About Me

Crowdwork platforms are central to the data supply chains that power AI systems and sustain large-scale academic and commercial research. Yet worker engagement dynamics remain poorly understood. We study how early procedural frictions shape retention using approximately 64 million task submissions from Prolific. Focusing on new workers’ first fourteen days, we examine two frictions: task returns, reflecting study design failures, and duration underestimation, reflecting inaccurate platform information. Using instrumental variables to address endogeneity, we find that a 10 percentage point increase in returned tasks reduces retention by 3.7 percentage points, while the same increase in underestimated tasks reduces retention by 9.9 percentage points. Both frictions operate primarily through attrition rather than demotivation among surviving workers. These findings suggest that process reliability and information accuracy are critical levers for sustaining the engaged workforce that platforms need to function.

The vast quantities of data required to build artificial intelligence (AI) technologies are often annotated and processed manually, making human labor a critical component of the AI supply chain. The workers who input this data are sourced through digital labor (crowdwork) platforms that often are unregulated and offer low wages, raising concerns about labor standards in AI development. Using the results of a survey, this article aims to shed light on the experiences and perceptions of fair treatment among workers in the AI supply chain. The study reveals significant variability in workers’ experiences, identifies potential drivers of fairness, and highlights how design choices by labor platforms can significantly affect worker welfare. Drawing on lessons from physical supply chains, this article offers practical guidance to managers on how to enhance worker welfare within the AI supply chain and how to ensure that AI technologies are responsibly sourced.

Serendipity would seem to preclude purposeful search. To understand the relationship between search behavior and the probability of a serendipitous discovery, we propose a formal modeling framework based on the NK model. Our simulations suggest that searchers with a clear hypothesis convert good fortune into fitness or value far more effectively. Searchers that pursue their theories through incremental steps instead of long-distance jumps have the highest rates of serendipitous success. Our model also provides insight into the role of bias, inaccurate or incorrect beliefs, and the ruggedness of the terrain in determining the rate of serendipitous discovery. Our results demonstrate the distinct roles of intent and information in finding novel, valuable discoveries.

How do professionals respond to computerized, data-driven guidance in practice? We analyze a workers’ compensation insurance program where physicians make coverage and diagnosis decisions. We study the introduction of an automated system that flagged diagnoses with historically low coverage. We develop a model that yields testable predictions to distinguish between informational and persuasive effects. Consistent with persuasion, physicians granted coverage less often when confronted with alerts, but they also avoided alerts by recoding diagnoses. Data from secondary reviews show that the system aligned physicians’ decisions with management’s preferences. These findings provide lessons for the design of information systems for decision-makers.