The Institute for Logic, Language and Computation (ILLC) at the University of Amsterdam will host the 3rd ILLC Workshop on Collective Decision Making on 6-7 June 2019 in the historic centre of Amsterdam.
The scientific programme of this informal workshop will be structured around a number of invited talks on topics broadly related to the design and analysis of mechanisms for collective decision making. There will be an open poster session, one of those infamous rump sessions so beloved by the computational social choice research community, and plenty of time for discussion. Previous editions took place in 2013 and 2015.
Invited talks will be delivered by Antoinette Baujard (Saint-Étienne), Caspar Chorus (Delft), Umberto Grandi (Toulouse), Davide Grossi (Groningen), Ronald de Haan (Amsterdam), Arianna Novaro (Toulouse), Daniele Porello (Trento), Ton Storcken (Maastricht), Zoi Terzopoulou (Amsterdam), Matthijs van Veelen (Amsterdam), and Anaëlle Wilczynski (Munich).
The full programme will be announced closer to the time of the workshop. (Tentative schedule: start at 9:30am on Thursday and end by 5:30pm on Friday)
Please note that this list of abstracts is not yet complete.
Is Evaluative Voting Empirically Stable under Different Grading Scales?
Antoinette Baujard (University of Lyon at Saint-Étienne and GATE Lyon Saint-Étienne)
Abstract: This talk deals with evaluative (or "range") voting, and studies to what extent the behavior of the voters, and the outcome of the election, are sensitive to the grading scale. We use data from an experiment conducted in parallel to the 2017 French presidential election: on the day of the first round, alternative evaluative voting rules were proposed to the voters using different scales. The protocol has been adapted to scrutinize two effects. First, the introduction of a negative grade distorts the scores in disfavor of some candidates coined "exclusive". Second, results seem remarkably stable for non-negative scales of different lengths, except for specific categories of candidates: exclusive candidates may be relatively disfavored with longer scales, especially compared with candidates rather unkown to the voters (considering their low and contingent media coverage). (This is joint work with Isabelle Lebon, Jean-François Laslier and Herrade Igersheim.)
Discrete Choice Theory, Choice Experiments and Taboos: Empirical Social Science Meets AI
Caspar Chorus (TU Delft)
Abstract: Discrete choice theory provides a formal, mathematical approach to model human decision making. Developed in the 1970s, choice models are nowadays routinely used in fields as diverse as econometrics, transport, marketing, political science and healthcare. The data used to estimate and validate such models often comes from so-called choice experiments, in which participants are faced with carefully constructed choice sets that allow for a statistically efficient inference of behavioral parameters. After a quick introduction to this field, I will present a choice model that deviates from mainstream models which presume that decision makers are always willing to make trade-offs (e.g., between price and quality of a consumer good). My Taboo-Trade-Off-Aversion model is rooted in moral psychology, and formalizes the notion that in some cases, people find the mere act of making a trade-off morally problematic; think of attaching a monetary value to a human life (which is relevant input for healthcare and traffic safety policies). In a dataset collected by means of a tailor-made choice experiment, I report sizeable heterogeneity in the population regarding the presence and size of taboo-trade-off-aversion. In the last part of my talk, I will explain how the outcomes of this empirical modeling effort can be used to construct a meta-normative theory to help AI make decisions in morally sensitive contexts.
Multi-Issue Opinion Diffusion under Constraints
Umberto Grandi (University of Toulouse and IRIT)
Abstract: Most existing models of opinion diffusion on networks neglect the existence of logical constraints that might correlate individual opinions on multiple issues. In this work we model the diffusion of constrained opinions on a social network as an iterated process of aggregating neighbouring opinions. To overcome the problem of updating towards inconsistent influencing opinions, our model is based on individual updates on subsets of the issues of limited size called propositionwise updates. By adapting notions from the theory of boolean functions, we identify classes of integrity constraints on which propositionwise updates decrease the influence gap between nodes of the network and their influencers caused by the presence of an integrity constraint. Furthermore, we provide a detailed study of the termination of the proposed diffusion processes. (This is joint work with Laurent Perrussel from the University of Toulouse and Sirin Botan from the University of Amsterdam.)
Blockchain Consensus Protocols and Iterated Aggregation on Networks
Davide Grossi (University of Groningen)
Abstract: I will present a model of binary opinion diffusion on networks and show how it can be used as an abstraction to analyse a specific type of consensus protocols in blockchains (e.g., protocols used in systems like Ripple and Stellar). At a high level, such protocols work through a form of iterated voting run on a network of trustees that is constructed locally in a distributed fashion. The model highlights specific weaknesses of those protocols and suggests the application of methods from (computational) social choice to problems and protocols from distributed computing.
General Yet Computationally Efficient Aggregation Frameworks
Ronald de Haan (University of Amsterdam)
Abstract: Several settings in computational social choice concern the aggregation of individual opinions into a single collective opinion. Examples include (committee) elections and voting on which projects to fund within a given budget. For many such settings, efficient algorithms for various aggregation rules have been found. However, these algorithms typically cannot deal with additional constraints (e.g., diversity constraints for committees, or dependencies between projects), whereas such additional constraints are often important when using the aggregation rules in practice. I propose to develop general frameworks that can be used to model different aggregation settings, including various additional constraints that can come up in applications. When developing such a general framework, there is a tradeoff between generality and computational tractability. The social choice model of judgment aggregation can be used as such a general aggregation framework. I will discuss several results that strike a balance between generality and tractability, and explain how these results are relevant for specific aggregation applications.
