🐯 Two doors — one hides a tiger
You can't see in.
Listen
— the growl is only ~85% reliable. Your
belief
updates by Bayes. Open the wrong door (tiger): a big penalty; the
correct
one: a reward. Tune the payoffs & noise below.
tiger LEFT
50%
tiger RIGHT
50%
Should you open, or listen again?
noise: growl accuracy
0.85
listen cost
−1
correct reward
+10
tiger penalty
−100
👂 Listen
🚪 Open left
🚪 Open right
⟲ New game