The shock of the new
A significant - and very much of-the-moment - element of the UNTHNKBL Academy programme for marketing leadership will naturally be dedicated to our future relationship with new and emerging technologies. Marketers are coming to a clear and important fork in the road, and no doubt our understanding and interpretation of the opportunities and threats of AI will continue to loom larger.
Now is the time for us to be crystal clear in our ways of thinking and talking about this hot and highly polarising subject. To achieve this, we need first to look very closely at the common language of AI. How, day-to-day, are we thinking and talking about it?
In their ground-breaking and still-critical work, Metaphors We Live By, George Lakoff and Mark Johnson make the compelling case that metaphors are far more than mere whimsical linguistic tools for simplification or elaboration. They are, in fact, intrinsic and deeply-wired daily behaviours, by which human beings receive, make sense of and share their worlds and their lives.
Metaphors are both subtle and inescapable. We can think of them as lines of mental code which offer more or less universally accepted and understood navigational shortcuts for living.
Faced with the challenges of dealing with the shock of the new, therefore, we turn, instinctively and unconsciously, to metaphors. New technologies -combining, as they tend to, the entirely alien and the utterly familiar - consistently rely on metaphorical description to convey their meaning and their value. Going back for a single example to early days, it was, as so often, at Xerox Parc that the phrase 'the user illusion' originated: "The illusion created for the user by a human-computer interface, for example the visual metaphor of a desktop used in many graphical user interfaces".
To given a sense of the looming depth that underpins this ostensibly common sense concept, plenty of big thinkers - most famously Bishop George Berkeley in the early 18th century - have argued that our very human consciousness is itself an illusion. Suggesting, in other words, that consciousness is a crucial metaphor for making sense of human experience.
We are, it turns out, in deeper and far more interesting waters than expected. Not least because this very subject of consciousness - or the lack of it - will continue to recur in the AI debate.
Metaphor as friend and foe
The power - both creative and destructive - of metaphors resides in their ability to do far more than merely overlay a convenient and accessible explanation on top of a complicated idea or situation. They are literally 'how we think' about the bulk of our day-to-day experiences. They inform and guide value judgements, evaluations, decisions, commitments and actions.
For this reason - and given in particular their newfound networked ability to cross the entire world in seconds in memetic form - the cultural reference points, associated imagery and vocabularies that make up our dominant metaphors demand, today, unprecedented care and vigilance. A well-chosen phrase or image can make the world a better place. Ill-chosen - even with the best possible intentions (let's not forget the paving of the road to hell) - the damage can be swift, brutal and very hard to unravel.
The language of AI
Which brings us to the very hot discourse - triggered (there's a metaphor right there, in passing) by the explosive release (another one there) of GPT-4. Convenient, indeed highly desirable, though it may be, we can't dismiss this latest wave of development in AI in the same way that crypto and metaverse have recently, probably deservedly, found their way to the bottom of the news cycle. AI, and this new breed of chatbots and enhanced search capabilities, really matter a lot, and will continue to do so.
In the current rush and flurry of updates, discussion, observation, claims and opinions, it's unsurprising that metaphors are playing a central role. For the urgent reasons noted above, their vocabulary, references and imagery demand a very careful, forensic, scrutiny. They will frame, determine and steer our understanding of AI. They will, it's not too extreme to state, control how we get to grips with its meaning and value, its potential positives and negatives, and its plethora of opportunities and threats.
The term 'artificial intelligence' itself conceals rich and powerful metaphor. Putting aside that, rather disingenuously, the adjective 'artificial' subtly emphasises, rather than minimises, this form of 'intelligence', the leading experts are consistently careful to point out that this 'intelligence' should be understood as somewhat 'alien'. In other words, AI does not think the way humans do.
And yet, we can't seem to help ourselves. We insist on humanising (harking back perhaps to the earlier point about the user illusion) our chatbots. And the way we do this is, of course, with metaphor. We talk and write about 'what GPT-4 can do' in irresistible lockstep with what we, as people, can do.
