It takes two to tango: a synergistic approach to human-machine decision making

人工智能(AI)在增强人类决策、改善认知过载和降低高风险场景中的偏见方面拥有巨大的潜力。然而，在这类应用程序中采用基于人工智能的支持系统的情况很少，主要是因为很难评估它们的假设、限制和意图。为了实现人工智能对个人、社会和经济的承诺，人们应该觉得他们可以信任人工智能，在可靠性、理解人类需求的能力方面，并保证它们是真正旨在帮助他们的。Tango将为混合决策支持系统(HDSS)开发理论基础和计算框架，在HDSS中，人和机器在价值和目标方面保持一致，了解各自的优势，并共同努力达成最优决策。为此，探戈将发展：1)相互理解和混合决策的认知理论，直觉和深思熟虑的决策方法，以及它们如何影响我们对人类和人工智能队友的信任。2)认知感知可解释人工智能实现协同的人机交互，使机器能够确定特定决策者(例如，外行与专家)需要或不需要什么信息来做出知情决策。3)建立在双向、解释增强的对话基础上的人类决策和机器学习模型的“人在环中”的共同进化。将根据四个影响较大的使用案例对探戈框架进行评估，即：支持：1)孕期和产后妇女；2)手术团队在术中决策中的作用；3)信贷员和申请者在信贷决策过程中的作用；4)公共政策制定者在设计激励措施和分配资金方面的支持。这些案例研究的成功将使探戈成为开发新一代协同人工智能系统的参考框架，并将加强欧洲在以人类为中心的人工智能领域的领导地位

Artificial Intelligence (AI) holds enormous potential for enhancing human decisions, improving cognitive overload and lowering bias in high-stakes scenarios. Adoption of AI-based support systems in such applications is however minimal, chiefly due to the difficulty of assessing their assumptions, limitations and intentions. In order to realise the promise of AI for individuals, society and economy, people should feel they can trust AIs in terms of reliability, capacity to understand the human’s needs, and guarantees that they are genuinely aiming at helping them. TANGO will develop the theoretical basis and computational framework for hybrid decision support systems (HDSS) in which humans and machines are aligned in terms of values and goals, know their respective strengths, and work together to reach an optimal decision. To this end, TANGO will develop: 1) A cognitive theory of mutual understanding and hybrid decision making, of intuitive vs deliberative approaches to decision making and of how they affect our trust in human and AI teammates. 2) Cognition aware explainable AIsimplementing synergistic human-machine interaction, enabling machinesto determine what information a specific decision maker (e.g., layperson vs expert) needs, or does not need, to reach an informed decision. 3) A “Human-in-the-loop” co-evolution of human decision making and machine learning models building on bi-directional, explanation-augmented interlocution. The TANGO framework will be evaluated on four high impact use cases, namely supporting: i) women during pregnancy and postpartum, ii) surgical teamsin intraoperative decision making, iii) loan officers and applicantsin credit lending decision processes, and iv) public policy makers in designing incentives and allocating funds. Success in these case studies will establish TANGO as the framework of reference for developing a new generation of synergistic AI systems, and will strengthen the leadership of Europe in human-centric AI