Task Complexity and Performance in Individuals and Groups Without Communication

• Published in: Computational Theory of Mind for Human-Machine Teams pp. 102 - 117
• Main Authors: Gulati, Aditya, Nguyen, Thuy Ngoc, Gonzalez, Cleotilde
• Format: Book Chapter
• Language: English
• Published: Cham Springer Nature Switzerland
• Series: Lecture Notes in Computer Science
• Subjects: Cognitive model; Group size; Instance-based learning theory; Task complexity
• ISBN: 3031216709; 9783031216701
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/978-3-031-21671-8_7

While groups where members communicate with each other may perform better than groups without communication, there are multiple scenarios where communication between group members is not possible. Our work analyses the impact of task complexity on individuals and groups of different sizes while solving a goal-seeking navigation task without communication. Our major goal is to determine the effect of task complexity on performance and whether agents in a group are able to coordinate to perform the task effectively despite the lack of communication. We developed a cognitive model of each individual agent that performs the task. We compare the performance of this agent with individual human performance, who worked on the same task. We observe that the cognitive agent is able to replicate the general behavioral trends observed in humans. Using this cognitive model, we generate groups of different sizes where individual agents work in the same goal-seeking task independently and without communication. First, we observe that increasing task complexity by design does not necessarily lead to worse performance in individuals and groups. We also observe that larger groups perform better than smaller groups and individuals alone. However, individual agents within a group perform worse than an agent working on the task alone. This effect is not the result of agents within a group covering less ground in the task compared to individuals alone. Rather, it is an effect resulting from the overlap of the agents within a group. Importantly, agents learn to reduce their overlap and improve their performance without explicit communication. These results can inform the design of AI agents in human-machine teams.