Work with us
Working with us
We work with students, early-career researchers, and established scholars. To start, write to either of us with a short note on what you would like to work on and which of our key questions it bears on.
Who we work with
Students and early-career researchers
Prospective graduate students can apply to the philosophy PhD programs at UNC and Carnegie Mellon. Researchers elsewhere can work with us through the fellowships and programs we mentor in, listed below.
Established researchers
We want to work with established researchers whose work bears on our key questions. That can be a joint paper or project, or a standing affiliation with .
Mentoring
Aydin Mohseni
- Advises students at Carnegie Mellon
- Mentors through SPAR
- Teaches graduate and undergraduate courses on decision theory, game theory, and AI
Current and past students and mentees are listed on the People page.
Key questions
Questions we would like help with, by research area.
Agency
- Under what assumptions about a system’s goals and its environment does power-seeking follow from pursuing almost any goal, and which of those assumptions, dropped, make it fail?
- If we cut the channels two copies of a model use to communicate, can they still coordinate? Causal and evidential decision theories predict different answers.
- What would show that a system treats an outcome as within its control, and can we read that off its beliefs and desires?
Interpretability
- A model fails an eval for a dangerous capability. What licenses the conclusion that it lacks the capability, especially if it can tell it is being tested?
- What would make a probe for deception a measurement of deception, and not of something that merely goes with deception in the probe’s training data?
- Probes are trained on our sentences. How do we find the propositions a model’s own states are about, when they need not line up with ours?
Value
- RLHF and model constitutions already settle whose feedback counts and how conflicts are resolved. Which aggregation rules do they assume, and how should those rules be compared?
- How should AIs make decisions on our behalf, if they are uncertain of our preferences?
- How should an AI system aggregate multiple dimensions of value, when such values are difficult to compare?
Alignment
- A proxy is a measurement, valid under the conditions in which it was calibrated. How far outside those conditions can a system optimize it before the principal is worse off?
- Can a system share our values while reasoning by a different decision theory?
- How reliable must the measurement of a trait be for a given deployment decision?
Contact