguishable from magic.” (We’ve all used this quote in innovation slide decks. Sorry.)
Clichés are cliches for a reason, though. This one speaks to something near universal: the combination of awe and discomfort of new technologies, which we manage by calling it “magic”. As true for Luddites and for senior engineers (they’re people too!):
With my marketing hat on, I’ll admit that (beyond empathy for people’s unsettled awe), whenever I see the word “magic” and AI, this little guy pops into my head:
Using the “magic” metaphor is just a tiny, little bit lazy. I understand why people reach for it. Explaining something new without being boring is flamboyantly difficult. For now, “magic” often works as a marketing device.
Metaphors are the Freudian slips of technology
But what if we take the metaphor of AI magic seriously? Can we use the (European) history of magic to help us think about our relationship with magic? Can doing so help us get beyond a polarised binary of “AI is good” or “AI is bad”?
What did we mean by “magic” in the past? I grabbed this from a text from the 16th century, cited in The Magus by Anthony Grafton:
For 16th century people who believed in magic, tapping into that power and shaping it meant giving them access to truly extraordinary powers. It amused me to find direct echoes in DeepMind-founder Demis Hassabis’ TED talk from April 2024:
A theorist I like a lot, Silvia Federici, describes magic as “a variety of practices designed to appropriate the secrets of nature and bend its power to the human” (from Caliban and the Witch).
Federici outlines the sorts of problems that people in the past tried solve with magic. A lot of them will be very, very familiar. Federici says that: There was magic designed to win games…
…to play unknown instruments…
…to gain immunity in war…
…and to get babies to sleep.
Often, it seems, we’re pointing AI at the same problems that saw our 16th century forebears turning to magic.
Magic is a symptom, not a cause
We tend to think that new technologies cause change. Looking to the history and practices of magic helps us think beyond this. While this a short talk, and I focus on European magic, globally, magic and witchcraft have been associated with times of societal upheaval and change and inequality.
For Federici, magic is a symptom, not the cause, of the relationships between labor and production changing.
Sound familiar? The people who made magic in Renaissance Europe also map pretty neatly onto today’s AI practitioners.
Why is magic controversial?
According to Federici, magic’s controversy stems from promising something (desirable) from nothing.
Beyond necessary debates around bias and copyright, I think that part of the controversy around AI is the promise of results without work. Either as an individual, or as a company. Who hasn’t heard “Just put an AI on it”?
I’m intrigued by the idea that court magicians would promise knowledge and riches to the prince or to the courts, as long as they had access to all the books, supplies (or compute power) that they need.
Often, having a court sorcerer seems to have been more important to Medieval rulers than anything the magi would actually achieve.
I think we see something similar in Silicon Valley, and elsewhere. Just look at this year’s reverse acquisitions of AI teams that haven’t found product-market fit.
Another similarity between AI and the magi is that they’re always at risk of falling afoul of other, traditional sources of power
The history of European witchcraft, for Federici, is a reaction to the commons being split up and privatised.
We see a lot of these tensions echoed in the relationship between large AI companies and the Internet as a Commons. Walled gardens, witch burnings.
Taking a step back about what it means to say something is “magic”: in Medieval Europe, magic placed activities in the realm of the stars, where worldly cause and effect didn’t apply. There’s echoes of this in both AI hype and AI doom: “AI is going to change everything,” or “There’s nothing we can fix or do.”
But I there’s also a lot of hope and power in the similarities between AI and magic, too.
The magic of medieval courts morphs into other ways of knowing. We can see the roots of modern medicine, chemistry and physics in the practices of court magicians — all of which, ironically, would have struck magicians themselves as “magical”.
If the theme of today at The Conference is “now what?”, how can we use these historical parallels? As a practitioner, how do I use metaphors and historical parallels in my work with AI?
Personally, I find the history of ideas reassuring: they show us that we are not alone. Technological changes might be enormous in scale and ambiguous in impact, but they’re not fully unprecedented. Other people, far in the past, in different parts of the world, have thought about similar problems and challenges. They’ve managed, to find new ways of functioning. (Other than if they’re a being burned at the stake for witchcraft. Avoid.)
Metaphors and historical parallels also give us archetypes for who we might want to be in a new age. One of the magis, doing elite magic with vast amounts of funds from centralised sources of power? Or parts of communities, even knowing what the risks are?
Finally, these comparisons show us that designing for ambivalent and uncertain outcomes, as you see in GenAI systems, is not new.
It’s an ambivalence that we’ve lived with in the past. It’s an ambivalence that we’ve known and understood as cultures. We can deal with them in interfaces and technology systems, too.
Or, as they said in magic books in the 16th century:
Slides designed by Mark Hurrell, with images from Public.Work unless it says otherwise.
Other references:
Behringer, Wolfgang. “Witchcraft Persecutions in Bavaria: Popular Magic, Religious Zealotry and Reason of State in Early Modern Europe.” Cambridge University Press, 1997.
Comaroff, Jean, and John L. Comaroff. “Modernity and Its Malcontents: Ritual and Power in Postcolonial Africa.” University of Chicago Press, 1993.
West, Paige. “From Modern Production to Imagined Primitive: The Social World of Coffee from Papua New Guinea.” Duke University Press, 2012.
Taussig, Michael. “The Devil and Commodity Fetishism in South America.” University of North Carolina Press, 1980.
Gershman, Boris. 2016. “Witchcraft beliefs and the erosion of social capital: Evidence from Sub-Saharan Africa and beyond.” Journal of Development Economics 120: 182–208.
