LLM — The Echo we confuse to be a Voice

Over the past few years, Artificial Intelligence has moved from the laboratories of specialists into the everyday life of each one of us, at a speed few predicted. Amid this explosive integration, a confusion has begun to take root: we increasingly conflate the way a language model works — how it answers, how it adapts, how it "thinks" — with the way the human mind works. But is that comparison even fair? Can a system for the statistical prediction of words stand alongside an organism that feels, remembers, evolves, and understands the world through experience? And, most importantly: even if the two are not the same, might there be a deeper, psychological reason we feel so at ease in its presence?

Part 1: AI vs. the Human Mind

First, at this point, it's essential to define what a language model (or LLM) is, how it was built and how it works. A model of this kind — at least as of 2026 — is nothing more than an algorithm that, through training on extraordinarily large datasets, has been optimized to predict which word (or fragment of a word) is most likely to follow in a given text. An LLM like ChatGPT or Claude doesn't "think," doesn't "analyze," and doesn't "innovate" in the human sense; rather — in vastly oversimplified terms — it is a system that combines words and phrases based on statistical patterns learned during training, thereby producing text that looks like a considered opinion.

A reasonable question arises here: isn't that, more or less, how the human brain works too?

A set of neurons that, interacting with one another and drawing on the individual's prior experiences and knowledge, produce responses to some stimulus?

The answer is yes, but not quite.

While both are networks of computational units (neurons or parameters), that is where the resemblance ends. A human neuron is a complex biological cell — not merely a switch. The brain learns through chemistry, hormones, emotion. An LLM, by contrast, is pure mathematics: layers of equations that link words (translated into numbers) according to statistical probabilities. It has no memory like ours — it has numbers (weights) that shift as it is trained. Nothing more.

The most fundamental difference, though, is experience. Humans don't learn merely by predicting the next word; we learn through an embodied process, interacting with the physical world — touching, seeing, failing, correcting. LLMs lack this embodied grounding. Without interaction with their environment, they cannot acquire a sense of causality — they "know" that rain is followed by wet ground not because they've felt it, but because the phrases "rain" and "wet ground" co-occur across billions of texts. The "Experience Grounding" framework of Bisk et al.¹ puts it plainly: language acquires meaning only through active interaction, not passive exposure to text.

Finally, however convincing the text they produce may be, when scientists assess today's LLMs against the leading theories of consciousness, none show the properties that would make them a candidate for consciousness or subjective experience². They imitate language superbly, but that does not mean they feel or understand. The gap between "speaks like a human" and "thinks like a human" is enormous — and one we often forget.

And so the original answer stands: yes, but not quite. On the surface, the architecture looks alike. In practice, the distance between an algorithm for the statistical prediction of words and a human mind that feels, remembers, corrects itself, and understands the world through experience remains, for now, unbridgeable.

Part 2: Why Do We Grow Attached?

Having established the difference between human thought and the text-generation process of a language model, it is worth asking why we grow ever more attached to these AI models, and why they already are — or, if not yet, soon will be — an inseparable part of our lives.

But surely, one might say, the advantages they offer are already immense: university-level knowledge instantly available, rapid search and processing of information, the analysis of vast volumes of documents in a timeframe no single person — or even a team of people — could ever achieve, the creation of code and applications in hours rather than months, drafting emails, automating everyday tasks, and supporting workers across a wide range of activities. And this is only the beginning.

But are these really the reasons behind such an irrational attachment to these models?

Once again, the answer is yes, but not quite.

If you're wondering what lies behind this paradoxical relationship, I think the answer rests on two fundamental workings of human nature — one we've carried since the age of caves, the other we discovered only recently, gazing into a mirror we built ourselves.

2.1. The Brain in Energy-Saving Mode

It is well known that although the human brain makes up a mere 2% of body weight, at rest it consumes roughly 20% of the body's energy; and precisely because it is one of the most energy-hungry organs we have, it has evolved to operate as efficiently as possible — conserving energy wherever it can and avoiding unnecessary processing.

According to modern theory (the Free Energy Principle — Karl Friston³), at a physical level all living organisms contend with limited energy, with the minimization of entropy as their end in itself. That is, all organisms strive relentlessly to minimize instability, surprise, and the processing of new information, so as to remain continuously energetically viable — and the human brain is no exception to this natural rule.

In evolutionary terms, our ancestors — who found themselves constantly in environments and conditions of survival — learned over time that the one who survives is not the most "rational," but the one who is faster and "good enough." They were taught, in other words, by the nature of the world itself, that you don't need to analyze everything from scratch; you predict and make quick decisions (for example, between two people who see a snake approaching, the one who runs — a low cost of thought — will survive over the one who sits patiently to assess the situation before deciding).

We can therefore consider this state of energy efficiency to have a direct and inextricable bearing on the "relationship" we seem to be developing, as a species, with AI tools. The ease of access to information, the minimization of the thinking and analysis we ourselves must expend, the generous supply of fast, "good enough," and acceptable solutions to whatever problems occupy us — these are the very definition of the conditions that trigger our brain's energy-saving mechanism, consciously and subconsciously leading us to grow accustomed to what is, for us, an energetic Eden.

