ML Basis for Intelligence Augmentation:Toward Personalized Modeling, Reasoning under Data-Knowledge Symbiosis, and Interpretable Interaction for AI-assisted Human Decision-making

今天人们所做的许多工作——在医疗保健、商业、科学企业和军事行动中——都是在团队中完成的。团队中的协作决策工作是一个复杂且具有挑战性的过程，需要对不同类型的信息进行集成、理解和操作。该项目旨在推进人工智能和机器学习作为智能增强（IA）工具的使用，以促进和改善临床团队的协作决策，重点是人工智能辅助诊断和治疗。调查人员的重点反映了IA在医疗保健中的实际重要性和影响，特别是在正在进行的与大流行的斗争中，医疗决策的效率、有效性和成本效益至关重要。然而，所建议的方法将适用于其他形式的内部审计和用例，如政策制定、公共卫生应对、情报和商业运作，最终促进国家健康、繁荣和福利。尽管现代机器学习研究已广泛参与解决基于广泛数据的各种模式发现和识别任务，无论是以完全自主的方式还是在基本的人类-人工智能协作环境（如众包）中，有效地增强和协助诊断，治疗，规划，物流领域的复杂协作人类决策工作仍然是一个公开的挑战。在临床决策中，对疾病的理解和治疗必须依赖于大量的知识和专业知识，并以来自不同信息来源的证据为基础，从文本（病史）到图像（x光片），再到时间序列数据（生命体征）。要理解这种多模式信息，需要临床团队之间的有效沟通和协作。研究人员建议研究人工智能机器学习中的一些关键技术挑战：(1)为人类决策过程建模；(2)将背景知识纳入数据驱动系统；(3)构建人机界面，促进团队间和团队内的高效协作。为此，研究人员将：(1)开发一个基于个体决策者建模的机器学习框架，该框架能够准确检测医疗诊断中的错误，并可用作协作决策设置中的推荐引擎；(2)制定有原则的策略，将客观医学知识（例如，从快速增长的医学文献中自动提取的知识）与卫生专业人员团队的临床经验和专业知识相结合；(3)设计人类可解释的界面，使团队内部和团队之间的决策有效沟通，包括解释模型如何达到每个建议决策的新工具，以及促进人类与人工智能协作的自然语言界面。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。

Much of the work people do today—in healthcare, business, scientific enterprises, and military operations—is performed in teams. Collaborative decision-making effort within a team is a complex and challenging process of integrating, understanding, and acting upon different types of information. This project aims to advance the use of artificial intelligence and machine learning as intelligence augmentation (IA) tools for facilitating and improving collaborative decision making in clinical teams, focusing on AI-assisted diagnosis and treatment. The focus of the investigators reflects the practical importance and impact of IA in healthcare, especially in the on-going fight with the pandemic where efficiency, validity, and cost-effectiveness of medical decision-making is critical. However, the proposed methods will apply to other forms and use-cases of IA, such as policy making, public health responses, intelligence and business operations, ultimately advancing national health, prosperity, and welfare.Although modern machine learning research has been widely involved in solving various pattern discovery and recognition tasks based on a wide spectrum of data—either in a fully autonomous fashion or in rudimentary human-AI collaborative settings such as crowdsourcing—effectively augmenting and assisting complex collaborative human decision-making efforts in the space of diagnosis, treatment, planning, logistics remains to be an open challenge. In clinical decision- making, understanding and treating the disease must rely on the vast knowledge and expertise and be based on evidence coming from heterogeneous sources of information, ranging from text (medical history), to imagery (radiograms), to time series data (vitals). Making sense of such multimodal information requires effective communication and collaboration within clinical teams. The investigators propose to study some of the key technical challenges in machine learning for IA: (1) modeling human decision-making processes; (2) incorporating background knowledge into data-driven systems; and (3) building human-AI interface for productive inter- and intra-team collaboration. To that end, the investigators will: (1) develop a machine learning framework based on modeling individual decision-makers that enables accurate detection of errors in medical diagnosis and can be used as a recommendation engine in collaborative decision-making settings; (2) develop principled strategies for integrating objective medical knowledge (e.g., automatically extracted from rapidly growing medical literature) with the clinical experience and expertise of a team of health professionals; (3) design human-interpretable interfaces that enable efficient communication in decision making within and across teams, including new tools for interpreting how the models arrived at each recommended decision and natural language interfaces that can facilitate human-AI collaboration.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.