Use the Right Metaphors to Explain AI

We often explain Artificial Intelligence through metaphors – a ‘brain’, a ‘colleague’, a ‘black box’ or the worst of the lot, ‘magic’. These often dictate the use – and misuse – of the technology. They distinguish AI as smarter, more mysterious, or more autonomous than it can be. The problem? All these metaphors are unsustainable. And leaning on them we run the danger of constructing policies, businesses and expectations on quicksand.

Most Metaphors Miss the Point

The metaphor of the ‘brain’ is alluring. Neural networks were in fact inspired by the form and structure of neurons in the brain. But that similarity is only surface-level. AIs do not think, reason or remember like humans. They’re simply excellent at seeing statistical patterns and repurposing them.

So once we refer to AI as a brain, we overpower it. We are inclined to see its output as a priori perceptive, as though we were peeking over the edge of a second cognition. The reality is that it is more of a very fast parrot, in that it produces answers that appear human, where in fact it does not really understand. The danger of the metaphor is that it negates human responsibility. When the brain says it, who is to argue with the brain?

Chatbots make the metaphor even stickier. A system that is able to engage in a conversation, answer questions or make a joke almost feels like a person. But anthropomorphism is a fallacy. It deceives us of the boundaries of the system.

Take into consideration the way people speak of AI assistants as helpful ‘colleagues’. The same framing promotes trust, to the extent of dependence. The analogy conceals, however, that AI does not have judgment, context, or lived experience – all aspects that make you human. One of your colleagues may confront you, oppose or raise ethical issues. A chatbot cannot. In referring to AI as human-like, we allow the instrument to masquerade as things it isn’t and in that process, undermine our vigilance.

Referring to AI as a ‘black box’, on the other hand, captures a very real frustration: even experts cannot always describe the general way complex models come to their conclusions. The metaphor correctly cautions us against being opaque. But it also breeds fatalism. A box that one is not supposed to look into implies it is pointless to examine, and cannot be explained.

In fact, AI systems can frequently be probed, audited and stress-tested – not perfectly, but meaningfully. Overrelying on the black-box metaphor, we chance shrugging our shoulders and giving up oversight at a time when sharp improvements in the way of interpretation, transparency and accountability are what is required.

Then there is the most ancient (and arguably, worst) metaphor: ‘magic’. When an output from a Generative AI bot surprises us – a poem, a photographic picture, a market report, we say – wizardry! The more magical it appears the less we are likely to question it.

Magic defies explanation. It is supposed to be glitzy, not subject to analysis. Similarly, magical language surrounding AI encourages people to avoid posing questions about how AI works, which data it has been trained on, and what its blind spots are. That deletion of mechanism preconditions receptive conditions to hype cycles, regulatory panic and bad investment decisions. When AI is magic, we can only sit there and watch in wonder.

A Better Mental Model: AI as an Imperfect Mirror

So when brains and humans and magic fail us, what would symbolize AI better? A possible alternative, which is promising, is the mirror. AI is based on its training data. It blows some things out of proportion, overlooks others and sometimes even distorts reflections completely. However, at the bottom of it all we are being shown patterns that we have already entrenched in the world.

There are merits to the mirror metaphor. It underlines that AI is also derivative and not generative. It reminds us that training data bias will reflect in results. More to the point, it puts human judgment in the middle-seat. Mirrors do not analyze, they reflect. We must be the interpreters.

The use of such metaphors is not merely linguistic flourishes – how we explain and define AI will shape how much dependence we can place in it in key aspects like policymaking, governance and everyday applications. A policymaker that regards AI as a brain may argue for rights and personhood. When a manager regards AI as a partner, they can delegate excessive decision-making. An investor who sees AI as magic can pour capital down the drain.

In treating AI as an imperfect mirror, the duties are more apparent: filter the data, make sense of the reflection and be suspicious of distortions. Such an attitude grounds strategy in realism as opposed to hype.

AI does not have to be magic or human. And it is potent enough as a mirror – one that displays patterns at speed and scale, but never without a critical human eye. The more we can get out of the metaphor trap, the more wisely we will use the tools we are building.

#AI #ArtificialIntelligence #CriticalThinking #MachineLearning #ML #Bias #Strategy #Data