@M3RC3N4RI0 Nah look, from what I understand, like most of their contents in Paladins, Hi-rez did not want anything to be the same as other games, and I guess that's why they came up with this MMR (who knows, maybe all those "classical" matchmaking systems are now patented or something).
So, comparing a normal matchmaking system with theirs: what do we have?
In a league-based matchmaking (silver with silver, gold with gold, etc), the problem is that the league does not necessarily represent the skill of the player, but only the luck that he had with matchmaking. Imagine playing in Paladins in the bronze league; It would be a freaking disaster. The advantage would be that, if you are a good player, once you're in gold or platinum you can easily stay there.
What they did with this MMR, from what I get, is: they tried to assign a skill identifier to every player(I guess by reading kills, objective time, healing, damage, etc), and I guess they match the closest MMRs together. The only problem is that when you get bad team mates and pro enemies, it does not really matter how good you are, it is very difficult to get a positive feedback to your MMR, so therefor you will be classified as a "bad" player because of this, and you will continue getting matched with "bad" players. This is one case. The other case is the other way around, where bad players get classified as pro players because of good teams and weak enemies.
So yes, the MMR matching system is theoretically good. The problems appear after one wrongfully bad or good game, and from there, it can last even for weeks until u return to normal.
You can't classify a player as pro or newbie by judging numbers. It happens often that if you look at the scores you would think a losing team should have won, but victory in siege does not depend on any individual numbers, but only on team work. Five noobs playing team can easily waste five selfish pros.
So yeah, another view on why Paladin's matchmaking needs to be fixed. And no matter how this MMR works, there must be fixes possible. Maybe an AI verifying system like you said: compare the system's expectations before the game with the real outcome, and work from there. Who knows, after all @Shadowpuppy is partially right. We can't suggest fixes if we don't know exactly what the problem is. However, it's pretty obvious that there is a problem.