RI: Small: TIDES: Trustworthy Interactive DEcision-making Using Symbolic Planning

自主性方面的创新不断催生出能够自主感知、学习、决策和行动的系统。许多公司现在都在制造自动驾驶汽车和医疗机器人，开发先进的自动驾驶系统已经是一个价值10亿美元的产业。这些新技术提供了监督、先进的自动化和自主仪器，并且它们能够适应不断变化的情况、知识和约束。然而，将新技术引入我们的技术和社会基础设施具有深远的影响，因此需要建立对其行为的信心，以避免潜在的危害。因此，自主智能系统的有效性和更广泛的可接受性依赖于这些系统解释其决策的能力。建立对人工智能（AI）系统的信任是人机交互的关键要求，也是实现人工智能的全面社会和工业效益的必要条件。该建议确定了建立自治系统可信度的两个关键因素：可解释性和风险意识。拟议的研究将提供新的算法和指导，以实现现实世界的应用，为控制、机器人、电子商务和医疗等各种实际应用打开可信的强化学习技术。总体而言，本研究将产生：首先，基于符号规划和分层强化学习的可解释和数据高效的分层顺序决策框架；第二，通过整合归纳逻辑规划和强化学习，构建可解释策略搜索框架；第三，改进了易于使用的风险敏感策略搜索方法（例如，无需调整多时间尺度步长）。本研究的理论贡献是通过发展一种促进和通知智能交互式学习过程的信任理论，显著改善交互式顺序决策系统中数据驱动的政策搜索。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Innovations in autonomy continue to produce systems that perceive, learn, decide, and act on their own. Many companies are now building self-driving vehicles and medical robots, and the development of advanced autonomous systems is already a billion-dollar industry. These new technologies offer oversight, advanced automation, and autonomous instruments, and they are adaptable to changing situations, knowledge, and constraints. However, introducing new technologies into our technical and social infrastructures has profound implications, and thus requires establishing confidence in their behavior to avoid potential harm. The effectiveness and broader acceptability of autonomous smart systems therefore rely on the ability of these systems to explain their decisions. Building trust in artificial intelligence (AI) systems is a critical requirement in human-robot interaction, and essential for realizing the full spectrum of societal and industrial benefits from AI. This proposal identifies two critical factors for establishing the trustworthiness of autonomous systems: explainability and risk-awareness. The proposed research will provide new algorithms and guidance to enable real-world applications, opening trustworthy reinforcement-learning techniques to a wide variety of practical applications such as control, robotics, e-commerce, and medical treatment. Overall, this research will produce, first, an explainable and data-efficient hierarchical sequential decision-making framework based on symbolic planning and hierarchical reinforcement learning; second, an explainable policy-search framework that can learn explainable policies via integrating inductive logic programming and reinforcement learning; and third, improved approaches to risk-sensitive policy search that are easy to use (for example, without the burden of tuning multi-timescale stepsizes). The theoretical contribution of this research is to significantly improve data-driven policy search in interactive sequential decision-making systems by developing a theory of trust that facilitates and informs smart interactive learning processes.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.