Configural learning by cleaner fish in a complex biological market task

Individuals engaged in cooperative or mutualistic interactions can benefit from adjusting their strategic behaviour according to the bargaining positions of their partners. In interaction networks in which individuals cooperate with a variety of partners, assessing partners' options and expected interaction payoffs might be crucial for such behavioural optimization to occur. However, the cognitive challenges involved in this assessment and the constraints they may pose for strategic behaviour are currently poorly understood. Here, we investigated whether cleaner fish, Labroides dimidiatus, can learn to optimally prioritize partners, when faced with an ecologically relevant challenge that requires them to distinguish between interaction contexts. We expanded on a relatively well-studied problem that cleaners face in nature: choosing between client fish that simultaneously seek service. In such situations, cleaners should give priority to visitor client species as these tend to leave if not serviced immediately, while resident species will readily await inspection. In laboratory ‘biological market’ experiments simulating this choice using ephemeral and permanent food-offering plates, cleaners could learn to prefer visitors. However, the actual challenge cleaners face is more complex: interactions in the reef form a ‘complex market’ whereby clients appear in all possible configurations. In homogeneous configurations (visitor–visitor and resident–resident), the payoffs received from tending residents are higher which may slow down or even prevent learning to prioritize visitors. Adjusting the plate paradigm to this more complex scenario, we found that exposure to homogeneous configurations indeed posed a challenge to cleaners' strategic prioritization, diverting their preferences towards residents. Our results also revealed individual differences in performance, as some cleaners still learned to prefer visitors. This suggests that cleaners may not only learn about each client type separately, and that learning about cue configurations, a precursor of learning complex structures, may play a role in shaping their strategic behaviour in cooperative interactions.