Collaborative Research: Design Decisions under Competition at the Edge of Bounded Rationality: Quantification, Models, and Experiments

该项目的目的是研究在团队之间竞争的情况下，信息不确定性对设计合理性的影响，并开发一个具有竞争意识的人工智能（AI）助手，用于设计空间的探索和开发。沟通问题被广泛认为是影响团队决策设计结果的关键因素，因为在沟通过程中共享的信息可能是不完整和不完美的。因此，通过理解信息不确定性在团队中人类设计决策中的作用来解决这些问题，可以更好地协调团队决策，促进人类与人工智能的合作，并在大型工程设计项目中节省大量成本。然而，尽管在团队设计决策建模方面取得了重大进展，但目前的文献忽略了设计过程的两个基本方面：人类设计师具有有限的理性，大多数设计活动都是在竞争中进行的，无论是有意识的还是无意识的。这个项目的动机是通过发展理论和实验结构来填补这一空白，以计算模拟人类设计师在竞争环境中的顺序决策，并通过实验测量他们的有限理性。预期的结果是一套新的知识，包括度量、模型、算法和测试平台，在存在不确定性的竞争下，设计团队中的人类行为。通过德克萨斯大学奥斯汀分校的新生工程研究入门（FIRE）项目和史蒂文斯理工学院的品尼高和克拉克学者项目，直接让不同的本科生参与研究活动，将产生更广泛的影响。这个项目是通过回答两个研究问题来驱动的：1)信息不确定性（例如，当团队成员之间共享的设计信息不完整时）对团队竞争下设计合理性的影响是什么？2)在更大范围的团队规模中，这种影响会有什么不同？为了回答这两个问题，计划采用一种跨学科的研究方法，将描述性、规范性和预测性分析相结合。特别是，我们将开发博弈论模型来模拟竞争下的顺序设计决策，协同整合两种类型的顺序学习模型，即理论驱动的规范模型（例如，贝叶斯优化）和数据驱动的预测模型（例如，长短期记忆单元）。这种新方法明确地模拟了设计师对对手过去表现的看法，并预测了对手未来的决策，从而提供了一种定量的方法来研究团队中设计师之间共享的不确定信息对他们与其他团队竞争决策的影响。实验将收集行为数据，研究人类的非理性行为，以理论模型预测的理性行为为基准，在设计竞赛过程中如何随时间变化。研究结果将在现实世界的太阳能系统设计挑战赛中得到验证，届时多个团队将竞争奖项。为了使更广泛的研究社区受益，该项目将建立一个开放的设计基础设施来共享项目数据和发现。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。

The objective of this project is to investigate the impact of information uncertainty on design rationality under competition among teams and to develop a competition-aware artificial intelligence (AI) assistant for design space exploration and exploitation. Communication issues have been broadly recognized as a critical factor that impacts design outcomes in team decisions, because the information shared during communication can be incomplete and imperfect. Therefore, addressing such issues by understanding the role of information uncertainty on human design decisions in teams could lead to better coordination of team decisions, advancement in human-AI collaborations, and significant cost savings in large-scale engineering design projects. However, despite the significant progress in modeling design decision-making in teams, current literature neglects two fundamental aspects of the design process: human designers have bounded rationality, and most design activities happen under competition, whether consciously or unconsciously. This project is motivated to fill this gap by developing theoretical and experimental constructs to computationally model human designers’ sequential decisions in a competitive environment and to experimentally measure their bounded rationality. The expected outcome is a suite of new knowledge, including metrics, models, algorithms, and testbeds, on human behavior in design teams under competition in the presence of uncertainties. Broader impacts will be generated by directly engaging diverse undergraduate students in research activities through the Freshman Introduction to Research in Engineering (FIRE) program at the University of Texas at Austin and the Pinnacle and Clark Scholars programs at Stevens Institute of Technology.This project is driven by answering two research questions: 1) what are the effects of information uncertainty (e.g., when design information shared between team members is incomplete) on design rationality under competition among teams? 2) How would such effects differ across a wider range of team sizes? To answer the two questions, an interdisciplinary research approach is planned that combines descriptive, prescriptive, and predictive analytics. In particular, we will develop game theoretic models to model sequential design decisions under competition, synergistically integrating two types of sequential learning models, i.e., theory-driven prescriptive models (e.g., Bayesian optimization) and data-driven predictive models (e.g., long-short term memory units). The new approach explicitly models designers’ perceptions of their opponents’ past performance and predicts the opponent’s future decisions, thereby providing a quantitative way to study the influence of uncertain information shared among designers in a team on their decisions in competition against other teams. Experiments will be conducted to collect behavioral data to study how human irrationality, benchmarking on rational behaviors predicted by the theoretical models, would change over time during the course of the design competition. The research findings will be validated in a real-world design challenge on solar system design, where multiple teams compete for awards. To benefit a broader research community, this project will build an open design infrastructure to share the project data and findings.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.