When researching a blog post on the frontier metaphor in AI discourse, I came across an article from 1980 that I had not seen before. It gave me some good ideas about how to deal with the frontier metaphor, but it also provided some forgotten insights into the nature of metaphor that had parallels with the much better known 1980 book by George Lakoff and Mark Johnson, Metaphors We Live By.
In this post I try to show how the article’s approach to metaphor overlaps to some extent with that put forward in the book but also adds some different perspectives. I will first say why I think the 1980 book is important, then why the forgotten 1980 article is also important.
This post is as much about the nature of metaphor (and why knowing about it is important in the age of AI) as it is about the history of metaphor analysis (which goes back a long way).
Conceptual metaphors
When people study metaphors nowadays, they mostly base their research and definitions of metaphor on the tradition of metaphor analysis started by Lakoff and Johnson. Although metaphors have been studied from a rhetorical, philosophical, philological and psychological perspective before, this book opened my eyes to what one may call hidden metaphors: metaphors that we use every day but that go completely unnoticed.
Imagine I said “I can’t process all this stuff about AI anymore”. You would understand what I mean. Why? Because the word ‘process’ that I use in my utterance is linked to a whole network or pattern of metaphors that can be condensed into what Lakoff and Johnson called a ‘conceptual metaphor’, in this case THE MIND IS A COMPUTER or similar, where aspects of computers are mapped onto aspects of mind.
Conceptual metaphor theory gave us great insights into how mind and language work together to shape thinking, speaking and acting in the world. It made clear that metaphors make the world, including, nowadays, GenAI and LLMs, “thinkable, debatable, and socially meaningful”.
Imagine I said “ChatGPT is my constant companion”. Again, you would understand what I mean partly because there is a conceptual metaphor at work, namely “AI systems are humans”, a pervasive framing underlying all anthropomorphisation of AI. But what actually goes on when I say this and somebody else hears this? To gain some further insights into this process, I now turn to the old(ish) article.
Metaphors as invitations
The article, published in The Professional Geographer of all things, is by D. N. Livingstone & R. T. Harrison and entitled “The frontier: Metaphor, myth, and model”. It uses insights derived from sociology, mid-twentieth century metaphor theory, mainly Max Black, but also from a 1962 article by Douglas Berggren on the ‘uses and abuses of metaphor’ published in, wait for it, the Review of Metaphysics. As I noticed in the past, when working on the history of 19th-century semantics and pragmatics, one can find hidden gems about metaphor in all sorts of periods and places.
The first thing Livingstone and Harrison point out is: Metaphors are invitations. In our case they are an invitation to see AI systems such as ChatGPT as companions. Livingstone and Harrison themselves use a metaphor that goes in the opposite direction (not humans→machines but machines→humans) and they point out: “The assertion ‘man is machine,’ for example, invites the reader to consider those characteristics which might be analogues to certain well-known machine traits”.
Metaphors as lenses
Once readers or hearers have accepted the invitation, they engage in a “highly selective transference of familiar connotations of a word to a new context”; for example from human capacities (thinking, learning, reasoning, supporting and caring) to AI. This transference or, what Lakoff and Johnson call mapping, then “acts as a kind of screen or lens through which the new subject is viewed”. When AI is seen as human, we forget about the humans who build it.
The lens metaphor for how a metaphor works was explored in the Berggren article referenced by Livingstone and Harrison. It’s worth pointing out that the metaphor for metaphor as a lens or a pair of lenses, a filter or a screen was very much in vogue between the 1930s and 1960s and beyond, in literary criticism (I. A. Richards), psychology (Karl Bühler) and philosophy (Max Black), even in the study of the classics (W. B. Standford). The lens metaphor has also been used by the historian of science Andrew Reynolds in the title for his 2018 book on the history of cell biology, where metaphor becomes a ‘third lens’ – over and above the eye and the microscope.
