[jfager]

I'm a big fan of this general idea but taken to the limit of being able to delegate all issue votes to any person goes too far, and I'd prefer a system that elected representatives who accrue a threshold number of votes. There are a few reasons for this, but I'll just talk about one here, privacy.

Privacy is addressed in the paper but I think the real issues with it are just glossed over. "You get to see how your delegate votes" is great and should absolutely be true, but there are two other sorts of privacy that matter: privacy from the electoral system itself, and privacy from closely related people.

In very Googley fashion, this paper seems to take for granted that of course you'd trust the system itself with the knowledge of your votes and delegates, and that its privacy-respecting duty is fulfilled by simply not broadcasting that knowledge to the unauthorized public. But that's a terrible way to run a government that at least partially relies on a secret ballot as a means of expressing dissent and effecting change in the government itself, for hopefully obvious reasons.

The other form of privacy that matters is on a much smaller scale. A delegate who can exercise the votes of an arbitrarily small number of people is also capable of ensuring they are exercising the votes of a specific set of people. An abusive husband/father/wife/mother/boss/etc can know simply from the count of votes they hold whether their victims are following through on their command, and via transitive delegation they can pool that power with collaborators.

[cbsmith]

There's all kinds of math on techniques for being able to audit the vote without exposing individual voting behaviour. It is possible to prove to yourself that your vote was counted without exposing what your vote was to others, and also providing assurances that the result represents the aggregate of everyone's "votes". I don't think there is much about the important aspects of this design that require there be massive privacy issues. There are some variants around "honest verifiers" that work even more simply. Most of the homomorphic encryption excitement stems from some of the more sophisticated variants of that approach.

[jfager]

Do you have any pointers to papers describing how vote auditing techniques extend to liquid democracy? The Google paper doesn't even acknowledge the problem, much less describe a solution, and most of the other papers I've read on the topic are either extremely handwavy on the technical details of privacy or make the same assumption as Google, that its good enough to hide votes from the participants but 'the system' still knows.

[cbsmith]

I'm not sure I understand how liquid democracy would change the fundamental problem context in a way that would change the problem for homomorphic encryption...

[crdoconnor]

>I'm a big fan of this general idea but taken to the limit of being able to delegate all issue votes to any person goes too far

Why?