Balancing personal and social information in cooperative problem solving
Research Unit 1, SCIoI Project 34
A fundamental challenge in cooperative problem solving in human, animal and robotic groups is the integration of personal and social information. Relying too heavily on personal information prevents the spread of information among group members, whereas relying too heavily on social information may hamper profitable personal exploration and reduce collective performance. Here we investigate this key process by studying how collectives of different complexities dynamically balance personal and social information use across different levels of environmental complexity to achieve collective intelligence. We will investigate both the performance of fixed strategies across different environments, as well as how agents learn about which strategies to use when facing unknown environments. We propose to investigate both human and robotic groups. To foster integration between both systems, we will use similar experimental paradigms: collective spatial search tasks. In human groups, we plan to use immersive reality: humans will control avatars in the virtual world and collectively search for resources. This approach allows full experimental control, providing an ideal testbed for studying cooperative problem solving in humans. For robotic groups, we plan to use swarms of Thymio II robots, performing collective spatial search tasks. Both ‘systems’ will be probed with collective search tasks of increasing complexity, starting with simple binary resources, and working towards more complex probabilistic resource environments (spatial multi-armed bandits). These tasks will share increasingly more overlap with cooperative shepherding. Human and robotic experimentation will continuously interact, using the following iterative steps: (i) extract fundamental principles of cooperative problem solving from human experimentation; (ii) test the robustness of these strategies across a broader set of environments using agent-based-modeling; (iii) use robotic simulations to test the performance of the most robust strategies in the ‘physical’ world; (iv) implement these strategies in a robotic platform, and, finally (v) feed the insights and predictions from the modeling and robots back into human experimentation.