Zooming In or Panning Out? How AI Recommendations Shape Selection Decisions

Artificial intelligence (AI) is transforming how organizations evaluate innovations. AI can offer recommendations emphasizing key evaluation criteria, such as novelty and feasibility, to influence human evaluators’ selection decisions. Organizations often generate large numbers of potential solutions, requiring evaluators to decide which are worth pursuing. Typically, evaluators assess criteria in parallel, integrating AI recommendations highlighting both novelty and feasibility. Yet bounded rationality limits their ability to fully integrate all information, making it difficult to consistently identify the most promising solutions. An alternative is sequential evaluation, in which AI recommendations emphasizing different criteria are presented in stages. This may reduce cognitive load but also shape which solutions are ultimately selected. This raises a central puzzle: how do parallel versus sequential evaluation formats, guided by AI recommendations, affect the types of solutions selected in terms of novelty and feasibility? Our pilot experiment compares these formats to advance literature on human-AI evaluation.

Key facts

Principal Investigator: Cyrille Grumbach
Co-PI: Jacqueline Ng Lane
Affiliation: PhD student, ETH Zürich