Hacker News Reader

Comment by Wohlf | Hacker News Reader

The alternative is the unknown human biases of random judges. At least an algorithm can be audited and changed.

Asking from a cultural standpoint, but curious about the technical approach to an “audit” when the same people who biased the training data are ostensibly influencing the audit. Basically, who decides what’s “just” and how could that be reflected in a change to the algorithm?

Judges are elected and can be recalled in many states. Their stances and rules are usually public, too.

Why would someones proprietary codebase be more accessible than public court documents?

The process/decision tree is open and transparent. There's still going to be the potential for implicit bias in the system, but there's no way to remove that entirely from the process (I guess unless you move to a luck based system in which the decision is just made randomly, but nobody would like that)

I have doubts the algorithm would be a decision tree, or capable of being contained in one. I feel like at some point it would use or be an ML model.

Would a model also be open and transparent? It would at least be open & testable, but I'm not sure about transparent for easily identifying any biases.

Good point, but that assumes the algo would need to be known, open/accessible, auditable, etc. If it is somehow hidden behind a proprietary curtain, then that's no good.

Reply on news.ycombinator.com