It is now widely accepted that a variety of interaction strategies in animals achieve optimal or near optimal performance. The challenge is in determining the performance criteria being optimized. A difficulty in overcoming this challenge is the need for a large body of observational data to delineate hypotheses, which can be tedious and time consuming, if not impossible. To alleviate this difficulty, we propose a system — termed ``in-silico behavior discovery" — that will enable ethologists to simultaneously compare and assess various hypotheses with much less observational data. Key to this system is the use of Partially Observable Markov Decision Processes (POMDPs) to generate an optimal strategy under a given hypothesis. POMDPs enable the system to take into account imperfect information about the animals' dynamics and their operating environment. Given multiple hypotheses and a set of preliminary observational data, our system will compute the optimal strategy under each hypothesis, generate a set of synthesized data for each optimal strategy, and then rank the hypotheses based on the similarity between the set of synthesized data generated under each hypothesis and the provided observational data. In particular, this paper considers the development of this approach for studying mid-air collision-avoidance strategies of honeybees. To perform a feasibility study, we test the system using 100 data sets of close encounters between two honeybees. Preliminary results are promising, indicating that the system independently identifies the same hypothesis (optical flow centering) as discovered by neurobiologists/ethologists.