In this seminar, Tomasz Kopczewski (University of Warsaw) presents the interactive online tool “Entropy Lab”.

Entropy Lab began as a teaching tool for informational entropy and evolved into a proof-of-concept for modeling uncertainty. Participants interact with environments represented as Markov chains that share identical stationary distributions but differ in conditional entropy and predictability. By paying explicit prediction costs, they trade resources for uncertainty reduction, creating a Maxwell demon–like setting. Sudden regime changes test adaptation under Keynesian uncertainty. Running the experiment under additive and multiplicative payoff regimes reveals volatility drag and survival effects, where ensemble averages diverge from typical trajectories. The project combines entropy, information costs, and ergodicity into a simple but behaviorally transparent experimental environment.

You can find the Entropy Lab here and try out the online experiment here.

The seminar was hosted by Emilie Rosenlund Soysal (London Mathematical Laboratory) and James King (Science Practise).