ColabTrails

Embodying intelligent behavior in social context

Abstract

Research has found that working in pairs or small groups can have beneficial effects on learning and development, particularly in early years and primary education (Benford et al., 2000). Although collaboration is very important during these early years it is implemented in a very minimal way at primary schools. In order to design a game which improves the collaboration between children from 6 to 8 years old, the existing game Colored Trails (Gal, Grosz, Kraus, Pfeffer & Shieber, 2005) was evaluated. From this evaluation, a new game, ColabTrails, was created which shares some mechanics with the Colored Trails game. The game aimed to encourage the players to collaborate by trading different colored tokens with each other to move across the board and finish earlier than the computer.

For the computer to sense and perform a smart or stupid move a learning algorithm needed to be implemented within the computer. Reinforcement learning was chosen for the concept. To be even more specific, the final prototype contains Q-learning. This algorithm simplifies the analysis and approximates the optimal action-value (Sutton & Barto, 1998). The agent will follow the optimal action-values for each movement to reach the final goal.

Reinforcement learning has proven to be a good tool to create a playful experience, and thus suitable for the social context and embodying intelligent behavior. During the user-deployment study, the prototype was perceived as too easy, several future work and improvements are suggested to keep the game interesting and to encourage the players to have a more elaborate collaboration. The algorithm can be applied better, in a way that the interaction between the two players will be led by the computer. To conclude the research of ColabTrails, we can say that we created a fun and interesting system steering children to collaborate, but not capable of letting the children realize the effects of this collaboration.

Responsibilities

Concept development

User interaction design

User interface

Implementation reinforcement learning

User deployment

Analyzing user experience

Concept - Envisioned interaction before collaboration

Concept - Envisioned interaction after collaboration

Low-fi prototype for user deployment

Low-fi prototype for user deployment

References

Benford, S., Bederson, B. B., Ã…kesson, K. P., Bayon, V., Druin, A., Hansson, P., ... & Simsarian, K. T. (2000, April). Designing storytelling technologies to encouraging collaboration between young children. In Proceedings of the SIGCHI conference on Human Factors in Computing Systems (pp. 556-563). ACM.

Gal, Y. A., Grosz, B. J., Kraus, S., Pfeffer, A., & Shieber, S. (2005, July). Colored trails: a formalism for investigating decision-making in strategic environments. In Proceedings of the 2005 IJCAI workshop on reasoning, representation, and learning in computer games (pp. 25-30).

Sutton, R. S., & Barto, A. G. (1998). Reinforcement learning an introduction. Cambridge, Mass.: MIT Press.

© 2020 Jesper van Bentum