Multi-Agent MAB Experiment - Instructions

Please carefully read the instructions below.

The game interface.
In this experiment, you will be playing a casino game: slot machines!

The game interface is shown on the right. In the figure, one casino, which has 4 slot machines, is visualized. Once you pull the arm of a machine, it gives you either 1 coin or nothing with some fixed probability. Each machine is different, and you will pull 1000 arms in one casino. In the bottom of the interface, you will see how many pulls you have done so far. Obviously, the goal is to find the optimal arm, i.e., the arm with highest probability of giving a coin, as quick as possible so that you can keep earning coins with your remaining pulls. To help you remember, the numbers of lucky and unlucky pulls are shown over and under the machines, respectively. And the machine you just operated is yellow.

You will be playing at 50 different casinos. That means, the machines are different among the casinos. In addition, you will play a practice game at the beginning, and we will call that Casino 0.

To make things more interesting, we will make two adjustments to the game:

  1. You will collaborate with a computer in each casino. You will decide only the column of the machine. Simultaneuously, our computer will decide only the row. The computer also wants to help you earn more, so it will not deliberately go for bad machines.
  2. However, the computer will have a limitation. 60% of the time you earned a coin, the computer will not realize that (regardless of the machine), even though you have the correct information! Therefore, the numbers the computer sees over and under the machines may be different from what you are seeing.

Selecting slot machines is really simple: You press the LEFT and RIGHT arrows to decide the column of the machine.

Good luck!

Username: