This simulation turns the abstract ethics of AI systems into something you can watch happen: a layered decision network where every "decision" is a pulse of light travelling from raw input, through the model's hidden internals, to an output the system acts on.
Each pulse rolls a chance of being biased based on the Algorithmic bias slider, then lands on an output node — fair decisions glow teal, biased ones glow red. Autonomy level sets how fast and how often the system decides on its own, Decision impact scales how much every decision visually "weighs", and Explainability (XAI) controls the fog obscuring the hidden layer: the lower it is, the more the model's internals become an unreadable black box.
Drag Algorithmic bias to see the ratio of red vs teal decisions shift, raise Autonomy to watch decisions fire faster with less pacing, and lower Explainability to thicken the fog around the hidden layer. Watch the ring around the whole network — it is the visual "human oversight" and shrinks as the Accountability gap grows. Press Reset to clear the decision log and start clean.
The panel's Complexity score is literally Autonomy × Impact — the same relationship researchers use informally to argue that highly autonomous, high-stakes systems (like autonomous vehicles or medical AI) need proportionally stronger explainability and audit trails to stay accountable.
Real AI systems make decisions by passing data through layers they don't fully expose to the humans relying on them. This simulation makes that pipeline visible: an input layer (raw signals), a hidden layer (the model's internals — the "black box"), and an output layer (the decisions the system acts on).
Complexity = Autonomy × Impact — how hard a system is to meaningfully supervise, from how independently it acts and how much its decisions matter.
Accountability gap = Complexity × (1 − Explainability) — the risk left over once you subtract how much of that complexity can actually be explained.
Because Accountability gap multiplies Complexity by (1 − Explainability), even a low-autonomy system can still carry real risk if its Decision impact is high and it stays unexplainable — which is exactly why high-stakes AI (medical, judicial, financial) is held to stricter explainability standards than low-stakes AI.