Misaligned by Design: Incentive Failures in Machine Learning PDF
Experiments show aligning AI training with users’ goals can weaken learning and yield worse decisions than post-training adjustments.
Daniel Martin
Wilcox Family Chair in Entrepreneurial Economics, UCSB Economics
I study attention and perception (how information is processed) and information disclosure (how information is communicated). My current research explores how human and AI interactions are shaped by attention, perception, and information disclosure.
Current Research: Humans and AI
Working Papers
Learning from an Unknown DGP: Experimental Evidence on Belief Updating with AI Recommendations PDF
We experimentally study belief updating from AI recommendations when the data-generating process is unknown.
Human Responses to AI Oversight: Evidence from Centre Court PDF
Using tennis data, we find that AI review improves umpires’ accuracy on close calls and shifts errors toward less disruptive mistakes.
When an AI Judges Your Work: The Hidden Costs of Algorithmic Assessment PDF
In an online work experiment, we find that AI evaluation increases output but lowers quality after accounting for output quantity.
When Crowds Miss Rare Events: Prevalence Bias in Human Annotation and Machine Learning
We run a field experiment on DiagnosUs to study the prevalence effect in AI labeling.
Published Papers
The ABCs of Who Benefits from Working with AI: Ability, Beliefs, and Calibration PDF
In a controlled experiment, we find that AI assistance benefits people most when they have lower baseline ability and well-calibrated beliefs.
Harnessing Human Uncertainty to Train More Accurate and Aligned AI Systems PDF
We show that training AI on bias-corrected human probability judgments improves accuracy and alignment with expert judgments.
Improving Crowdsourcing for AI through Cognitive-Inspired Data Engineering PDF
In medical-image experiments, we show that correcting annotators’ judgment biases improves both labels and trained AI models.
Modeling Machine Learning: A Cognitive Economic Approach PDF
By varying a medical-image classifier’s training incentives, we find behavior consistent with costly learning rather than fixed capacity.
Perceptions and Detection of AI Use in Manuscript Preparation for Academic Journals PDF
We survey academics on AI disclosure norms and test whether AI detectors identify GPT-modified abstracts.
Other Working Papers
Working papers not directly related to humans and AI.
Due Diligence: Endogenous Information Acquisition and Offer Quality in Search and Matching PDF
Using a search-and-matching model, we show that improving low-quality offers can discourage due diligence and leave more searchers unmatched and worse off.
Testing Capacity-Constrained Learning PDF
Applying a new theoretical test to existing perceptual experiments, we find evidence that stronger incentives often expand attention beyond fixed capacity limits.
Selected Publications
A selected set of papers on attention, perception, and information disclosure.
-
Search and Satisficing PDF
American Economic Review 101 (7), 2899-2922, 2011
-
Is No News (Perceived As) Bad News? An Experimental Investigation of Information Disclosure PDF
American Economic Journal: Microeconomics 13 (2), 141-73, 2021
-
Comparison of Decisions under Unknown Experiments PDF
Journal of Political Economy 129 (11), 3185-3205, 2021
-
Complex Disclosure PDF
Management Science 68 (5), 3236-3261, 2022
-
A Testable Theory of Imperfect Perception PDF
Economic Journal 125 (582), 184-202, 2015
Teaching and Background
Before receiving a PhD in Economics from NYU, I co-founded a small business that is now one of the leading providers of IT services to small and medium-sized businesses in the Carolinas.
At UCSB I teach a seminar course on entrepreneurship and a lecture class on behavioral economics. I also teach PhD classes on behavioral economics and attention and perception.
View courses