Advanced Seminar in Behavioral Machine Learning
This seminar focuses on the relation between machine learning systems and human behavior and on the unique challenges that arise once considered jointly. The goals of this seminar are to develop critical thinking skills and provide tolls for the development of learning systems that operate in human-driven environments. Some of the questions we will pursue are:
- What power do learning systems have over the choices provided to us and on the information we are exposed to?
- What are the limits of human modeling, and what are the pros and cons of using such models within learning systems?
- What assumptions on human behavior - explicit and implicit - are incorporated in learning systems? What are the implications of making these assumptions?
- What is the "right" way to train a system to support human decision making? Can the system learn to manipulate or control decision makers? If so, can this be revealed?
In class students will present (and the class will discuss) technical papers in machine learning (both classic and contemporary),but within a behavioral context and in light of results, experiments, and theories from the social sciences. Students will also be asked to submit a medium-scale report analyzing a human-facing learning system of their choice.
- Attendance in at least 90% of class meetings
- Presenting one technical paper + behavioral background (60% of grade; possibly in pairs)
- Final project, medium scale (30% of grade)
- Fun (!) weekly assignment (10% of grade)