Theme 1: Economics in Training, Alignment, and Evaluation
Using economic ideas to improve learning and alignment, explain the limitations of existing algorithms and paradigms, and align incentives in local economic interactions around AI systems.
Topics include, but are not limited to:
- Preference aggregation for alignment and its limitations
- Pricing of data, training, and inference
- Social choice and auction mechanisms for steering alignment
- Strategic behavior in model evaluation, and the design of incentive-aware evaluation
- Strategic classification
- Mechanisms for eliciting high-quality data and feedback
- Discrete choice and behavioral modeling in learning pipelines
- AI decision making and bias in economic contexts
- Algorithmic collective action
- Formal abstractions of AI rationality and bias in economic contexts
- New formal models of incentive misalignment and information gaps around AI systems
Safwan Hossain
Meena Jagadeesan
Ariel Procaccia
Eric Mazumdar
Eden Saig
Kunhe Yang