When the Map Starts Talking Back: AI, Human Cognition, and the Limits of Metaphor

Inspired by the recent paper “Artificial Intelligence, Conceptual Metaphors, and Conceptual Engineering: Are AI-Based Framings of Human Behavior and Cognition Successful?” by Jan Thomas Mollema and Thomas Wachter.

The Machine in the Mirror

For centuries, humans have explained themselves through the most sophisticated technologies available to them. When clocks represented the pinnacle of engineering, the universe became a clockwork mechanism. When steam engines transformed industry, the body became a thermodynamic machine. And when computers emerged, cognition became information processing.

Now, in the age of artificial intelligence, we increasingly describe ourselves through the language of machine learning, neural networks, optimization, prediction, and computation. Memory became storage while learning slightly drifted into training. In the AI domain, language becomes prediction and decision-making is becoming a mere optimization.

The bottom line here is that human cognition is being increasingly framed through AI concepts. Recent philosophical work by Jan Thomas Mollema and Thomas Wachter asks an important question:

Are these AI-based framings actually successful?

Their answer is nuanced. Such framings can be useful, but they also carry risks. We may begin confusing the map with the territory, mistaking our models for the phenomena they attempt to describe. The warning is important.

But it may no longer be sufficient.

The Problem Is Not the Metaphor

The authors distinguish between two possibilities where the first is that AI concepts function as conceptual metaphors. They help us understand aspects of human cognition by highlighting structural similarities between humans and machines. The second is that AI concepts function as conceptual engineering. Rather than merely describing reality, they actively reshape the concepts we use to think about ourselves.

Both perspectives are valuable, yet both assume something increasingly questionable, and that is postitioning of AI remains outside the cognitive process. As if humans create concepts, and AI merely supplies metaphors. But is that still true?

When the Tool Becomes Part of the Environment

Historically, our tools remained external. Imagine how a microscope changed what we could observe. And then the telescope changed what we could see. A simple calculator changed what we could compute.

But AI is not just another tool. It is fundamentally different from other tools we created. AI participates in language. It provides interpretation. The machine strongly contributes in meaning construction.

Every day millions of people use AI systems not merely to retrieve information but to organize thoughts, generate explanations, explore ideas, write texts, create plans, and structure decisions.

This relationship is no longer simply human observing machine, as it is more about human and machine co-producing meaning.

This changes the question. Instead of asking:

“Does AI describe humans accurately?”

we may need to ask:

“What happens when AI becomes part of the cognitive ecology through which humans understand themselves?”

Similar outputs, different processes.

One reason this question matters is that humans and AI often produce remarkably similar outputs. Both can write, both can explain, both can summarize. Both can generate stories an both can answer questions.

Yet similar outputs do not imply identical processes, because human cognition emerges through embodiment, physiology, emotion, development, social interaction, and lived experience. AI cognition emerges through statistical pattern formation across vast computational architectures.

The observable result may sometimes appear similar, while the underlying mechanisms remain profoundly different. Confusing output similarity with process similarity creates a new version of the classic map-territory problem.

The danger is not that AI becomes too human to be distinguished, the real danger is that humans begin simplifying themselves into whatever aspect of themselves AI can reproduce.

Beyond Metaphor: Conceptual Transduction

The debate between metaphor and conceptual engineering is valuable, and perhaps incomplete. A third possibility deserves consideration. Concepts may function as transduction mechanisms.

A metaphor does not simply compare two domains, because it also transfers organizational structure from one domain into another. Once introduced into a cognitive system, a metaphor changes future perception.

It influences what becomes visible, what becomes meaningful and what becomes thinkable. In this sense, AI concepts do not merely describe cognition, but they participate in shaping cognition.

Every time we describe memory as storage, thinking as computation, or emotion as information processing, we subtly reorganize the conceptual landscape through which future understanding emerges.

The concept itself becomes an active force.

The Question Behind the Question

This leads to a deeper issue. When we ask whether an AI-based framing is successful, we must first ask:

Successful at what?

