Publication database

Artificial intelligence systems are increasingly demonstrating their capacity to make better predictions than human experts. Yet, recent studies suggest that professionals sometimes doubt the quality of these systems and overrule machine based prescriptions. This paper explores the extent to which a decision maker (DM) supervising a machine to make high-stake decisions can properly assess whether the machine produces better recommendations. To that end, we study a set-up in which a machine performs repeated decision tasks (e.g., whether to perform a biopsy) under the DM’s supervision. Because stakes are high, the DM primarily focuses on making the best choice for the task at hand. Nonetheless, as the DM observes the correctness of the machine’s prescriptions across tasks, she updates her belief about the machine. However, the DM is subject to a so-called verification bias such that the DM verifies the machine’s correctness and updates her belief accordingly only if she ultimately decides to act on the task. In this set-up, we characterize the evolution of the DM’s belief and overruling decisions over time. We identify situations under which the DM hesitates forever whether the machine is better, i.e., she never fully ignores but regularly overrules it. Moreover, the DM sometimes wrongly believes with positive probability that the machine is better. We fully characterize the conditions under which these learning failures occur and explore how mistrusting the machine affects them. These findings provide a novel explanation for human-machine complementarity and suggest guidelines on the decision to fully adopt or reject a machine.

The rapid adoption of AI technologies by many organizations has recently raised concerns that AI may even-tually replace humans in certain tasks. In fact, when used in collaboration, machines can significantly enhancethe complementary strengths of humans. Indeed, because of their immense computing power, machines canperform specific tasks with incredible accuracy. In contrast, human decision-makers (DM) are flexible andadaptive but constrained by their limited cognitive capacity. This paper investigates how machine-basedpredictions may affect the decision process and outcomes of a human DM. We study the impact of thesepredictions on decision accuracy, the propensity and nature of decision errors as well as the DM’s cognitiveefforts. To account for both flexibility and limited cognitive capacity, we model the human decision-makingprocess in a rational inattention framework. In this setup, the machine provides the DM with accurate butsometimes incomplete information at no cognitive cost. We fully characterize the impact of machine inputon the human decision process in this framework. We show that machine input always improves the overallaccuracy of human decisions, but may nonetheless increase the propensity of certain types of errors (such asfalse positives). The machine can also induce the human to exert more cognitive efforts, even though its inputis highly accurate. Interestingly, this happens when the DM is most cognitively constrained, for instance,because of time pressure or multitasking. Synthesizing these results, we pinpoint the decision environmentsin which human-machine collaboration is likely to be most beneficial. Our main insights hold for differentinformation and reward structures, and when the DM mistrust the machine.

Markush structures are molecular skeletons containing not only specific atoms but also placeholders to represent broad sets of chemical (sub)structures. As genus claims, they allow a vast number of compounds to be claimed in a patent application without having to specify every single chemical entity. While Markush structures raise important questions regarding the functioning of the patent system, innovation researchers have been surprisingly silent on the topic. This paper summarizes the ongoing policy debate about Markush structures and provides first empirical insights into how Markush structures are used in patent documents in the pharmaceutical industry and how they affect important outcomes in the patent prosecution process. While not causing frictions in the patent prosecution process, patent documents con-taining Markush structures have an increased likelihood to restrict the patentability of follow-on inventions and to facilitate the construction of broad patent fences.

The literature on quality-management standards has generally focused on the drivers, motivations, and performance effects of adopting such standards. Yet the last decade has witnessed a substantial degree of decertification behavior, as organizations have increasingly decided to voluntarily withdraw from quality-management standards by not recertifying. While the drivers of the decision to initially adopt quality-management standards have been extensively studied, the drivers of the decision to decertify have received scant scholarly attention. We argue that innovative organizations are generally prone to retaining quality-management certification and thus exhibit a tendency to not abandon certification; however, radically-innovative organizations are more prone than incrementally-innovative organizations to discontinue quality-management standards and thereby exhibit a tendency to withdraw from quality certification. We compile World Bank data surveying facilities based in 50 countries and 103 industrial sectors across the 2003 to 2017 period. Taking advantage of the data’s panel properties yields a dataset composed of up to 1,755 facility-level observations of recertification decisions for empirical analysis. Our empirical testing employs a probit estimation technique that accounts for the appropriate fixed effects and generates results that support our theoretical priors regarding decertification behavior.

Anscombe & Aumann (1963) offer a definition of subjective probability in terms of comparisons with objective probabilities. That definition - which has provided the basis for much of the succeeding work on subjective probability - presumes that the subjective probability of an event is independent of the prize consequences of that event, a property we term Prize Independence. We design experiments to test Prize Independence and find that a large fraction of our subjects violate it; thus, they do not have subjective probabilities. These findings raise questions about the empirical relevance of much of the literature on subjective probability.

To study the effect of congestion on the fundamental trade-off between diagnostic accuracy and speed, we empirically test the predictions of a formal sequential testing model in a setting where the gathering of additional information can improve diagnostic accuracy, but may also take time and increase congestion as a result. The efficient management of such systems requires a careful balance of congestion-sensitive stopping rules. These include diagnoses made based on very little or no diagnostic information, and the stopping of diagnostic processes while waiting for information. We test these rules under controlled laboratory conditions, and link the observed biases to system dynamics and performance. Our data shows that decision makers (DMs) stop diagnostic processes too quickly at low congestion levels where information acquisition is relatively cheap. But they fail to stop quickly enough when increasing congestion requires the DM to diagnose without testing, or diagnose while waiting for test results. Essentially, DMs are insufficiently sensitive to congestion. As a result of these behavioral patterns, DMs manage the system with both lower-than-optimal diagnostic accuracy and higher-than-optimal congestion cost, underperforming on both sides of the accuracy/speed trade-off.