Right, there was legal pressure upon inputs of decision-making to make it more egalitarian or whatever. And by other criteria too.

So what happens is full obfuscation of inputs. In the form of LLMs.

Philosophically this is correct in my opinion, trees should be judged by their fruit.

A simplified comparison is British vs Prussian army philosophy, in Prussia, when evaluating officer’s performance, they’d judge his decision-making process and its inputs, even if the result was catastrophic, while in British army and navy they’d only judge the result, no matter how correct the decision-making. That has been often called unjust and not nuanced enough, but one way lost historically and the other won. For a reason. Judgement of inputs has more failure points. It causes degeneracy long-term.

A bit like every metric used as a KPI ceasing to be a useful metric, there’s such a commonly quoted MBA rule, except MBAs are not smart enough to remember that rule, generally.

The alternative to this is responsibility for all that happens downstream. No matter which inputs you get. In exchange for that you are allowed to have any decision-making process at all, just pay for it in full if something wrong happens.

We are being pressed by evolution (including technical progress) to adopt that approach, and it’s good, but it’ll take probably lots of wars and revolutions. People who hide malice behind formally correct inputs do resist. And they do hold power.

Instead of inputs you should treat any social mechanism as a black box, and both limit and judge its outputs. If they are outside limits, discard and punish. If they are inside limits, then evaluate and bill - in prison years or in fines or both. Or reward.

You never know all the inputs anyway and can’t tell if they are correct.