Research

Research Interests

Primary: Gender and Gender in Language, Economics of Artificial Intelligence and Robotics

Secondary: Experimental Methodology, Networks, Contest

Publications

Working Papers

Regulators of artificial intelligence (AI) emphasize the importance of human autonomy and oversight in AI-assisted decision-making (European Commission, Directorate General for Communications Networks, Content and Technology, 2021; 117th Congress, 2022). Predictions are the foundation of all AI tools; thus, if AI can predict our decisions, how might these predictions influence our ultimate choices? We examine how salient, personalized AI predictions affect decision outcomes and investigate the role of reactance, i.e., an adverse reaction to a perceived reduction in individual freedom. We trained an AI tool on previous dictator game decisions to generate personalized predictions of dictators’ choices. In our AI treatment, dictators received this prediction before deciding. In a treatment involving human oversight, the decision of whether participants in our experiment were provided with the AI prediction was made by a previous participant (a ‘human overseer’). In the baseline, participants did not receive the prediction. We find that participants sent less to the recipient when they received a personalized prediction but the strongest reduction occurred when the AI’s prediction was intentionally not shared by the human overseer. Our findings underscore the importance of considering human reactions to AI predictions in assessing the accuracy and impact of these tools as well as the potential adverse effects of human oversight.


Social norms, though often implicit, are to a great extent communicated and made salient using natural language. They carry the notions that "the participant," "the customer," or "the worker" should behave in a certain way. In English, we refer to each of these personal entity nouns using the pronouns "he," "she," or the gender-inclusive singular "they." In languages with grammatical gender, the nouns and the grammatical structure they are embedded in mark them as either male, female, or gender-inclusive. Little is known as to whether the framing of norms with respect to these grammatical genders affects norm compliance. We conducted an experiment in German with three games commonly used to study fair sharing, cooperation, and honesty. Our treatments allowed us to compare the differences in the increase of norm compliance when introducing prescriptive norms depending on the match between the participant's self-reported gender and the gender frame used in the experimental instructions. Overall, we find no strong evidence that a match between the participant's self-reported gender and the norm formulation led to a higher increase in norm compliance compared to the differences in a mismatch or gender-inclusive frame. We observed the strongest effect for men in the sharing game, where the data suggests that a match led to a higher increase in norm compliance compared to the increase if gender-inclusive formulations were used. This line of research has important implications for the effective communication of rules and norms in organizations and administrations.


We conducted a controlled experiment to study how different gender frames used in the instructions affect economic behavior. In our experiment, we systematically varied the framing of the instructions either using the male, the female, or a gender-inclusive form. Participants played three standard economic two-player games measuring prosocial behavior. In particular, we elicited the degree of (fair) sharing, reciprocal behavior, and honest reporting. We investigated if participants behaved differently if their self-reported gender matched the grammatical gender used in the instructions. The results reveal that the framing of instructions had the strongest impact on sharing and the effects were mainly driven by participants self-identifying as men. In contrast, we observe only mild treatment differences, if any, regarding reciprocal behavior or honest reporting. We discuss potential mechanisms and consequences of our findings.


We analyse the correlation between job satisfaction and automatability - the degree to which an occupation can be or is at risk of being replaced by computerised equipment. Using multiple survey datasets matched with various measures of automatability from the literature, we find a negative and statistically significant correlation that is robust to controlling for worker and job characteristics. Depending on the dataset, a one standard deviation increase in automatability leads to a drop in job satisfaction of about 0.64% to 2.61% for the average worker. Unlike other studies, we provide evidence that it is not the fear of losing the job that mainly drives this result, but the fact that monotonicity and low perceived meaning of the job drive both automatability and low job satisfaction.


Conference Proceedings (peer-reviewed)


Work in Progress