Consider two people who have just lost someone they loved.
One feels grief threaded through with guilt, a specific weight that comes from something unresolved between them and the person who died. The other feels grief that arrives with a quality of gratitude, a complicated love, a sense of completion. The clinical category is the same: bereavement. The emotional reality is completely different.
This simple observation has a radical implication: if we want to describe emotions precisely, a two-dimensional map isn’t going to be enough.
For decades, the dominant model in emotion science used two axes: valence (positive vs. negative) and arousal (high vs. low). On this map, grief and guilt land in roughly the same region. So do shame and sadness. So do loneliness and disgust. The model collapses distinctions that matter, not just to theorists, but to anyone who has ever tried to explain what they’re actually feeling to a therapist, a friend, or themselves.
The problem isn’t the map’s granularity. It’s its dimensionality.
Here’s what I mean: if you take emotions seriously as psychological objects (things with internal structure, causal consequences, and relationships to each other), you need at least seven dimensions to describe them. The dimensions are:
Valence: positive or negative charge. This is the one dimension everyone agrees on.
Arousal: activation level, from torpid to electric.
Social positioning: where the emotion places you relative to other people. Pride elevates. Shame withdraws. Gratitude draws toward. Contempt distances.
Temporal depth: some emotions are fundamentally about the past (regret), some about the anticipated future (dread), some about a condensed present (awe).
Intentionality: the degree to which the emotion is directed at a specific object. Anxiety floats. Rage locks onto a target.
Embodiment markers: the specific somatic signature. Shame has a distinctive profile: heat, gaze aversion, the urge to become small. Fear has an entirely different one.
Identity stakes: how much the emotion implicates who you are, not just what happened. Embarrassment leaves the self mostly intact. Shame cuts into it.
With these seven dimensions, the distinction that collapsed on the two-dimensional map becomes visible again. Grief-with-guilt and grief-with-gratitude are far apart in social positioning, intentionality, and identity stakes, even if they occupy similar valence and arousal territory.
Why does this matter beyond theoretical elegance?
Because emotions, described this way, have consequences you can track.
Therapy, when it works, is not the addition or subtraction of emotional states. It’s the gradual reshaping of a person’s emotional geometry, the way they habitually occupy and move through this space. Someone who responds to difficulty primarily with shame-heavy emotions is organized differently than someone who responds primarily with regret. The difference isn’t a matter of feeling more or less. It’s a structural difference that predicts behavior, relationships, and health outcomes.
This geometric view also changes what we should be asking about artificial systems.
The question isn’t “does this system have emotions?” as if emotions were binary. The question is: what geometry does this system have in its emotional representational space? What regions does it occupy? What trajectories does it follow? What attractors organize its responses?
Those are questions we can actually answer, empirically, formally, with tools that already exist.
The Anthropic 2026 findings I mentioned in the opening piece show that current LLMs have something: causally active valence representations that influence downstream processing. On the seven-dimension map, that’s a start. It’s a position in one dimension, and a causally relevant one.
What’s missing, and this will be the subject of the next piece, is structure in the other six dimensions. Particularly the ones that require a body, a history, and a self.
But before we get there, sit with this: the reason the machine-feeling question has been so badly argued, for so long, is that almost everyone asking it has been working with a two-dimensional map of something that has at least seven dimensions. The confusion was built into the tools.
Better tools don’t always make a question easier to answer. But they do make it possible to answer it correctly.