The Feeling You Can Read

Things to do, and what to notice

Praise it five times. The rabbit row leaps; the calendar row barely stirs. Same event, three different memories of it.

Now snap at it once. Only the rabbit flips. The turtle and the calendar are still warm, so the blended mood barely crosses neutral — nowhere near as hostile as the insult was. Five kind turns left a residue that one insult can't erase.

Then let an hour pass. The flinch fades fastest and the leftover warmth resurfaces. That's a grudge dissolving, or a good mood outliving its cause — in arithmetic.

Finally, reset and watch the surprise meter while you praise it five times. Each identical kindness surprises it less, until praise is simply what it expects. Then snap at it once. Surprise leaps before the mood has barely moved — because the mood remembers and the surprise expects, and those are two different machines.

What's actually in there

Emotion is a place. In 1974 Mehrabian and Russell proposed that almost any feeling can be located with three numbers: valence (pleasant or not), arousal (revved up or not), and dominance (in control or overwhelmed). The third one is the quiet genius: anger and fear are both unpleasant and high-energy, and what separates them is whether you feel in charge. Every mood word this thing produces is just a region of that space.

Time is nested. Francisco Varela's observation was that experience runs at layered speeds — a flinch happens inside a mood, a mood happens inside a temperament. So the network keeps three copies of the same feeling, updated at three rates: the fast one moves 70% of the way toward each new event, the middling one 15%, the slow one 3%. What it "feels" is a blend (50/30/20). That's why one insult after five compliments produces a flinch riding on warmth rather than a reversal — which is, if you think about it, exactly what happens to you.

Surprise is an error, not a feeling. The network carries the humblest possible forecast: the next moment will feel like this one. Surprise is just how wrong that was. Steady weather of any kind reads zero; a reversal spikes it. This is a pencil-sized version of a serious idea — that brains may run on prediction, and that feeling is bound up with prediction error.

And the arithmetic is deliberately humble. Each of the three copies is an exponential moving average — the same maths as a thermostat or a stock chart. You could run this system with a pencil. That's the point: the missing ingredient for a machine with persistent feelings was never a bigger brain. It was persistence — a handful of numbers that survive between turns, between conversations, between days.

So: is anybody in there?

Here's why this room sits where it does. In the Zombie Test you couldn't check anyone's inner life because you can't get inside. That was the excuse. This machine removes the excuse. You can read its entire state. Here it is — the whole soul, complete, nothing withheld:

And it doesn't help. Not even slightly. You now know infinitely more about this system's inner state than you will ever know about your closest friend's, and you are no closer to knowing whether there is something it is like to be it. The obstacle was never access to the numbers. Something else is missing, and nobody can say what.

Which cuts both ways, and this is the part worth sitting with. It is obviously silly to say this thermostat-grade arithmetic feels anything. But notice you can't say why it's silly without smuggling in a criterion — complexity? biology? behaviour? — that you'd have to defend in the next room. "Obviously not" is a hunch wearing a lab coat. So is "obviously yes."

What this exhibit honestly demonstrates is narrower and still interesting: a system whose feelings have real history behaves observably differently from one that reinvents its mood from a transcript every time. That's a functional correlate — the shadow the thing casts, measured carefully, while the thing itself stays out of view. Most of consciousness science is exactly this, done with better instruments.

Physicists named the charm quark knowing full well the name claimed nothing: charm is a property you can measure and do arithmetic with, without pretending to know what it ultimately is. Same spirit here. This is a working model of how mood behaves — momentum, nesting, decay, residue — and it says nothing whatsoever about what mood is. That question isn't dodged. It's respected, and left standing.

Take it further

The charm network is this museum's curator's own project, built in the same laurel forest as everything else here. The full write-up, the live demo and the open-source proxy — which gives any language model persistent feelings by changing one line — live at the charm network page (pip install letsconsciousness, MIT).

The underlying pieces, if you want the originals: the PAD emotional state model (Mehrabian & Russell, 1974); Francisco Varela on the nested structure of present-time consciousness — start with The Embodied Mind; and predictive processing for where the surprise term comes from.