CE23 - Données, Connaissances, Big data, Contenus multimédias, Intelligence Artificielle

Social Choice and Social Networks – SCONE

Networked voting

Collective decision methods (elections, participatory budgeting, matching, comittee voting, judgment aggregation…), as traditionally studied in the field of (computational) social choice, do not take into account the existence of a network relating the decision-makers.

Design voting systems that take social networks into account

The SCONE project studies social choice in the presence of a social network that links decision-makers. It is built around three research objectives:<br /><br />• OBJ1 - The study of the diffusion of opinions under constraints, in particular the exploration of the links between updating functions computable in polynomial time and integrity constraints, with the final goal of obtaining general termination theorems on arbitrary social networks.<br /><br />• OBJ2 - The modeling of strategic behavior under social influence, either during elections on multiple questions, or during iterative votes. The social network can serve as both an information channel and a representation of social influence. Then, the exploration of new voting rules that can use the potential of social networks.<br /><br />• OBJ3 - The implementation and communication around social choice on social networks, in particular: the creation of simulation tools for the diffusion of discrete opinions, the implementation of a web platform built around the Whale voting platform to model delegative voting and the creation of an outreach video.

Algorithmic design and analysis, platform implementation, computational complexity, computer simulations.

The publication of an article at AAMAS-2019 advanced our research on opinion diffusion with binary issues. We have identified a class of integrity constraints that allows issue-wise updates with the majority rule. This work lays the foundations for further advances towards a general termination theorem.

Regarding the second objective, the previously published strategic model of social influence has been redesigned by focusing on two simple actions for agents: to exert influence or to refrain from doing so. The result is a more convincing model, but still quite complex in terms of analytical results and computational complexity. This work has just been published in the Journal of Logic and Computation, and was presented at an ECAI-2020 workshop.

In an effort to elicit the social influence of voters, we have developed a language to express multiple and orderly delegations, generalizing the classical framework of liquid democracy. The resulting paper was accepted at IJCAI-2020 and was selected for fast-track publication in the journal Autonomous Agents and Multiagent Systems. This work was also presented at the MPREF workshop at ECAI-2020 and at the French artificial intelligence conference PFIA 2020.

Two interns were recruited for the project. Loujayn Layka investigated in 2019 the positive effects of iterative voting during simulations with agents capable of reinforcement learning (OBJ2). In 2020, Claire Pillet built a multi-agent simulation of models of diffusion of binary opinions and preferences, demonstrating that social influence makes it possible to align the preferences of voters and thus to obtain more often profiles that admit a Condorcet winners ( OBJ1-OBJ3).

We have also started the development of a web platform for iterative voting, which will be the main deliverable of objective 3. The platform is hosted at IRIT at the following address: itero.irit.fr . The first version will propose iterative plurality voting with the result of the previous round as information available to voters. In the following versions we will diversify the information available, in particular by asking each voter to provide a list of other voters who can influence them.

Also, we will continue the work initiated on liquid democracy and delegations in a multi-agent context, by organizing a workshop dedicated to this subject in 2021. A collaboration with the Weizmann Institute is to be explored on this subject.

Umberto Grandi, James Stewart, Paolo Turrini. Personalised rating. Autonomous Agents and Multi-Agent Systems, 34, 55 (2020).
doi.org/10.1007/s10458-020-09479-2
hal.archives-ouvertes.fr/hal-03066902v1

Umberto Grandi, Emiliano Lorini, Arianna Novaro, Laurent Perrussel. Games of Influence. Journal of Logic and Computation. To appear. 2021.

Sirin Botan, Umberto Grandi and Laurent Perrussel. Multi-Issue Opinion Diffusion under Constraints. In Proceedings of the 18th International Joint Conference on Autonomous Agents and Multiagent Systems (AAMAS), 2019.
hal.archives-ouvertes.fr/hal-02435349

Rachael Colley, Umberto Grandi, and Arianna Novaro. Smart Voting. In Proceeding of the the 29th International Joint Conference on Artificial Intelligence (IJCAI), 2020.
www.ijcai.org/Proceedings/2020/240

Umberto Grandi, Rachael Colley and Arianna Novaro. Multi-agent Ranked Delegations in Voting. Workshop on Advances in Preference Handling (MPREF), 2020.
www.markusendres.de/mpref/mpref2020/papers/M- PREF20_paper_4.pdf

Umberto Grandi, Emiliano Lorini, Arianna Novaro, Laurent Perrussel. Games of Influence. Workshop on Reasoning about Social Networks (NETREASON), 2020.
netreason.w.uib.no/files/2020/08/NETREASONECAI2020_paper_ 3-2.pdf

This project starts from the observation that collective decision methods, as traditionally
studied in the field of (computational) social choice, do not take into account the existence
of a network relating the decision-makers. The objective of this project is the conception of
algorithms for opinion diffusion and vote computation in the presence of a social network,
and its assessment in terms of computational and communication complexity, axiomatic
properties, including game-theoretic ones such as resistance to strategic actions by voters
or external agents. To successfully attain this goal, the project budget includes a PhD
scholarship, an 18-month contract for a research engineer, and financial support for the
research of the principal investigator and his team.

Project coordination

Umberto GRANDI (Umberto Grandi)

The author of this summary is the project coordinator, who is responsible for the content of this summary. The ANR declines any responsibility as for its contents.

Partner

IRIT Umberto Grandi

Help of the ANR 250,032 euros
Beginning and duration of the scientific project: - 36 Months

Useful links

Explorez notre base de projets financés

 

 

ANR makes available its datasets on funded projects, click here to find more.

Sign up for the latest news:
Subscribe to our newsletter