To your first block of questions - the crowdsourced moderation methods employed in public forums (like for example HN, Reddit, Wikipedia) have evolved to a point where they work very well. It's possible that a government-grade moderation system could be developed by augmenting with machine learning. There could be a grading system employed for the quality, fairness of any legislation submitted that uses both machine-based checks and human moderation to balance the quality grade of the legislation up for vote. "Hard to understand" is something we can detect with software even today. A check by both a human panel and a machine that's parsing language should be enough to ensure that a law is being fairly presented. (Which is a hell of a lot better than the system we have today, where politicians create bills even the people who wrote them don't understand.)
> What I think would be interesting, was a way to wrest my vote from my representative on a by-vote basis. That is, we all get a chance to have a direct vote, but if we don't, our share goes to our representative.
That's a very clever idea. I like that a lot.
To this second block, here's my brainstorm on how it could work - it's possible to counter the potential fatigue by clustering the vote in tiered node cluster groups.
Let's say we made the smallest vote cluster maybe 150 people at any given point, and we'll distribute these groups evenly by geography and population density (adjusted in realtime as demographics shift). The group is politically neutral - assigned based on political sentiment (not party affiliation) using a set of simple surveys and voting behavior so there is an even representation of the political voices in each sample. This is collected with voter registration, or after non-participation for X time period.
Anyone can log in and vote on any legislation they chose each day but the cluster always stays neutral. We apply abuse detection to detect if the cluster is being stuffed or gerrymandered in some way, and the software rejects and reorgs a cluster when abuse is detected. If someone suffers voter intimidation or outside influence they simply report abuse and the system identifies their cluster for review.
Take that model and scale it up, each cluster node is part of a larger node. The vote from each node counts as a whole for that node. If the system needs an extra human layer in a node to recount a vote, it can get swapped with no consequences to other nodes. One node fails, the vote is rerouted elsewhere.
The question really comes down to the participation rate. In social network systems this is best solved with physical rewards at first and later with social rewards. Same can be applied. If nodes get low voter participation, they trigger an automatic vote reward system. Some will try to game the system to get the rewards, pick random answers. This will trigger fraud detection by inserting control questions, honeypots.
Anyhow it's all doable just a matter of prototyping and iterating in a test environment. And the political willpower to implement it.
Probably wouldn't become reality in our lifetimes, at least at the national level, but a fun thought exercise nonetheless