Metaphors and new meanings
As Livingstone and Harrison point out, through the lens of a metaphor and the mappings it invites, a topic, issue or phenomenon can now be “seen in a new way and new attitudes are evoked”. New meanings and ways of seeing emerge from a “superimposition, overlap, or combination between ‘the expected and the experienced’… and the unknown and the comprehensible”. And new ways of seeing also shape new attitudes and actions on the world.
Lakoff and Johnson would say: we map what we know about something more concrete onto something more abstract that we don’t know so much about. So, when we say “AI is asbestos”, we are invited to map what we know about asbestos (toxic, difficult to remove etc.) onto our not yet quite complete understanding of risks posed by AI. The metaphor provides insights into possible dangers posed by AI systems, short-term and long-term. This might invite certain attitudes and action towards AI.
Metaphors highlight and hide
But this mapping process is not always smooth and uncomplicated. When we transfer connotations or features from one context to another, some “features are ignored or suppressed while others are emphasized or distinctively organized” (so when I call ChatGPT a companion, I disregard the fact that I can’t hug it and that it is a commercial product). Lakoff and Johnson call this highlighting and hiding: a metaphor foregrounds some features of its target and pushes others into the background. When we say “AI is a revolution”, we highlight novelty, progress, speed and change for the better, but we hide the slow and painstaking human labour that goes into developing AI systems.
Metaphors and plurality
Furthermore, what is achieved through having been invited to engage with a metaphor is not a single meaning. Instead, metaphors invite “the exploration of an unspecified number of potentialities”. They are an invitation to discover further similarities, but not everybody discovers the same ones. When I say “AI is a black box” some might think that’s bad as we don’t know what’s going on inside it, as it’s obscure, mysterious and opaque, while others might argue that it’s actually math not magic and see it as an invitation to open it up, look inside and study its innermost workings. The invitation to explore leads to different outcomes.
Metaphors and reduction
However, when using metaphors, we also tend to reduce “the unintelligible – randomness – to the intelligible – a pattern”. A metaphor’s “inherent open-endedness and invitation for imaginative speculation tends to be reduced”. A metaphor like “black box” can invite opening it up, but once it hardens into a model, only the meaning of ‘we can’t know’ survives. In Livingstone and Harrison’s terms, the metaphorical pattern becomes a ‘model’ of reality.
There is a danger though. Metaphors as models can be “useful explanatory schema”, but “all models, like religious symbols, are idolatrous unless they suggest their own inadequacy.”
This means that if we fall under the spell of a metaphor, for example framing AI systems as an unstoppable force of nature, there is the danger that “the model is mistaken for the reality it represents” and this “results in a reductionism”. That is the moment when it’s time to resist metaphors and to ask: who creates that metaphor, what is the purpose of that metaphor and what does the metaphor make us believe and do? The metaphor of AI as an unstoppable force of nature frames AI as inevitable, for example, and invites inaction rather than regulation or even resistance.
Reductive metaphors can become tools of power. As James Bono put it in 2001: “The work of metaphor is not so much to represent features of the world, as to invite us to act upon the world as if it were configured in a specific way like that of some already known entity or process”. Metaphors structure how we think and how we act. The study of metaphors allows us to reveal the structure of thought and action. (Thanks to Kate Roach for that sentence!)
And the lesson is…
When we are invited to explore similarities (and differences) through the lens of a metaphor, we should be open to the plurality of interpretations that might emerge, but we should also be aware of the dangers that lurk when reducing a metaphor to one dominant point of view. Metaphors both invite and constrain.
Metaphors invite us to see something as something else, but we should always ask what a metaphor makes easy to see and what it makes easy to miss!
Image: Close up of round reading glasses set down on an open book. Sepia colour. Photo by Büşra Akkaya on Pexels.com
Acknowledgement: Thanks to Cinzia Pusceddu, associated director of We and AI, for stimulating discussions about AI and metaphor.
Leave a Reply