Could we completely solve the smurf problem with AI bots?

To be honest, a lots between us has multiple accounts for Dota2.

The main reason to make a second account/ smurf, (at least the reason for me):

1. "My teammates sucks, I deserve better." -> so create new accounts to try callibrate higher.
2. "Main account gaming is too serious, I want to try some memes."

If there exists a human-like AI for a specific MMR! and here is my MMR system proposal:

1. People are ranked with badges just like now: crusader, archon,…, immortal. But, we may need more levels. Then, let's say if an ancient 2 player wins a lot (like 8 out of 10) in this bracket, then he can play a game VS the bot with skill level between Ancient 2 and Ancient 3, if he wins then he will get promoted. On the other hand, if he loses a lot, he will forced to play a game VS the Ancient 1~Ancient 2 bot, if he loses the game, he will become Ancient 1.
2. To prevent "smurf"/"account buyer" to intentionally lose/win the bot match, if the player wins a lot like 24/30 then he can skip the promotion process.

Why this solve the smurf problem:

1. Now the mmr gaining process is VS bot, so you can hardly argue that it's the smurf in the opponent team ruining your game.
2. If you want to try some meme build or relaxed game, you can play with those high quality AI.
3. In this process, the promotion process is faster than the current system, which makes a smurf account harder to make.
4. (Bonus) This also can solve the "skill inflation" problem, since those bots are like records of the skill level, new player can learn thourgh this process.

But, how to build this kind of AI? It should be really hard!

Actually, I have an idea inspired by the TI prize pool.

As the bot interface is already open source. (There are already many bots good or bad exists.) If Valve allow the developers to upload their AI code and test at the corresponding bracket. And ask the players to answer blindly the opponent team is AI or not. And give like say a percentage of TI prize pool to the top AI codes.

This will attract many research team to develop a such thing I guess. (Since the TI prize pool is already so huge, even a small percentage of it is already attractive.)