Towards Composable Bias Rating of AI Services
Biplav Srivastava, Francesca Rossi
AIES 2018
By the Gibbard–Satterthwaite theorem, every reasonable voting rule for three or more alternatives is susceptible to manipulation: there exist elections where one or more voters can change the election outcome in their favour by unilaterally modifying their vote. When a given election admits several such voters, strategic voting becomes a game among potential manipulators: a manipulative vote that leads to a better outcome when other voters are truthful may lead to disastrous results when other voters choose to manipulate as well. We consider this situation from the perspective of a boundedly rational voter, using an appropriately adapted cognitive hierarchy framework to model voters’ limitations. We investigate the complexity of algorithmic questions that such a voter faces when deciding on whether to manipulate. We focus on k-approval voting rules, with k≥1. We provide polynomial-time algorithms for k=1,2 and hardness results for k≥4 (NP and co-NP), supporting the claim that strategic voting, albeit ubiquitous in collective decision making, is computationally hard if the manipulators try to reason about each other's actions.
Biplav Srivastava, Francesca Rossi
AIES 2018
Umberto Grandi, Andrea Loreggia, et al.
ISAIM 2014
Cristina Cornelio, Judy Goldsmith, et al.
JAIR
Cristina Cornelio, Antonio Nicolò, et al.
AIES 2019