Harmless enough, we could say. Why not? And as long as we're playing with GPT-4 in the same spirit in which, say, we might play a mobile game or scroll up and and down our social feeds, I must agree. But there's a lot more going on here, and a lot more at stake.
Simulacra of life
The New York Times is this week running On Tech: AI for subscribers, "a pop-up newsletter that will teach you about artificial intelligence, especially the new breed of chatbots like ChatGPT". The authors, Kevin Roose and Cade Metz, are both very credible technology analysts and commentators, and I've found it a handy resource (not least to help ensure that I avoid at least most of the inevitable AI newbie errors). With no critical intent or disrespect to Messrs Roose and Metz, I'd like (see below) to use their excellent work to pick out and point up some examples of the impacts and implications of the metaphorical language that is fast accumulating and crystallising around the discourse.
What emerges immediately and clearly is that more or less all of these dominant metaphors are based on 'humanising' the visible behaviours and effects of AI. The machine is thereby bestowed with our own impressions of intelligence, of consciousness, of 'mind', of relationship, of creativity and of communication skills. This phenomenon is far from new, by the way. Between 1964 and 1966 at MIT, Joseph Weizenbaum's ELIZA repeatedly shocked its creator when its human users, responding to the language that the program used in their exchanges with it (her?), consistently attributed human-like feelings to what had been intended as a simple method to explore communication between humans and machines.
We are, it seems, predisposed to the assumption of human emotion and meaning in any system that can assemble respectable syntax and perform plausible dialogue. It's also worth noting in passing that, if meaning at its root is a sense of connection, then this impression of shared meaning in turn feeds a feeling of connectedness with such machines. If there is such a relationship, it is, of course, entirely one-sided.
Our language, in other words, is doing far more than just making the capabilities of AI accessible to the lay person. It's framing the way we think about and how we interpret the behaviours of AI. Perhaps most critically, it's subtly moving us to attribute qualities to the machine that do not - at least not yet - exist at all. At the easy-to-use and genuinely friendly interfaces offered by, for example, ChatGPT, we are encountering highly sophisticated simulacra - representations or imitations of life. To pick a now-ubiquitous example of mainstream AI application, there is no 'self' in the 'self-driving car'!
From metaphors to simulacra
Let's close with a few deconstructions (borrowing from the NYT as noted above) to bring a little clarification to how we think and talk about AI, by contrasting the metaphors and language that are most commonly in use in and around the topic, with their respective simulacra.
| Common AI Metaphors | Deconstructed as Simulacra |
| "By analyzing thousands of cat photos, for instance, it can learn to recognize a cat." | What is recognised is not 'a cat', but in fact a mathematically probable resemblance to millions of cat images, and perhaps also cat movement. |
| "Neural networks enable Siri and Alexa to understand what you’re saying." | As with the cats above, the machines do not 'understand' here. They perform a massive mathematical triangulation to create and respond to a hypothesis of what you might want. |
| " ... these systems learned to write unique prose and computer code and carry on sophisticated conversations." | While with generative AI, chatbots are learning to perform these tasks at both remarkable speed and scale, we would not normally think of 'writing' or 'conversation' as being merely the outputs of plausible, even meaningful, one-way or two-way content. |
| "... it might learn that “bank” can refer to a financial institution or the side of a river, depending on the surrounding words." | The machine is certainly hunting and learning patterns, again at extraordinary speed and scale, but it does not - not yet - 'know' what a financial institution or the side of a river are. |
| "Weirdly, it may develop other skills, too." | The ability of Large Language Models to spontaneously acquire unplanned capabilities is fascinating and important. But to refer, as here, to these as 'skills' brings hidden associations that lead to problematic assumptions. |
To conclude, and at the risk of seeming even more pedantic, I should point out that whenever we find ourselves with a plethora of use cases on one side of the equation, and a paucity of value propositions on the other - and this is clearly where we are now with AI - there is a need to pause for deep reflection.
The ancient discipline of philosophy - whose unique job it is, when we are pushed back from 'What To Do?', via 'What To Think?' to the empty and bewildering space of 'How, In Fact, To Think?', has much to offer us, before we rush headlong into a future that will definitely feature the dystopian alongside the utopian.