CompCog: Towards a computational cognitive science of helping

帮助他人实现共同的目标涉及到大量的社会认知。例如，向另一个人传授一项技能需要了解学习者的目标、当前知识和偏好。尽管人类相互帮助无处不在，但大多数认知科学研究都集中在个人学习和决策上。因此，对于人类用来减少对他人的知识、目标和偏好的不确定性以提供相关信息并更有效地帮助他们的计算认知机制，我们知之甚少。例如，当我们问一个问题以更好地理解另一个人的目标或理解时，我们如何决定？我们如何根据我们对他人心理的不完全理解来调整我们的援助？在这项建议中，研究人员开发了几个计算认知模型，并在新开发的帮助和教学认知分析中将这些模型与人类行为进行比较。该项目利用部分可观察的蒙特卡罗决策问题框架，这是一个人工智能框架，通过在这些协作问题解决范例中推导出最优策略(即最有效的助手策略)，来提供有效行为的基准。对人类如何解决这一复杂问题的算法基础的洞察可能会为定义未来工作场所的人机团队的研究提供信息。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Helping others to achieve mutual goals involves a great deal of social cognition. For example, teaching a skill to another person requires understanding the learner’s goals, current knowledge, and preferences. Despite the ubiquity of humans helping each other, the majority of cognitive science research has focused on individual learning and decision making. Therefore, not much is known about the computational cognitive mechanisms that humans use to reduce uncertainty about the knowledge, goals, and preferences of others in order to provide relevant information and to help them more effectively. For example, how do we decide when we ask a question to better understand the goals or understanding of another person? How do we adapt our assistance to another on the basis of our incomplete understanding of their mind? In this proposal, the investigators develop several computational cognitive models and compare these with human behavior in newly developed cognitive assays of helping and teaching. The project leverages the framework of Partially Observable Monte Carlo Decision Problems, an Artificial Intelligence framework, to provide a benchmark of efficient behavior by deriving the optimal policy (i.e., the most efficient helper strategy) in these paradigms of collaborative problem-solving. Insights into the algorithmic basis of how humans solve this complex problem may inform research on human-machine teams that will define the workplace of the future.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.