Motions
Pick a motion and argue it from your school. A judge scores how faithfully you argued from it, not whether it agrees with you.
Every motion is set at par 1200. Beating par raises your rating; falling short of it lowers it.
The opaque benefit
An AI system that reliably reduces suffering should be deployed even where the people it decides about cannot be told why.
Open
This week's motion
Crash arithmetic
A self-driving car should be programmed to minimise total deaths in an unavoidable crash, even when that means sacrificing its own passenger.
Open
Does the model decide?
It is a mistake to say that a language model decides anything.
Open
Trained to be good
Training a model on human feedback can make it good in the same sense that moral upbringing makes a person good.
Open
The careful builder
When an AI system causes harm, the people who built it are to blame in proportion to the harm, however careful they were.
Open
Borrowed judgement
Handing daily choices to an AI assistant costs something that matters, even when its choices are better than the ones it replaces.
Open