PhD position Decision-Making in Generative AI: Bias, Rationality, and Benchmarks

Since its foundation in 1614, the University of Groningen has enjoyed an international reputation as a dynamic and innovative center of higher education offering high-quality teaching and research. Belonging to the best research universities of Europe and joining forces with prestigious partner universities and networks, the University of Groningen is truly an international place of knowledge.



Faculty of Economics and Business

The Faculty of Economics and Business offers an inspiring study and working environment for students and employees. International accreditation enables the Faculty to assess performance against the highest international standards. It also creates an exciting environment of continuous improvement. FEB's programmes, academic staff and research do well on various excellence ranking lists.



FEBRI, the graduate school and research institute of the Faculty of Economics and Business has available

One PhD position in the field of Decision-Making in Generative AI: Bias, Rationality, and Benchmarks



Project description

Generative AI systems are increasingly being considered for decision-making roles, yet their rationality, coherence, and susceptibility to biases remain poorly understood. This project aims to develop benchmarks that assess AI decision-making against normative principles, identifying both human-like biases and novel systematic incongruities.



AI should not only satisfy fundamental axioms of rational choice theory but also demonstrate consistency across different formulations of the same decision problem. Just as humans exhibit framing effects, AI models may reach contradictory conclusions depending on how a prompt is phrased. Similarly, they may display preference reversals, nonstationary risk attitudes, violations of dynamic consistency, or path dependence.



For instance, an AI providing financial advice might recommend different actions depending on whether an investment is framed as a potential gain versus an avoided loss. A language model generating ethical judgments could shift its stance depending on wording variations. AI assistants in strategic decision-making might even open avenues for adversarial exploitation.



By systematically evaluating these issues, this project seeks to establish a rigorous framework for assessing AI decision-making. It will benchmark generative AI models, comparing their decision-theoretic performance much like existing evaluations of their mathematical problem-solving, reasoning, or factual accuracy. The findings will contribute to the development of AI systems that make decisions that are coherent, reliable, aligned with principles of rationality, while also reflecting the preferences of their principals.



Organisation

The PhD position is embedded in the Economics, Econometrics, and Finance research programme of FEB’s Research Institute. The project will be supervised by Zsombor Méder and Pim Heijnen.

Requirements:

Candidates should hold an MSc in economics, computer science, artificial intelligence, or a related field. They must have a strong grasp of decision theory and a keen interest in applying it to AI evaluation.



Proficiency in coding is essential. Ideally, the candidate has experience in behavioral economics and AI benchmarking. A background in textual analysis and an interest in AI alignment or ethics may also prove useful.



As this is an emerging research area with little prior work, candidates will be expected to formulate original research questions and develop novel methodologies. The ideal candidate should be comfortable working at the intersection of economics and artificial intelligence, demonstrating intellectual independence and the ability to critically assess both the limitations and potential of AI.

Salary Benefits:

The full-time appointment is temporary for a period of four years, under the condition of a positive assessment at the end of the first year. The gross monthly salary is € 2,901 which increases yearly to a maximum of € 3,707. Based on the gross salary you will receive an additional 8% holiday allowance each year in May, and an additional 8.3 % end of year bonus in December. PhD candidates contribute 20% of their time to teaching in years 2-4.



Starting date: Preferably per September 1, 2025.



An assessment may be part of the procedure, consisting of psychological tests and an interview.

38 hours per week

Broerstraat 5