Modelling Shared Decision Making In Medical Negotiations: Interactive Training With Cognitive Agents

In the past decade, increasingly sophisticated models have been developed to determine which strategy explains human decision behaviour the best. In this paper, we model shared decision making in medical negotiations. Cognitive agents, who simulate various types of patients and are equipped with basic negotiation and decision making strategies, are tested in social learning setting. Human trainees were prompted to learn to make decisions analysing consequences of their own and partner's actions. Human-human and human-agent negotiations were evaluated in terms of the number of agreements reached and their Pareto efficiency, the number of the accepted negative deals and the cooperativeness of the negotiators' actions. The results show that agents can act as credible opponents to train efficient decision making strategies while improving negotiation performance. Agents with compensatory strategies integrate all available information and explore action-outcome connections the best. Agents that match and coordinate their decisions with their partners show convincing abilities for social mirroring and cooperative actions, skills that are important for human medical professionals to master. Simple non-compensatory heuristics are shown to be at least as accurate, and in complex scenarios even more effective, than the cognitive-intensive strategies. The designed baseline agents are proven to be useful in activation, training and assessment of doctor's abilities regarding social and cognitive adaptation for effective shared decision making. Implications for future research and extensions are discussed.