We are trying to understand how drug delivery supply chains can be more resilient.
Play our new Gamette and help us to learn more about human behavior within drug delivery supply chains.
GAMETTES
A Playful Approach for Capturing Decision-Making for Informing Behavioral Models
WHAT ARE GAMETTES?
Gamettes are short and relatively simple games that immerse human players in a decision-making scenario, and can be easily adapted to test different settings. By using Gamettes we can put the human into the loop of an agent-based simulation and get insight into the black box of human behavior.
The integrated simulation framework. The Flow Simulator simulates the network and communicates with the Gamette Server to receive player input from the Gamette Client and send updated parameters to the players. Player decisions are also stored (“Gamette Decision Storage”) for later gameplay analysis.
GAMETTE DESIGN
Gamettes are coupled with the Flow Simulator which is a multi-agent simulation environment designed based on Partially Observable Markov Decision Processes (POMDPs). A POMDP is a general framework to model sequential decision problems where the state of the system is not completely observable by the agent. The design of the Gamettes follows the same idea of the POMDP framework by allowing players to collect <information>, take <actions>, and receive <rewards>. Read More…
The term Gamette is a contraction of “game” with “vignette”
Similar to a vignette, a Gamette aims to provide a brief description of a situation as well as to portray someone.
GAMETTES ARE VALID
Gamettes have been tested and validated by replicating the well-known Beer Distribution Game, while changing the context to drug delivery supply chains. Results from our experimental study confirmed the expected behaviors and patterns such as the Bullwhip Effect for the players who played the wholesaler role.
Players who received order suggestions based on an optimal ordering policy, tend to deviate from those suggested order amounts although they have been informed about how the system calculates these order suggestions. The figure shows the order deviations in the form of error bars on the average suggested order amount. The error bars show how much on average the players deviated from the order suggestions. Players tend to over-order at almost all time periods. Read More…
What Players saw in ordering scene without order suggestions
What Players saw in ordering scene with order suggestions
LEARNING FROM GAMETTES
We also evaluated the performance of AI agents as well as supervised learning approaches in imitating the behavior of human decision-makers using data extracted from Gamettes. We considered a regression model, and two imitation learning algorithms (Behavioral Cloning and Generative Adversarial Imitation Learning) and compared their performance in imitating players’ behavior. Read more…
Contribute to our research!
We are trying to understand how drug delivery supply chains can be more resilient.
Play our new Gamette and help us to learn more about human behavior within drug delivery supply chains.
