What Eugene Volokh suggests in these 22 paragraphs are some interesting thoughts on online voting. I'll engage him for a moment.
I don't like the idea in that his system, as proposed, tightly aligns the automated process of guidance to the actual vote. I think the vote should be a pure expression of will. The idea that an automated system could actually capture every nuance of decision making is flawed. The entire thing can be spoofed, not to mention subverted. But the fundamental flaw is what happens to interest group standing and the consequent effect of omissions and conflicts in the system's likelihood of setting up limited choices.
Contrast Guided Voting (GV) to XRepublic. It is fair to view GV as a watered down version of an XRepublic, whose function is more oriented toward the more wonkish work of deliberation and crafting of resolutions. In XR the processes and outputs are dynamic, with GV the processes are deterministic. Linking them to voting makes the outputs deterministic as well.
It is this particular aspect that bothers me the most because at bottom there will always be a bottleneck in the capacity of the system. At issue is the mechanism used to determine which selection of interest groups merit inclusion in the weighting algorithm of GV. At the heart of the GV system are will be some affinity scoring algorithm that maps voter opinion through the lens of interest groups onto ballot choices. There are a couple places this gets dicey.
The first area for contention would be in weighting the scores. This is what I call the interest group conflict problem. Let's say a thumb (binary choice) is requested on 2nd Amendment question in the form of a gun control ballot initiative. On all the points of the language of a proposal behind the ballot initiative, the NRA scores .89 certainty for a NO. On all the points of the Gun Nuts of South Texas also scores an .89 certainty for NO. Some representative of the GNST, upon hearing this claims to be a .92. It is a matter of pride for their partisans to be stauncher supporters of gun rights than their rival organization, the NRA. Who arbitrates? How long? Where is the science? Who certifies it?
The second issue is that of omission. What incentive do maintaners of the system have to include 'redundant' or 'superfluous' interest groups? What standing is required to become legitimized as an interest group? This problem is rather obvious so I won't elaborate.
The third is an issue of horseracing tripwires. In this case, the system programmers, in programming the system's affinity matrix notice some interesting sets. By estimating the number of voters who have generally announced their affinities in other elections for the NRA and GNST interest groups, they will be able, by dint of their knowledge of the affinity matrices, to project with certainty the outcomes of elections. This knowledge will get into the hands of campaigners and can be used strategically to undercut democracy in a must cruel way. I can think of no clearer example of this than the current recall election in California. Who would have thought that for a mere $1.7 million dollars, approximately 1.02 per signature that a recall measure could force a vote of confidence (which incidently costs the State of California $32 million simply to launch). The point is not the expense but the increased determinacy of tripwires in elections that only require narrow activism to force public change.
It may be that within the scope of the decision space of any one election, a limited set of influencing interest groups will provide sufficient diversity to allow voters more than enough 'free will' as determined by the affinity algorithms. But this is something that needs to be watched very closely. I would strongly suggest that any such system be decoupled from actual voting and tried on the simple merits of guidance.