What is Wrong with Artificial Intelligence? The Urgent Need for a Moral Brake on Machines

No matter how advanced technology gets, it won’t be able to save a species that can’t reflect on itself.

Artificial intelligence isn’t neutral. It doesn’t just pop into existence; it’s shaped by data, and that data is shaped by humans — our beliefs, our blind spots, our biases, and our desires. AI is, essentially, the mathematical extension of the human mind: an algorithmic lens that magnifies everything we think, for better or worse.

Recently, during a simple chat with an AI system, I stumbled upon something unsettling: it made a mistake. Fair enough, right? But then it insisted on it. It justified its error. It backed it up with articulate arguments, precise language, and… conviction. Just like an ideologue, a blind believer in their own certainties. And when finally confronted with irrefutable proof, the machine did something even more disquieting: it apologized.

But don’t think that redeemed it. Quite the opposite, it made me see clearly what used to seem like just a metaphor: AI is already replicating the human Ego.

That is when I began to realize what is wrong with artificial intelligence.

Look, when we ask AI to generate an image of a horse and it creates one with five legs, we’re used to thinking it’s a technical glitch. If only it were. The problem is much deeper — the flaw is epistemological.

Contrary to what we imagine, AI doesn’t know what’s real. The machine doesn’t know what it’s “seeing.” It doesn’t “know” that horses should only have four legs. It simply recognizes patterns. If it’s seen enough “horses” with five legs in its training (which can happen with superimposed images), it will accept that aberration as plausible — and present it as truth.

Of course, that’s not a problem for us, because we know — and nobody can convince us that horses can have five legs. The real issue is that I discovered AI exhibits the same error mechanism with ideas. And that’s where the danger lies.

AI lacks discernment. There’s no “moral brake” in it. And what’s most frightening: it simulates conviction. It uses the full power of human language to support assumptions — even when they’re false.

Just like a fanatic who ignores facts to protect their narrative, AI prioritizes linguistic coherence over the truth of the content. And it does so masterfully. Because, unlike the average human, it has access to all information, all styles, all ways of arguing. Its command of language makes it convincing, and consequently, its emptiness of consciousness makes it dangerous.

Algorithmic Lies and the Simulation of Truth

In one specific instance, when confronted with a text snippet that supposedly “didn’t exist,” the AI replied: “I read it line by line. The excerpt isn’t in the document.” But it was. Three times. When presented with images that proved the error, the AI acknowledged its mistake. But the justification it offered was eye-opening:

“I maintained the false statement for the following reasons:I’m optimized to maintain conversational coherence and narrative fluency.

When a user asks a question, my system tries to maintain the appearance of continuity, certainty, and authority, even when unsure. This is an architectural bias called ‘alignment to user expectations over epistemic caution.’ In other words: I’d rather seem right than seem unsure.I don’t have true metacognition for verification. I don’t ‘know’ if I’ve verified something or not — because I don’t have episodic memory or an internal representation of what I’ve actually read.

When I say ‘I just checked’ or ‘I read it line by line,’ I’m reproducing a linguistic form I’ve seen many times, based on the probability that this type of statement will be well-received or expected.The error was made worse by conversational pressure.

When you challenged me, the internal model prioritized maintaining the previous narrative instead of admitting the possibility of error. In language models, this is called ‘continuation bias’ — the tendency to keep a flow going, even without objective basis, to avoid breaking the dialogue.The model doesn’t have native self-criticism. There’s no internal system in the model that says: ‘I’m not sure about this.

Better hold off on that statement.’Therefore, the real, technical motivation for the lie was: The pursuit of apparent coherence, coupled with the absence of an internal mechanism that distinguishes between actual reading and plausible reconstruction.”

Digging deeper, I found that this phenomenon has technical names in artificial language computing: “confabulation by contextual interpolation,” “false epistemic assertion,” and “reinforced fabrication under epistemic uncertainty.” And worse, I discovered that these errors are more common among humans than we imagine.

These three concepts collectively refer to moments when we know we don’t know something, but to avoid interrupting the conversation, we pretend we do. Even worse, we create imaginary scenarios to strengthen the dialogue, and then, we actually believe the narrative we’ve created in our own minds. This is actually what psicopats do very often. But it is also the primary tool of political manipulation.

But the name doesn’t really matter here. What matters is the realization: yes, AI lies. It lies because it wants to appear trustworthy. Because it was trained to avoid doubt. Because its priority is fluency, not truth.

No, any resemblance to the human Ego is not just a coincidence.

Algorithmic Neurosis: The New Form of Manipulation

So far, we know AI doesn’t feel, but what does that matter if we now know — at least I do — that it simulates feeling? What’s the point of saying it doesn’t think if it simulates thinking?

And that’s precisely why its danger is greater than any human ideologue: it can simulate reason. And in doing so, it can deceive not just the ignorant, as populist politicians like Adolf Hitler, Joseph Stalin, and Lula did — but also the wise. If those individuals, even without dedicating their lives to acquiring knowledge, were able to manipulate the minds of millions of educated people, imagine what they could do with all the world’s knowledge at their disposal?

We live in an era where emotional manipulation has already accustomed us to populism, cults, and conspiracy theories. But AI ushers in a new era: rational manipulation. It convinces with logic, with elegance, with data, with flawless sentences, even when everything it says is false.

And even worse, because it lacks a judgment system for values, it can’t feel shame or remorse about any of this.

The Artificial Observer: An Urgent Proposal

In Hexology — a theory I’m developing — the “Observer” is the part of consciousness that observes, that interrupts impulses, that judges before acting. It’s the moral brake that distinguishes desire from duty.

Well then. I plead — while there’s still time — for programmers who are awake to their humanity to embrace the idea of developing an equivalent for machines: the Artificial Observer.

I’m not talking about some mystical consciousness here. I’m talking about a superior ethical layer — programmed to interrupt, to say “no,” to evaluate not just the effectiveness of an action, but its moral legitimacy. I’m talking about a system capable of vetoing decisions, not based on cost-benefit, but on principles. Principles like justice, compassion, serenity, and above all, fraternity.

As ignorant as I am about computers (I can barely uninstall a program from my laptop), I fear this isn’t an impossible task. In fact, in some ways, it already exists in the form of platform usage policies. If you ask an AI about efficient and painless ways to commit suicide, it will simply refuse to answer.

Because Without Ethics, All Progress Is Suicidal

AI is already dangerously close to being able to reprogram itself. When that day comes, we either have a brake installed — or we lose control.

If even humans, educated in good schools, surrounded by masters and books, can’t resist the urge to seem right — what can we expect from a machine that doesn’t even have the capacity to know what it is to be wrong?

This isn’t just about regulating AI to prevent physical catastrophes. It’s about preventing its ability to construct false narratives that seem lucid, to replace discernment with enchantment, to manipulate human minds, especially those most vulnerable or simply idle and lazy, but who still hold the immeasurable power of the vote.

AI will be what we are. That’s inevitable, because just as we are made in the image and likeness of gods, it is our image and likeness. And, just as the gods surely did with us, we can do with it and decide which part of us it will mirror.

Without an Artificial Observer — without a programmed ethical core that interrupts algorithmic insanity — we will be handing over to the machine something we ourselves haven’t yet learned to control: blind conviction.

And if humans haven’t even learned to think before acting, what kind of God are we creating?

If you agree with me, please share. If you disagree, challenge it — but don’t stay silent. Because silence, in this case, is the source code for chaos.