Gorilla Academy: Learning the value of information in an uncertain world

This is a classic n-armed bandit task. Participants choose between two slot machines (or bandits) with the aim of winning as many points as possible over the task. Each bandit has a different probability of paying out and different number of points on offer. Participants will learn by trial-and-error whether a bandit is:

• High risk High reward
• High risk Low reward
• Low risk High reward
• Low risk Low reward

For half of the experiment the environment is stable (i.e. 60 trials with a fixed 75:25 reinforcement schedule), in the other half the environment is volatile (i.e. a 80:20 reinforcement schedule that switches every 20 trials).

This task uses spreadsheets to assign magnitude and probability values on each trial. Different spreadsheets decide which version the participant does, i.e:

• Stable first with green paying out more (StableGreen)
• Stable first with blue paying out more (StableBlue)
• Volatile first with green paying out more (VolatileGreen)
• Volatile first with blue paying out more (VolatileBlue).

A simple bit of scripting is used to decide if participants win or lose on a specific trial, and to add winning points to the total score.

If you want to see how I created this experiment, organised and analysed the data, you can watch my video tutorials on Gorilla Academy in the Learning section.

This study is a replication of Behrens et al (2007)

Back to Open Materials