Empathetic, interpretable, and explainable Artificial Intelligence

本研究的首要目的是研究在高风险环境下，移情人工智能（AI）对人类-人工智能团队的影响。这项研究的结果可以帮助理解如何优化人类与人工智能的互动，以在各种高风险领域取得更好的结果，包括军事行动、航空、医疗保健和法律等。与该项目的合作伙伴泰雷兹公司相关的一个具体用例是潜艇上的高压操作，声纳操作员面临着分析大量复杂数据集的挑战。该项目旨在探索如何在这种情况下整合移情人工智能，以减轻声纳操作员的压力，同时提高团队的表现。主要假设是，一个具有同理心的人工智能，能够适应人类操作员的压力源和需求，将促进人类与人工智能之间更牢固的关系，从而提高团队的整体效率。此外，这种共情关系不仅会使人类与人工智能建立融洽的关系，还会促进一种互惠的理解，在这种理解中，人类操作员认识到并尊重人工智能的局限性和需求。据推测，这种双向共情将导致团队绩效的提高和人类操作员认知负荷的减少。该研究将利用一个严肃的游戏环境来测试所提出的假设。这种可控但生态有效的环境将允许对移情人工智能对人类操作员的心理影响进行彻底调查。该实验旨在评估作业人员在三种不同情况下的表现和压力水平：2.无人工智能辅助操作；2 .使用常规的、无同理心的人工智能进行操作；用有同理心的人工智能操作。对这些场景的比较分析将为同理心在提高人类-人工智能团队绩效方面的具体作用提供有价值的见解。这项研究的结果与UKRI的使命直接一致，因为将移情人工智能整合到人类人工智能团队中，将通过优化团队绩效为繁荣做出贡献，并通过减少操作员压力和培养更好的运营实践为公共利益做出贡献。这些理想得到了支持该项目的博士培训中心（CDT） ART-AI（负责任、负责任和透明的人工智能）的进一步回应，该中心致力于开发维护这些价值观的人工智能系统。

The overarching aim of this research is to examine the impact of empathetic artificial intelligence (AI) on human-AI teaming in high-stakes contexts. Findings from this research could assist in understanding how human-AI interactions can be optimised for better outcomes in various high-stakes sectors, including military operations, aviation, healthcare, and law, among others. A specific use case pertinent to the collaborative partner of this project, Thales, is the high-pressure operations aboard submarines, where sonar operators face the challenge of analysing vast and complex data sets. This project aims to explore how integrating empathetic AI in such scenarios could alleviate the stress experienced by sonar operators while also improving the performance of the team.The primary hypothesis is that an empathetic AI, attuned to the stressors and needs of human operators, will foster a stronger human-AI relationship, thereby enhancing the overall efficacy of the team. Moreover, this empathetic relationship will not only condition humans to build rapport with the AI but will also foster a reciprocal understanding where the human operator recognises and respects the limitations and needs of the AI. This bidirectional empathy, it is hypothesised, will lead to improved team performance and a reduction in the cognitive load of the human operator.The research will utilise a serious game environment to test the proposed hypotheses. This controlled, yet ecologically valid, setting will allow for a thorough investigation of the psychological impacts of empathetic AI on human operators. The experiment will be designed to assess operator performance and stress levels in three distinct scenarios:1. Operation without any AI assistance,2. Operation with a regular, non-empathetic AI, and3. Operation with an empathetic AI.The comparative analysis of these scenarios will provide valuable insights into the specific role of empathy in enhancing human-AI team performance.The findings of this research directly align with the UKRI's mission, as the integration of empathetic AI in human-AI teams is poised to contribute to prosperity by optimising team performance, as well as to the public good by reducing operator stress and cultivating better operational practices. These ideals are further echoed by the Centre for Doctoral Training (CDT) underpinning this project, ART-AI (Accountable, Responsible, and Transparent AI), which is committed to developing AI systems that uphold these values.