Strategic Majoritarian Voting with Propositional Goals
Arianna Novaro (University of Toulouse and IRIT)
Abstract: In this talk I will introduce the framework of goal-based voting, where a group of agents has to take a collective decision over some binary issues and each individual is motivated by a goal expressed as a formula of propositional logic. The collective decision is given by a function, whose input are the individual goals, and I will focus on three variants of majority. Agents holding individual goals naturally lead to situations where strategic voting may occur: i.e., when an agent reporting an untruthful goal may get a better outcome. I will present manipulability results for the general case, as well as strategy-proofness results when the language of the goals is restricted or the agents can only use a subset of strategic actions. Finally, I will talk about how computationally hard is it for an agent to know if they can profitably manipulate.
Collective Choice Rules on Domains with A Priori Information
Ton Storcken (Maastricht University)
Abstract: In this restricted-domain approach it is assumed that parts of the agents' preferences are known a priori. This means that the admissible preferences of an agent contain a given partial order, where this partial order may differ per agent. It appears that these sets of admissible preferences are convex with respect to the betweenness relation induced by the Kemeny distance on preferences. Necessary and sufficient conditions are formulated under which such restricted domains admit unanimous, strategy-proof, and non-dictatorial choice rules. Loosely speaking it boils down to admitting monotone and non-image-dictatorial decision rules on two alternatives, where the other alternatives are completely disregarded and these two alternatives are the maximal elements with respect to all a priori given partial orders. (This is joint work with Shashwat Khare.)
Collective Decision Making with Incomplete Individual Opinions
Zoi Terzopoulou (University of Amsterdam)
Abstract: Across different formal frameworks for collective decision making (e.g., voting, preference and judgment aggregation), a common assumption is that the agents hold and report opinions that are complete, i.e., concern all issues for which a collective decision has to be made. I confront this assumption, arguing that in many reasonable scenarios the agents can be expected (and should be given the freedom) to either have or report incomplete opinions. How can we then modify our models to account for such incompleteness, what challenges do we need to overcome, and what changes may be caused in classical results of the field? In this talk I will open this discussion and provide several theoretical insights motivated by examples of practical interest.
Good and Evil
Matthijs van Veelen (University of Amsterdam)
Abstract: I will give an overview of the evolutionary explanations of altruism and cooperation. There are three broad categories: population structure, repeated interaction, and partner choice. Population structure implies that individuals do not interact (uniformly) randomly, but one's own type is predictive of the type of the individual one interacts with. There are interesting complications here, such as the cancellation effect. Repeated interaction turns one-shot games into repeated games, with, typically, many more equilibria than the one-shot game, including more cooperative ones. We will see that this allows for transitions into and out of cooperation. With partner choice, a species can become their own selective force. Ingredients from these three broad categories can also interact in interesting ways.
Local Envy-Freeness in House Allocation Problems
Anaëlle Wilczynski (TU Munich)
Abstract: We study the fair division problem consisting in allocating exactly one item per agent, which is called house allocation, so as to avoid, or minimize, envy. Envy-free allocations are rare and trivial in house allocation: every agent must get her most preferred object. Moreover, assuming that any agent is able to envy any other agent is a strong assumption since the visibility or the jealousy of the agents may be limited due to a lack of knowledge or specific relationships. Consequently we aim to relax the notion of envy in house allocation by considering a notion of local envy which is determined by a network structure over the agents, represented by a directed graph. In this context, an allocation is locally envy-free if no agent envies one of her successors in the graph. The existence of a locally envy-free allocation is investigated, as well as the minimization of local envy in case a locally envy-free allocation does not exist. We study the complexity of these problems and also examine some variants where the role of the central authority is more or less important.
Everyone is cordially invited to contribute to the open poster session. This is an inclusive event and we specifically invite submissions discussing work in progress (but you are equally welcome to simply submit, say, a poster you will also present at one of the major conferences this year). Submissions will be reviewed for relevance to the workshop only and, subject to space constraints, we intend to accept all reasonable submissions.
You can submit either an abstract of up to 500 words or a draft of the poster itself. Please submit a single PDF document anytime before the end of Monday, 20 May 2019 by following this link:
Early submissions are welcome and encouraged. We will try to let you know whether we can accommodate your poster within one week of receiving your submission.
We would like to get a realistic estimate of the number of attendees well in advance of the event, so as to be able to order the right amount of coffee. Here is how we hope to achieve this:
Early registration is free and possible using this form until Saturday, 25 May 2019.
Late registration costs € 10, to be paid in cash on site.
The first five registered participants will be offered a free copy of the book Trends in Computational Social Choice.
De Doelenzaal, University Library, Singel 425, Amsterdam
Trams 2, 11, and 12 stop at Koningsplein across the street. Metro 52 stops at Rokin, a 5-minute walk away. The main train station (Amsterdam Centraal) is a 20-minute walk away.