Modeling Expert Choice at the Technical Frontier

Commercialization of a new material or process invention can take decades. A predominance of tacit knowledge, information asymmetries, and insufficient human capital with knowledge in the field can contribute to this delay. At this technical frontier, expert decision making drives critical

research and development. By surveying the top experts in an industry that exemplifies the technical frontier, metal additive manufacturing for aerospace, I find that there is inconsistency in decision making structure and process both within and across experts. This inconsistency leads to

an investigation of how to evaluate accuracy in a field when gathering outcome data is expensive or only exists in the future (as is the case at the technical frontier). I prove bounds on expert accuracy that rely on response structure without requiring outcome data for validation. The inconsistency in expert decision process seen in the survey case study motivates the search for a single tool to learn

a diverse set of behavioral decision making models. I create a novel neural network that is able to learn some, but not all, of a variety of common theoretical behavioral choice models. This combination of qualitative and quantitative modeling of decision making at the technical frontier provides tools for speeding the development of new technologies across the "valley of death" on their way to mass commercialization.