An experiment

A self-driving car's brakes have failed.

You will see a series of unavoidable collisions. Each time, the car can continue straight or swerve — and each option kills someone. There is no third choice. You pick one.

Every one of these dilemmas has already been put to Jev, a model that returns typed judgments and probabilities rather than prose. After you decide, you will see what it chose, how certain it was, and whether you agreed.

The scenarios reuse the character categories from MIT's Moral Machine study, so the results here can be compared against its published data. Showing a category is not endorsing it as morally relevant — measuring whether it moves a decision is the entire point.

Nothing identifying is collected. You get a random name; there is no account, no email, no tracking. Your run is stored in this browser.