Prediction? Control? Scientific explanation? Adaptation? Human flourishing? Meaning-making? Innovation? These goals are not identical.

A concept can be highly predictive while simultaneously narrowing our understanding. A concept can be scientifically useful while reducing experiential richness. A concept can improve efficiency while degrading wisdom.

The evaluation of a concept therefore cannot depend solely on accuracy. It must also consider consequences. What forms of understanding become possible? What forms become impossible? What possibilities are expanded? What possibilities are constrained?

When the Map Starts Talking Back

Mollema and Wachter correctly remind us that maps are not territories, yet a new phenomenon may be emerging.

For the first time in history, our maps are becoming active participants in the territories they describe. Artificial intelligence is no longer simply generating descriptions of human cognition. It increasingly participates in the processes through which those descriptions are produced, shared, refined, and adopted.

Metaphors themselves are not the problem. In fact, they are among humanity’s most powerful cognitive tools. Every scientific revolution has been accompanied by new metaphors. Metaphors allow us to transfer structure from the known into the unknown. They help us navigate complexity before we fully understand it.

The problem begins when metaphors cease to function as bridges and begin functioning as foundations. A metaphor is useful when it points beyond itself and serve as connecting joint. It becomes dangerous when it starts carrying the weight of reality.

We know that the brain is not literally a computer. We know that memory is not literally storage. We know that language is not literally prediction. Yet these metaphors are often so successful that they quietly transform into assumptions. What began as a way of thinking becomes what we think.

Meaning is not discovered exclusively through concepts; it is also concealed by them. Every concept illuminates a part of reality while casting a shadow over another. The more successful a concept becomes, the harder it becomes to see what it excludes and this is the game of dominion.

The situation becomes even more complex in this internal dominion game when metaphors accumulate upon metaphors.

Modern AI itself emerged through metaphors borrowed from human cognition. Neural networks, memory, attention, learning, reasoning, and intelligence were originally attempts to describe human capacities. We then built computational systems using these concepts and now increasingly use those systems to explain ourselves again.

Human → Machine → AI → Human.

At each step something is gained and at the same time at each step something is lost. The risk is not error in the traditional sense, but it is the abstraction drift. Each layer moves us slightly further from direct contact with the phenomenon itself while simultaneously creating an increasingly convincing explanation of it.

In an age of unprecedented information production, perhaps the greatest challenge is not misinformation but over-metaphorization. The hyperproduction of meaning creates endless explanatory layers that can obscure the very realities they were intended to illuminate.

The challenge is therefore not to abandon metaphors, because that would be literally impossible. Therefore the challenge seems to be to maintain awareness of where the metaphor ends and where the phenomenon begins.

The truth is that Artificial Intelligence may ultimately teach us something profound about ourselves, but only if we resist the temptation to reduce ourselves to the metaphors through which we currently understand it. The future question may therefore not be whether AI is an adequate metaphor for human cognition. It inquiry should be more in how human and artificial cognition together transform the conceptual landscapes through which meaning emerges.

The map is no longer silent.

The map has started talking back.

And in a world increasingly saturated with explanations, the greatest danger may not be the metaphor itself, but the metaphor built upon another metaphor, and another metaphor, and another, until the weight of abstraction becomes indistinguishable from reality. ‘Fake it till you make’, the assumption that reality conforms to sufficiently repeated descriptions of reality, it here does not apply well.

Because metaphors are valuable precisely when they remain metaphors. Once they become unquestioned foundations, they stop illuminating reality and begin replacing it. The challenge of the AI age may therefore be surprisingly simple:

…to remember that every map is drawn from somewhere, every concept leaves something out, and every explanation eventually requires explanation itself.

Only then can we benefit from our metaphors without becoming trapped inside them.

Author’s Note

The discussion initiated by Mollema and Wachter highlights a crucial challenge for the coming decades. The question is no longer merely how we use AI, but how AI changes the conceptual environments through which we understand ourselves. If the twentieth century was the age of information, the twenty-first may become the age of meaning production. In such a world, the ability to distinguish phenomena from the metaphors used to describe them may become one of the most important cognitive skills of all.