Searcharxiv⌕ Search

arXiv subjects

Oguzhan Celebi

Publications and source records attributed to Oguzhan Celebi.

5 recordsLinked to original sources

Instructor-designed AI improves learning in a randomized trial across 19 universities

Universities are alarmed by recent evidence that artificial intelligence (AI) can undermine learning, with students prescribed generic chatbots or centrally designed AI tools performing worse. In a pre-registered randomized controlled trial across 7,200 students at 19 universities over three semesters, we tested whether AI designed by frontline instructors can improve learning. We encouraged instructors to design AI solutions they believed would best help their students. Drawing on their unique knowledge of their classes and students, the instructors created a wide array of innovations. We show that these instructor-designed AI innovations improved student learning, measured by AI-proof assessments (proctored, in-person examinations). We found no change in student engagement, measured by course activity records, or satisfaction, measured by course evaluations. Our findings demonstrate that AI can improve student learning when its adoption is frontline-driven, with instructors empowered to design and refine AI solutions for their own classrooms.

cs.AI↗

Representation in District-Based Elections

District-based elections must reconcile two objectives: representing each district's electorate and matching the statewide delegation to parties' vote shares. We study seat allocation rules when these objectives conflict. The optimal rules, characterized by a cutoff, form a family ranging from first-past-the-post to proportional representation as the weight on statewide representation increases. Along this path, gaining further seats through gerrymandering becomes increasingly difficult. The characterization yields a measure of the robustness of first-past-the-post to statewide representation concerns, which we compare with established electoral diagnostics. In U.S. House elections from 2002 to 2024, we find that maps drawn by single-party legislatures perform worse than those produced by courts or non-partisan commissions. The efficiency gap, a common measure of partisan gerrymandering, is more strongly associated with litigated maps than the other measures, yet magnifies the extent to which Republican-legislature maps appear worse than Democratic-legislature maps.

econ.TH↗

Diversity Preferences, Affirmative Action and Choice Rules

I study the relationship between diversity preferences and the choice rules implemented by institutions, with a particular focus on the affirmative action policies. I characterize the choice rules that can be rationalized by diversity preferences and demonstrate that the recently rescinded affirmative action mechanism used to allocate government positions in India cannot be rationalized. I show that if institutions evaluate diversity without considering intersectionality of identities, their choices cannot satisfy the crucial substitutes condition. I characterize choice rules that satisfy the substitutes condition and are rationalizable by preferences that are separable in diversity and match quality domains.

econ.TH↗

Adaptive Priority Mechanisms

How should authorities that care about match quality and diversity allocate resources when they are uncertain about the market? We introduce adaptive priority mechanisms (APM) that prioritize agents based on both their scores and characteristics. We derive an APM that is optimal and show that the ubiquitous priority and quota mechanisms are optimal if and only if the authority is risk-neutral or extremely risk-averse over diversity, respectively. With many authorities, each authority using the optimal APM is dominant and implements the unique stable matching. Using Chicago Public Schools data, we find that the gains from adopting APM may be considerable.

econ.TH↗

Substitutability in Favor Exchange

I introduce a favor exchange model where favors are substitutable and study bilateral enforcement of cooperation. Without substitutability, the value of a relationship does not depend on the rest of the network, and in equilibrium there is either no cooperation or universal cooperation. When favors are substitutable, each additional relationship is less valuable than the previous, and intermediate levels of cooperation are observed. I extend the model to allow for transfers, heterogeneous players, and multilateral enforcement. My results can explain the stratification of social networks in post-Soviet states and the adoption of different enforcement mechanisms by different groups of medieval traders.

econ.TH↗