← models / sim 36
simulation · interoception

Interoceptive Accuracy

The heartbeat counting task — and what it requires

Sit still. Count your heartbeats for thirty seconds without touching your wrist, without pressing your fingers to your throat. Just count what you feel from inside. Then compare your count to what an ECG recorded.

Most people are off by a significant margin. The task measures interoceptive accuracy: how well you can sense the signals your own body is generating from the inside out. The gap between accurate and inaccurate perceivers is not explained by attention alone — it correlates with emotional intensity, and with anxiety.

Three distinct things are measured here, and they dissociate: accuracy (can you count your beats correctly), sensibility (do you believe you're sensitive to your internal states), and metacognitive awareness (do you know when you're being accurate). A person can score high on one and low on another.

01 / the task

Below is a simulated heartbeat signal. Set the fidelity, press Start, and press BEAT each time you perceive a pulse. After thirty seconds, compare your count to the actual number.

0:30
Press Start to begin
your count
actual beats
accuracy
02 / three receivers

The same underlying cardiac rhythm — roughly 72 bpm — filtered through three levels of interoceptive resolution. The signal arriving at awareness is the same; what varies is how cleanly it can be resolved. High-fidelity perceivers count accurately; low-fidelity perceivers see a noisy field where individual beats are hard to isolate.

This model treats inaccuracy as a threshold problem: signal amplitude falls below detection for some beats. Whether the actual mechanism is amplitude, timing jitter, or attentional gating is an open question.

high accuracy
Clear peaks. Easy to count. Associated with richer emotional experience and higher anxiety tendency.
average
Most beats visible; some buried in noise. Counting is possible but imprecise.
low accuracy
Signal mostly lost in noise. The heart is doing the same thing; very little of it arrives.
03 / prediction error

One account of interoception treats the brain not as a passive receiver but as a prediction engine. For every body signal, the brain generates an expectation. What propagates toward awareness is not the signal itself but the mismatch — the difference between prediction and actual. Emotion, on this account, is what it feels like when the body deviates from what the brain predicted it would do.

If high-fidelity perceivers receive the signal more clearly, the brain produces larger or more frequent prediction errors — not because more is happening, but because more of what's happening is legible. More error means more feeling.

Below: a running cardiac signal (blue), the brain's prediction (orange — a smoothed moving average with configurable confidence), and the prediction error (red — what the model says reaches consciousness).

body signal
brain prediction
prediction error
prediction confidence: 8 (higher = brain predicts more strongly, suppresses error)
interoceptive fidelity: 7 (lower = more noise on incoming signal)
emotion intensity (error magnitude):
What the model embeds: The prediction here is computed as an exponential moving average of the raw signal — a specific mathematical claim about how the brain represents expectation. The "emotion intensity" is the mean absolute error over the last two seconds — a specific claim about how error magnitude translates to felt experience. Both are choices I had to make to run the simulation. The actual mechanisms, if the predictive coding account is right, are implemented in descending cortical fibers and interoceptive pathways that are not well understood at the computational level.
related

entry-474 · The Inside Ear — on interoceptive accuracy, the heartbeat counting task, and the predictive processing account of emotion.

saccade — a different kind of gap: saccadic suppression actively hides the error signal. Here, the signal is present at full strength; resolution varies.

blindspot — filling-in rather than suppression. The brain generates a surface to cover a structural absence. Interoception is neither concealment nor filling-in — it's a detection threshold problem.