strategic voting , level-k thinking , text as data , GPT , algorithm-in-the-loop systems
Abstract:
The dissertation examines AI's role in experimental economics through three distinct chapters. Chapter 1 introduces a level-k model to propose an alternative theoretical framework for understanding strategic voting behavior in juries, aiming to clarify the discrepancy between Nash equilibrium predictions and empirical findings. This chapter demonstrates the value of jury diversity in enhancing the informativeness of jury decisions. Additionally, GPT-4 is used to verify the robustness of research assistants' classifications of jurors' reasoning processes. Chapter 2 evaluates the performance of GPT-3.5 and GPT-4, employing optimal prompting structure and fundamental prompting methods across four datasets to classify promises versus empty talk and assess strategic reasoning levels. Findings indicate that GPT-4 consistently outperforms GPT-3.5, offering a reliable, scalable alternative to human annotation with effective prompting techniques. Chapter 3 investigates the effects of increased autonomy for human decision-makers in a human-AI team decision environment. Results show that making algorithmic advice optional, on average, reduces its utilization, while providing accuracy information helps counteract biased prior near-perfect expectations of algorithmic accuracy. However, this desired debiasing effect is nullified when algorithmic advice is actively solicited.
Dieser Eintrag ist Teil der Universitätsbibliographie.
Das Dokument wird vom Publikationsserver der Universitätsbibliothek Mannheim bereitgestellt.
![[img]](https://madoc.bib.uni-mannheim.de/style/images/fileicons/application_pdf.png)
Search Authors in