Ringr inyecta 1.2M€ para consolidar su liderazgo en España

When we talk to a telephone artificial intelligence, many times we do not know why, but something is off, our brain realizes it instantly. Everything is correct: the intonation is natural, the words are well chosen, there are no grammatical errors or awkward pauses. And yet, we sense that we are not talking to a person. The curious thing is that this "something" that makes us doubt is usually not in what is said, but in what is not said. In what is missing. In that ability to improvise, interrupt, doubt, hesitate. In that lack of spontaneity that, ironically, reveals that we are talking to a machine. Because in the end, appearing human is not a matter of perfection, but of imperfection.

The clues that give away the AI

Despite the enormous advances in language models and synthetic voice systems, most telephone assistants still do not go unnoticed. The experience can be correct, even useful, but it does not deceive. Why?

In large part, due to a set of subtle signals that accumulate and reveal the artificial origin of the agent:

• Symmetrical latency: humans react faster or slower depending on the difficulty of the question, the emotional state, or the relationship with the interlocutor. An AI responds with a cadence that is too regular, which is suspicious.

• Absolute control of the speaking turn: humans interrupt, hesitate, and talk over each other. AIs often wait for you to finish speaking and then respond accurately, as if they had an unalterable turn-taking structure.

• Speech without friction: no fillers, no "uh..." or "okay...", no repetitions or corrections. The responses are too clean, too well-structured.

• Emotionally neutral or poorly calibrated language: the AI can sound empathetic, but in a generic way. It struggles to modulate nuances, ironies, or the type of emotional response a human would use depending on the context.

These elements, separately, may go unnoticed. But together they generate a feeling that many users describe as "strange", "cold", or "too perfect".

Why do we distrust perfection?

Psychology has been studying for decades how we attribute humanity, trust, or authenticity to interlocutors. One of the most striking findings is that humans tend to trust more in the imperfect. Not out of arrogance or a sense of superiority, but because subtle errors make us feel we are in the presence of someone real.

When everything is flawless, we distrust. We suspect that there is a script behind it. That there is no spontaneity, no intention, no emotion. That is, in fact, the basis of the so-called uncanny valley: when a machine resembles a human too much but not quite enough, it generates rejection.

Language is one of the spaces where this tension is most evident. Because speaking well is not speaking perfectly. It is speaking with intention, with rhythm, with doubt, with emotion. And that implies a margin of error, adaptability, and context.

Feigning errors: a conversational strategy

This is where a powerful and, at first glance, contradictory idea arises: to sound more human, an AI must learn to make mistakes. Not to fail technically or functionally, but to reproduce the micro-errors and frictions that are part of natural speech.

Some strategies that are already being explored include:

• Introducing hesitations before responding ("Mmm... I think so...").

• Simulating self-corrections ("Sorry, I meant to say... that the report is due tomorrow, not today").

• Including fillers or verbal tics that denote familiarity or informality ("okay", "do I make sense?", "well...").

• Realistically managing interruptions, allowing the user to cut the AI off without breaking the flow.

• Intentionally varying response times, depending on the type of question or the perceived emotion.

Feigning errors does not mean compromising the quality of service, but enriching it. It gives the machine a more real, more credible texture. Because verisimilitude is built not only with data but also with noise.

The risk of over-humanizing

However, it is not about turning AIs into caricatures of humans. An excess of imperfection can be just as annoying as its absence. The challenge is to find a balance: that hesitations do not hinder, that errors do not irritate, that pauses do not become awkward silences.

The goal is not to deceive but to generate a more empathetic and close experience. An AI can be transparent about its nature while still sounding human. What matters is that the user does not feel they are talking to a recording or an automaton, but to someone (or something) that truly listens.

A new sensitivity in conversational design

Humanizing an agent is not just a technical problem. It is also a design exercise, a narrative, a psychology issue. It is not enough to have good language models or realistic voice engines. We must work sensitively on the complete experience: how the conversation starts, how the rhythm is managed, how to respond to the unexpected, how to convey doubt, surprise or joy.

We are facing a new frontier in conversational design. One where imperfection is not a flaw but a tool. One where humanity is not imitated only with realistic voice but with credible behaviors. One where the error is no longer what we avoid but what—well-managed—brings us closer to the other person.

Conclusion: the human as horizon

As AI becomes more capable, the challenge is no longer technical but cultural. We want machines to speak like us, but we have not yet fully defined how we speak. Perhaps that is why errors—those small cracks in language—seem so valuable to us. Because that is where the truly human appears: in the pause, in the doubt, in the emotional clumsiness. That is where a machine can finally begin to resemble us. And that is where we, as designers, developers, and thinkers of AI, must continue to explore: not towards perfection, but towards verisimilitude.