The Seventh Sense

Another way of perceiving the implicit?

Human beings have a strange ability to perceive what is never fully expressed, those diffuse undercurrents that quietly shape our interactions. That familiar instinct of “watching what happens when nothing seems to be happening.”

I was reminded of this again over the past few weeks. I’ve been seeing a physiotherapist for three months because of an injury to my right knee, and I’ve found myself observing her closely. Her role goes far beyond rehabilitation exercises. When I tell her I can no longer bend my knee after a setback that happened at the exact moment I thought I was healed, she immediately separates the physical from the mental. She shifts my attention elsewhere, asks me to try another movement, starts talking about something unrelated. Before I even realize it, the movement comes back.

Sometimes she listens without answering. It’s a thoughtful kind of silence, not indifference, quite the opposite. She leaves room for me to unload my frustrations and doubts as an injured person without trying to over-explain things or reassure me too quickly. In a way, by staying silent, she allows me to work through it myself.

These small details, as I was recently discussing with a consultant friend, remain difficult to translate into a machine. The reason is fairly simple: conversational language models are designed to respond. Silence, hesitation, suspension, or choosing not to intervene at a particular moment are still extremely difficult behaviors to model.

There are other nuances that are even harder to formalize. Someone who had been working with Claude on sensitive emails pointed out, for example, that the AI almost always tries to leave the other person a diplomatic way out. Yet in some situations, he intentionally chose a firmer tone. It wasn’t about ego or rigidity, but about backstage dynamics, an entire layer of relational context impossible to compress into a prompt or a Markdown file.

Faced with this, we often fall into a simple opposition: humans on the side of the implicit, machines on the side of the explicit. But reality is more subtle than that. Large language models also access certain forms of implicit meaning. Not lived implicit meaning, not the kind rooted in bodies and experience, but another form of implicit structure embedded within language itself.

This is precisely what Marc Cavazza explores in Cahiers de sémiotique des cultures (2026). Drawing on François Rastier’s interpretive semantics, he argues that LLMs are not merely “stochastic parrots” mechanically predicting the next likely word. They preserve broader semantic coherences and continuities of meaning, what semiotics refers to as isotopies.

An isotopy is the diffuse continuity that gives a text its atmosphere, its underlying coherence, its sense of family resemblance across different expressions. Cavazza gives examples such as “president,” “debate,” “inauguration,” or “executive order.” Individually, these words refer to different realities; together, they immediately establish a shared political horizon.

LLMs appear capable of extending these continuities far beyond the lexical level. We saw this almost caricatured when some users noticed that phrasing requests aggressively or impatiently could sometimes produce more direct, even slightly more precise, answers. The machine obviously does not “understand” human anger. It simply extends statistical regularities in which markers of tension, urgency, or insistence are frequently associated with certain types of responses.

Cavazza even speaks of a form of “material hermeneutics,” where meaning no longer emerges solely from abstract intention or inner subjectivity, but through relationships between texts, situations, and observable regularities within large corpora. In this view, interpretive coherence can emerge from statistical operations applied across massive textual datasets. As he points out, a prompt activates discursive frameworks, tonalities, genres, and cultural references already embedded in the data used to train these models.

When someone leaves a meeting thinking, “something feels off,” they are already unconsciously processing countless explicit and implicit signals: hesitations, reformulations, slightly prolonged silences, shifts in tone, or subtle changes in conversational direction. LLMs may simply shift the scale at which some of these regularities become perceptible.

And what if tomorrow, instead of asking AI for a simple meeting summary, we asked: What remains implicit here? What tensions emerge beneath the surface? Which positions keep resurfacing without ever being stated openly? What, exactly, is circulating through the conversation without ever truly being said?

As always, the point will not be to take these interpretations at face value. Discernment will remain essential. But these tools may still sharpen our perception, redirect our attention, and open new ways of reading situations.

We often speak of a “sixth sense” to describe that difficult-to-explain human intuition, the ability to sense a situation before we can articulate it. Perhaps the seventh sense begins somewhere around there, in the meeting point between human intuition and new tools for exploring what usually escapes perception, for better or worse.

MD