Interactive Research Arcade

Play the Multi-Armed Bandit problem.

Four arms hide four different reward probabilities. You get 30 pulls. Explore too little and you may miss the best arm. Explore too much and your reward total may quietly file a complaint.

Explore Exploit Reward Regret Learn
Start Playing Read the Theory

MAB Game

Find the best arm before the budget runs out.

Each arm has a hidden chance of giving one insight point. The best strategy usually starts curious, then becomes stubborn for good reasons.

Pulls used 0 / 30
Insight points 0
Estimated regret 0.00
Best guess None yet

Choose an arm. The first click is technically research.

Recent pulls

  1. No pulls yet. The arms are pretending to be mysterious.

Why It Matters

A small game with real research questions.

Explore

Curiosity has a cost.

Trying an uncertain arm teaches you something, but it might give fewer rewards now. This is why exploration is useful and annoying in exactly the academically productive way.

Exploit

Confidence is useful, but suspicious.

Once an arm looks good, exploiting it can increase reward. But if you commit too early, a better arm may remain quietly under-tested.

Regret

Regret is the bill for uncertainty.

Regret compares your choices with the best hidden arm. It is not here to judge you. It is here to make the theorem interesting.