I don't know about you, but I would much rather work for a person than a machine. You can bargain with a person. And people have limited attention spans.
a) you don't trust its value function to be just, or b) you believe it will be just, but you would rather cheat people by taking advantage of human faults like irrational emotions and imperfect memory
We can argue about a) but I can't respect position b).
This is interesting. If all men are created equal and had are grown in same environment then treating by a machine equally is just. But all men are not created equal and hence treating equally by a machine which decides their pay will be considered unjust by the lesser capable men. I am not saying it is just or unjust, just a possible thought process by an intellectually less capable men.
The problem is, how do you measure impact and performance? If you have someone on a team that spends a ton of time helping other engineers with their coding problems, their own commit rate will be lower. If you have someone who checks in a ton of code (not a great messure of ability), they may cause a ton of breakages and bugs for others, a detriment to the group overall.
Meritocracies can be gamed, and any measure you decide to track, will be gamed. At some point machines will definitely be able to track all of these data points, and incorporate them into a model, but it's still only going to have a view of the world based on the data it is fed.
This means that people will game it, and then you have a similar issue as today. While I agree with you in principle, the implementation will not be perfect.