Collective decision-making is ubiquitous when observing the behavior of intelligent agents, including humans. However, there are inconsistencies in our theoretical understanding of whether there is a collective advantage from interacting with group members of varying levels of competence in solving problems of varying complexity. Moreover, most existing experiments have relied on highly stylized tasks, reducing the generality of their results. The present study narrows the gap between experimental control and realistic settings, reporting the results from an analysis of collective problem-solving in the context of a real-world citizen science task environment in which individuals with manipulated differences in task-relevant training collaborated on the Wildcam Gorongosa task, hosted by The Zooniverse. We find that dyads gradually improve in performance but do not experience a collective benefit compared to individuals in most situations; rather, the cost of team coordination to efficiency and speed is consistently larger than the leverage of having a partner, even if they are expertly trained. It is only in terms of accuracy in the most complex tasks that having an additional expert significantly improves performance upon that of non-experts. Our findings have important theoretical and applied implications for collective problem-solving: to improve efficiency, one could prioritize providing task-relevant training and relying on trained experts working alone over interaction and to improve accuracy, one could target the expertise of selectively trained individuals.
