Encode the ball
Stimulating electrodes communicate ball position. In the final protocol, place coding represented one dimension and stimulation rate represented distance.
INPUT: SELECTED SENSORY ELECTRODESBIOLOGICAL INTELLIGENCE / FIELD NOTE 01
When a dish of neurons
met a moving ball.
A visual, interactive guide to the closed loop behind DishBrain’s Pong experiments: sensing a simulated world, translating spikes into action, then feeding the result back to the culture.
- ~800K
- cells reported
- 1024
- recording channels
- 4–40 Hz
- rate-coded input
ANATOMICAL 3D BRAIN MESH1,284spikes / second simulated
DRAG TO ROTATE · SCROLL TO ZOOM · CLICK A MARKER
ANATOMICAL BRAIN MESH
HEMISPHERES / CORTEX / CEREBELLUM
HEMISPHERES / CORTEX / CEREBELLUM
Mesh: Z-Anatomy / BodyParts3D, CC BY-SA 4.0
Important distinction: the hero renders 3D organoid-inspired architecture—radial rosettes, lumen-like voids, and an outer neuronal layer. The Pong study itself used a 2D in-vitro neuronal culture on an electrode array; this page does not claim a 3D organoid played the game or that it is connected to living cells.
THE CLOSED LOOP
Not “training data.”
Embodiment, stimulation, response.
The game state becomes patterned electrical input. Recorded activity from selected motor regions becomes a paddle command. The next game frame changes the next input.
Culture responds
The network’s spontaneous and evoked spiking activity is sampled through the microelectrode array in real time.
READOUT: ELECTROPHYSIOLOGICAL SPIKESMove the paddle
Activity in two defined motor regions maps to up or down. The experimental game world updates from that readout.
OUTPUT: PADDLE DIRECTIONChange the next signal
A hit was followed by a structured, predictable input. A miss was followed by an unpredictable stimulus before a reset.
LOOP: STATE → STIMULUS → RESPONSEINTERACTIVE MODEL / NOT A LIVE BIOLOGICAL FEED
Watch the loop play out.
PONG WORLDFEEDBACK: SENSORY
ELECTRODE ARRAY1,024 CHANNELS / MODEL
SENSORY INPUTMOTOR UPMOTOR DOWNRECORDED SPIKE
ACTIVITY TRACELOOP: 5.2 ms
SPIKE RATE1,284 Hz
COHERENCE0.71
COMMANDPADDLE ↑
The model visualises the published interaction pattern; it does not reproduce biological performance data or infer consciousness.
WHAT THE GAME IS ACTUALLY TESTING
A tiny environment,
a demanding timing problem.
Pong is useful because an action has an immediate, measurable consequence. The system only receives a sparse representation of ball position; it must change firing patterns in a way that moves the paddle into the right place at the right time.
HITpredictable
stimulation
stimulation
MISSunpredictable
stimulation
stimulation
Feedback here is not a “reward” delivered to a person. It is part of an experimental stimulation protocol used to compare closed-loop conditions.
RESEARCH NOTES
What’s established,
what’s still open.
Structured, closed-loop interaction changed performance over time.
The published work reports better performance with informative sensory input and compares feedback conditions against controls.
“Played Pong” does not mean general intelligence or a conscious game player.
The experiment demonstrates adaptive, goal-directed behaviour in a constrained laboratory setup. Interpretation of sentience remains contested and ethically important.
The platform makes neural dynamics observable during a task.
Modern systems extend the idea toward controllable biological-neural-network experiments, recording, stimulation, and simulation.