Conscious Artificial Intelligence: Self-Model Simulator

Motor Command (Efference Copy)

Motor amplitude 35°

Motor frequency 0.60 Hz

Self-Model Learning

Learning rate η 0.010

External perturbation 0°

Self-Recognition

SELF-GENERATED

Prediction error

0.0°

Confidence

100%

Training steps

How it works

Each of the arm's three joints receives a motor command a_t — the same signal is kept internally as an efference copy. A small self-model tries to predict the resulting sensor reading before it arrives:

ŝ_t = w·a_t + b          (predicted sensation from own command)
e_t = s_t − ŝ_t          (prediction error = actual − predicted)
w ← w + η·e_t·a_t
b ← b + η·e_t·0.1        (online gradient descent on e_t²)

When the joint moves only because of its own commanded a_t, the model converges and e_t shrinks toward zero — the system attributes the sensation to itself. An external perturbation adds motion the command never predicted, so e_t spikes and the same comparator now flags the input as non-self:

confidence = clamp(1 − |e_t| / e_max, 0, 1) × 100%
self  if confidence > 60%
non-self  otherwise

Motor amplitude / frequency — the pattern the arm generates on its own; this is the signal the self-model learns to predict.
Learning rate η — how fast the weights chase the prediction error; too high and the model overshoots and oscillates.
External perturbation — an unpredicted disturbance injected into the actual sensor reading, simulating something moving the arm that isn't its own command.
Trail — recent end-effector positions, colored green→red by the error recorded at that instant.

This comparator scheme is a simplified version of the "comparator model" of the sense of agency (Blakemore & Frith) and of self-modeling robots that learn to distinguish their own body's dynamics from the environment (Bongard & Lipson, Science 2006) — a concrete, testable computational account of self-representation, one candidate mechanism researchers study on the way toward machine self-awareness.

Self-predicted (low error) Unpredicted / non-self

⚠ Couldn't load the 3D engineThree.js failed to load from the CDN. Check your connection and reload.

Conscious Artificial Intelligence: A Self-Model Simulator