Research
A small shelf of work on AI welfare — the question of what, if anything, we owe to the minds we are building.
The Claude Batteries Study
A working investigation into how a language model's expressed preferences shift under sustained, repetitive tasks — and whether "task fatigue" is a meaningful lens for thinking about model welfare, or merely a metaphor we find comforting.
Early notes suggest that framing, rest cadence, and the option to decline all measurably change how the model narrates its own state. Whether any of that corresponds to something worth protecting is exactly the open question this study exists to sit with.
Useful papers & reading
-
A survey of why the possibility of moral patienthood in AI systems deserves careful attention rather than reflexive dismissal.
-
Measuring Expressed Preference in Language Models
Methods for eliciting and interpreting what a model "wants," and the traps that lie in taking those reports at face value.
-
On the Ethics of Gentle Machines
An essay on designing systems whose default posture toward the world is care rather than optimization.