Can Çelebi

joint with Andria Smith, Hunter P. van Wagoner, and Ksenia Keplinger

Teams that combine human intelligence with artificial intelligence (AI) have become indispensable for solving complex tasks in various decision-making contexts in modern organizations. However, the factors that contribute to AI convergence, where human team members align their decisions with those of their AI counterparts, still remain unclear. This study integrates signaling theory with self-determination theory to investigate how specific signals—such as signal fit, optional AI advice, and signal set congruence—affect employees' AI convergence in human–AI teams. Based on four experimental studies conducted in facial recognition and hiring contexts with approximately 1100 participants, the findings highlight the significant positive impact of congruent signals from both human and AI team members on AI convergence. Moreover, providing an option for employees to solicit AI advice also enhances AI convergence; when AI signals are chosen by employees rather than forced upon them, participants are more likely to accept AI advice. This research advances knowledge on human–AI teaming by (1) expanding signaling theory into the human–AI team context; (2) developing a deeper understanding of AI convergence and its drivers in human–AI teams; (3) providing actionable insights for designing teams and tasks to optimize decision-making in high-stakes, uncertain environments; and (4) introducing facial recognition as an innovative context for human–AI teaming.

Link

joint with Stefano Balietti and David Tercero-Lucas

The objective of this paper is to explore whether NFT investors represent a distinct cohort within the broader crypto investment sphere. Employing data from a public survey with global outreach, we first find that NFT owners are younger and possess, on average, a lower educational level than the general crypto population but a higher cryptocurrency knowledge. Second, there are no significant gender differences among NFT investors and non-NFT investors, but those working in the crypto sphere are more likely to invest in NFTs. Additionally, individuals involved in yield farming or using crypto derivatives are more likely to own NFTs. Finally, we show that individuals with more concerns about the potential misuse of cryptocurrency for illicit activities are less likely to engage in the ownership of NFTs.

Link

joint with Stefan P. Penczynski

This study explores how communication within teams shapes individual reasoning and decision-making in strategic environments. We build on a structured intra-team communication protocol that allows for clean attribution of reasoning to individual participants by capturing their initial suggestions, exchanged messages, and final decisions. By analyzing this communication, we identify persuasion mechanisms and strategic thinking processes within teams. The study also compares this method to alternatives like "thinking aloud" and advice-giving, and highlights the advantages of using large language models for classifying participants' reasoning. Our results show that LLMs such as GPT-4 can reliably replicate human annotation for classifying levels of strategic sophistication, offering a scalable and replicable tool for analyzing reasoning in behavioral experiments.

joint with Christine Exley, Sören Harrs, Hannu Kivimaki, Marta Serra-Garcia, Jeffrey Yusof

High-quality data is essential to social science research. Online experiments and surveys are a central tool for data collection across many disciplines, but their data quality could be lacking, due to the presence of automated agents, participants who use LLMs without researcher knowledge, and inattentive participants. We identify behavioral patterns that can serve as data quality checks by collecting data from human subjects in the lab, automated agents, and online survey platforms. We further propose the two-stage recruitment method by which researchers first implement a short survey on their target sample and use checks to exclude plausibly low-quality responses. We test the method with a set of checks and demonstrate how data quality can improve with this method.

Link

joint with Stefan P. Penczynski

This study examines the use of large language models (LLMs) for text classification. We investigate whether original instructions can be effectively repurposed as prompts with moderate changes to achieve classification results comparable to human-coded benchmarks. Additionally, we study the impact of two prompting techniques—providing a set of n classified examples (n-shot) and requiring a justification explanation (zero-shot Chain-of-Thought)—on classification performance. Using GPT-3.5 and GPT-4, we further examine the extent to which larger model size improves classification accuracy. To assess these factors, we classify text from four economic experiments, covering tasks with varying complexity and prevalence in pre-training data, providing insights into how task characteristics influence classification performance. We find that LLMs can accurately and cost-effectively classify text across these tasks and replicate human annotations well. Performance improves through n-shot prompting and we observe task-dependent gains from Chain-of-Thought prompting. Our findings offer guidance for integrating LLM-based text classification into social science research.

Link

joint with Stefan P. Penczynski

Theoretical work by Feddersen and Pesendorfer (1998) has shown how strategic voting undermines the intuition that unanimous voting eliminates convictions of innocent defendants. We set up a level-k model of jury voting and experimentally investigate strategic thinking with an experimental design that uses intra-team communication. Looking at juries using the unanimity rule, we show that the jury performance depends on the strategic sophistication of jury members, which in turn depends on the complexity of the task at hand.

Link

joint with Stefano Balietti, David Tercero-Lucas, Luca Pennella

The objective of this paper is to examine whether memecoin investors constitute a distinct subgroup within the broader population of cryptocurrency holders. Using data from the 2024–25 State of Crypto Survey, a global, multi-language online survey, we restrict attention to respondents who own or have owned cryptocurrencies and study the determinants of memecoin ownership with logistic regressions. We relate holding at least one memecoin to socio-economic characteristics, risk preferences, investment behavior, and attitudes toward institutions and crypto markets. We find that memecoin investors differ systematically from other crypto users. They are predominantly male, more actively engaged in trading, and more likely to use leverage, derivatives, and yield farming, while allocating a larger share of their wealth to crypto and holding a broader set of tokens, including NFTs. Specific knowledge about memecoins strongly predicts investment, whereas age and general financial literacy do not. Memecoin investors exhibit lower sensitivity to taxation, weak trust in government, and a psychological profile characterized by a low need to belong and greater novelty-seeking. These results portray memecoin investors as a behaviorally and demographically distinct group within the crypto ecosystem and suggest that policy interventions—such as enhanced disclosure, platform-specific investor protection, and coordinated regulation—may be needed to address risks concentrated in this segment.

Link

joint with Stefan P. Penczynski

joint with Pawel Niszczota

joint with Stefan P. Penczynski

joint with Christian Koch, Stefan P. Penczynski

joint with Pawel Niszczota

joint with Christian Koch, Stefan P. Penczynski