Research

Testing

Liquid Democracy

The question is what local delegation mechanisms can achieve on graph classes where liquid democracy is already known to improve epistemic welfare in the absence of attacks. Each voter can only delegate to available neighbors in the fixed delegation graph imposed by the mechanism, and the voter decides either to vote directly, or to delegate using local information to a neighbor allowed by the mechanism. The difficulty they present is that many individually sensible delegations can create globally concentrated terminal voting power. Even when that concentration is statistically acceptable, it may make a small number of delegates disproportionately valuable targets for capture (i.e. forcing them to vote a certain way).

I conjecture that on some graph classes where competence-based local delegation achieves positive gain and do-no-harm in the clean setting, local competence information alone is insufficient to preserve those guarantees against corruption of π‘˜ terminal delegates. The goal is to characterize the weakest additional information (like a local signal of expected terminal exposure through other delegation) that is necessary and sufficient to restore robust positive gain and do-no-harm, and to determine the largest corruption budget π‘˜ for which such guarantees are achievable. A tight result would therefore identify exactly how much nonlocal information robust liquid democracy requires, rather than just observing that concentrated delegates are vulnerable.

Ideas and Brain Nonsense