The reason is you're inappropriately applying an technical solution to a social problem.
You also make at least one critical error:
> As you contribute to thoughtful discussion, as determined by the diversity and ratio of up/down arrows
Upvotes/downvotes pretty much never measure "thoughtfulness," because for them to do so would require a level of social cohesion that would probably make upvotes/downvote unnecessary.
What if there were separate dimensions for: Factually accurate, Morally agreeable, and Polite?
A machine learning algorithm could quickly work out which groups of users have different ideological positions, and which topics are likely to devolve into flame wars, so it could require that users have a minimum politeness score to be able to participate in a debate.
It could also, as you say, check for echo chambers forming and incentivise people with differing opinions to enter a discussion to balance it out.
Start everyone off with a maximum influence credit of say 10 people, like in your local neighborhood.
As you contribute to thoughtful discussion, as determined by the diversity and ratio of up/down arrows, your maximum influence credit grows as a function of your geography.