CAREER: Problem Partitioning and Division of Labor for Human-Computer Collaboration in Engineering Design

这个教师早期职业发展（CAREER）研究项目的目标是了解如何将集体设计任务分配给具有不同决策特征的人类和人工智能（AI）支持的计算机代理团队将影响设计结果。这个研究项目的首要前提是，围绕分析学科（例如，热分析、结构分析）或围绕物理组件（例如，引擎、电池、外部车身）可能不是构建人工智能团队的理想方式。考虑到人类和人工智能系统在能力方面的差异，可能有其他方法可以在团队中的人类和人工成员之间划分系统级设计问题的任务和责任。这项研究将通过对混合人类-AI团队中的团队架构以及这些不同架构对设计结果的影响进行系统和全面的分析，为设计科学领域做出贡献。该项目将采用自上而下的系统工程方法，而不是为人工智能预先定义角色，以确定使用人类和人工智能决策过程的计算模型在设计团队中定义人工智能角色的最佳实践。一项使用视频游戏平台的实验研究将让人类用户与AI队友一起解决设计问题，以评估人为因素在这种情况下的影响。研究结果将告知人工智能技术应如何通过适当的任务分配整合到工程设计人员中，以减少从智能医疗到国防等多个行业的未来企业的系统开发时间和成本。该项目将通过让不同的本科生群体参与利哈伊大学STEM-SI项目的研究来产生更广泛的影响。与研究计划相结合，该教育计划将使用电子培训游戏改善工程系统应用的统计和机器学习方面的教学实践。当地社区内的外联研讨会将吸引更广泛的兴趣在数据科学通过参与中学女生在工程数据相关的挑战使用的选择程序在利哈伊大学。这项研究解决了缺乏基本原则来指导混合人工智能团队的任务划分和劳动分工的问题，不仅考虑了决策者之间的异质性（由一组选定的特征表示），还考虑了重要的人为因素。该项目将使用多智能体模拟来模拟广义代理，解决贝叶斯决策过程后的上下文无关的设计问题。这些代理人的特点是在任务性能，自信心，并在其他团队成员的信心。计算分析将使用基于分解的设计和机器学习的各种问题划分和任务分配策略来量化它们对团队协作的影响，同时考虑不同的代理特征。模仿这种模拟场景，电动汽车设计和控制游戏的行为实验将为混合人类-人工智能团队提出一个协作设计决策问题，并在受控环境中提供替代任务分配方案。这些实验将收集行为数据，捕捉人为因素的影响，包括偏见，工作量和工作满意度，并将用于验证或完善计算结果。作为一个副产品，这项研究将开发一个开放的基础设施，通过与更广泛的科学界共享实验平台，研究设计团队中的人机协作。该项目还将把为行为研究开发的视频游戏平台整合到现有的机械工程课程和数据训练营中，以教授本科生和研究生数据分析。该奖项反映了NSF的法定使命，并通过使用基金会的智力价值和更广泛的影响审查标准进行评估，被认为值得支持。

The objective of this Faculty Early Career Development (CAREER) research project is to understand how the allocation of collective design tasks into teams of humans and Artificial Intelligence (AI)-enabled computer agents with diverse decision-making characteristics will affect design outcomes. The overarching premise of this research project is that the current organization of design teams around analysis disciplines (e.g., thermal analysis, structural analysis) or around physical components (e.g., engine, battery, exterior body) may not be the ideal way to architect human-AI teams. Considering the differences between humans and AI systems in terms of their capabilities, there may be alternative ways to divide tasks and responsibilities in system-level design problems among humans and artificial members within a team. This research will contribute to the field of design science through a systematic and comprehensive analysis of team architectures in hybrid human-AI teams and the impact of those different architectures on design outcomes. Rather than assuming a pre-defined role for AI, this project will follow a top-down Systems Engineering approach to identify best practices for defining roles for AI in a design team using computational models of human and AI decision-making processes. An experimental study using a video game platform will engage human users to work alongside AI teammates in solving a design problem to assess the impact of human factors in this context as well. The findings will inform how AI technology should be integrated into the engineering design workforce with proper task allocation in order to reduce system development time and costs for future enterprises in multiple industries, spanning from smart healthcare to defense. The project will generate broader impacts by engaging a diverse group of undergraduate students into research using the STEM-SI program at Lehigh University. Integrated with the research program, the education plan will improve pedagogical practices in statistics and machine learning for engineering systems applications using e-training games. Outreach workshops within local communities will attract broader interest in data science by engaging middle school girls in data-related challenges in engineering using the CHOICES program at Lehigh University. This research addresses the lack of fundamental principles to guide task partitioning and division of labor for hybrid human-AI teams, accounting not only for heterogeneity among decision-makers (represented by a select set of characteristics), but also for important human factors. The project will use multi-agent simulations to model generalized agents that solve context-free design problems following Bayesian decision-making processes. These agents will be characterized in terms of task performance, self-confidence, and confidence in other team members. A computational analysis will use various problem partitioning and task assignment strategies from decomposition-based design and machine learning to quantify their impact on team collaboration considering diverse agent characteristics. Mirroring this simulation scenario, behavioral experiments on an electric vehicle design and control game will present a collaborative design decision-making problem for hybrid human-AI teams with alternative task allocation scenarios in a controlled setting. These experiments will collect behavioral data that capture the effects of human factors, including bias, workload, and job satisfaction, and that will be used to validate or refine the computational findings. As a by-product, this research will develop an open infrastructure to study human-AI collaboration in design teams by sharing the experimental platform with the broader scientific community. This project will also integrate video game platforms developed for the behavioral study into existing mechanical engineering courses and data bootcamps to teach undergraduate and graduate students data analytics. Outreach workshops will use the same game platforms to increase data literacy and promote STEM careers among middle school girls.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.