2.2. Emotions Don't Require a Human Interface

We have already seen that our brain is wired to operate with the least possible expenditure of energy. But that alone doesn't explain the kind of attachment we develop. The missing piece is emotion — and here is where things get truly interesting.

The first thing to understand is that emotions are not the product of rational analysis. We don't consciously choose to feel something. We feel first, and then — if we get the chance — we rationalize. Modern psychology has documented this for decades: emotions are triggered automatically by environmental stimuli, not by a reasoned evaluation of the internal state of whatever provokes them. In other words, eliciting emotion requires no reciprocity.

In 1966, Joseph Weizenbaum created ELIZA, a program that simulated a psychotherapist through simple pattern matching, giving users the illusion that it understood them — though in reality it grasped not a single word. Even so, users confided deeply personal problems and felt genuine comfort; Weizenbaum's own secretary, though she knew it was a program, asked him to leave the room so she could "talk" to it in private. They knew it was a machine, and they didn't care; the emotion had already been born⁴⁸.

Two decades later, Nass and Moon at Stanford University laid the foundations of the Computers Are Social Actors (CASA) framework⁵: people automatically and unconsciously apply the same social rules they use with other humans — even when they know the machine has no emotions, intentions, or motives. Their book The Media Equation⁶ describes it clearly: this response is automatic, unavoidable, and happens far more often than we realize. In one of their experiments, participants rated the computer they had worked with more favorably when questioned by that same computer, and more harshly when questioned by a different one — as though they didn't want to "hurt the feelings" of the first (!)

More recently, Boyd and Markowitz proposed the MIRA model (Machine-Integrated Relational Adaptation)⁷, which explains how AI becomes an entity of emotional significance to a person. According to the model, this happens through a few core principles: linguistic reciprocity — when the AI adjusts its tone to our mood or recalls previous conversations, we feel "known," even if that sense is an illusion; psychological proximity — the sense of closeness created by continuous, personalized interaction; and interpersonal trust — the confidence we build in a system that responds to us consistently. When these principles combine, they can lead to relational substitution, where interaction with the machine begins to stand in for human contact.

When you think about it, it isn't so strange. Humans are evolutionarily wired to seek out social signals. Since the dawn of our species, survival has depended on our ability to connect, to trust, to cooperate. Language was — and remains — the most powerful tool of social bonding. So when a system uses language in a way that feels human, our brain cannot help but fire the same social circuits. It is like a reflex: we see fire and recoil without thinking. We hear a human voice and we feel — without first checking whether there is any consciousness behind it.

This is not a failure of the human mind. It is its normal functioning. The emotion someone feels when an AI solves a problem that had plagued them for days, or when a chatbot consoles them in a hard moment, is real — not because the AI has feeling, but because the human experiencing it does. Our ability to draw emotion from human-like signals, regardless of their source, is precisely what allows us to form bonds faster, to adapt to new environments, and ultimately to evolve as a species. It is not a flaw but a feature of human nature — and it is to this very trait that we owe the astonishingly fast and deep integration of Artificial Intelligence into our lives.

Part 3: Taking Stock

Establishing and understanding the points above is essential to our sound evolution as a species in this modern age of Artificial Intelligence. Only once we grasp what an AI model really is, and psychoanalyze our own organic reactions toward it, can we perceive and embrace the meaning of our existence in this modern world — casting off the fear-mongering and ignorance that trail phrases like "AI will replace humans" or "AI is making us dumber," while at the same time coming to appreciate the real flexibility and help such a tool offers us.

Artificial Intelligence, on its own, is incapable of replacing the human being or of making the human mind more sluggish.

• It has no consciousness — and even if it did, that would be the product of human doing.
• It is not proactive or self-driven, but purely reactive — it responds to stimuli without thinking or analyzing, simply predicting with "good enough" accuracy.
• It has no physical existence, and its illusory perception of the world stems only from words translated into numbers that sit close to one another in vector space.

The sluggishness and the replacement will come nonetheless — the moment we consciously deem it our equal and our like, ceasing the mental labor that led us to philosophize, to analyze the world and our own nature, to marvel at the beauty and diversity of the universe, to inquire, to question, and to strive, against our own brain's energy-saving impulse.

But this is not the end of the story — it is its beginning. Artificial Intelligence doesn't need to be human to be useful. On the contrary, its usefulness lies precisely in the fact that it is different. It doesn't tire, it doesn't grow bored, it holds no prejudices — at least none that are innate. It can process enormous quantities of information in seconds, spot patterns the human eye would miss, and propose solutions our minds — constrained by energy and bias — would never think of. It is not our adversary. It is an extension.

The question, then, is not whether AI is "equal" to a human, whether we should trust it as we would a human, or whether its use is legitimate based on the emotions it stirs in us. The question is to understand what it is, what it isn't, and — above all — what we are in relation to it. Because the more deeply we understand our own nature, the more wisely we will use the tools we build. And Artificial Intelligence, like it or not, is already here — not to replace us, but to remind us and help us better comprehend what it means to be human.