Leïla Amgoud Formal models of arguments [CANCELLED]
Argumentation is a reasoning approach based on the justification of claims by arguments. It is largely used in online debate platforms, tools where individuals from around the world come to debate online and read the opinions of others. The presentation will focus on the notion of argument. I will present a logical language for representing arguments and relations between them, and will talk about how to evaluate the strengths and weaknesses of an argument.
David Budescu The wisdom of forecasting teams
This talk is motivated by, and relies on data from, recent large-scale geopolitical forecasting tournaments (Mellers et al., 2014). Two key results of the tournaments are (1) the possibility to identify reliably expertise (Budescu & Chen, 2015; Chen et al, 2016) of individual forecasters and leverage it to improve the accuracy of the forecasts through efficient aggregation of relatively small crowds of “selected” forecasters (Mannes et al. 2014), and (2) the surprising success of small collaborative teams. This “teaming effect” is particularly intriguing because it seems to contradict the “wisdom of the crowd” hypothesis that emphasizes the importance of independence among forecasters. I will discuss both results with special attention to the “teaming effect” which I attribute to the hybrid approach implemented, which allowed forecasters to share information electronically and asynchronously, but required them to provide forecasts individually, for a statistical aggregation procedure. This hybrid approach benefits from the strengths of Computer Mediated Communication (CMC) and Statistical Aggregation.
We show that if one has access to a large number of forecasters, one can increase prediction accuracy in a novel statistical way. Place individual forecasters into teams and incentivize teams to compete against each other. Then construct new teams that combine small subsets of original teams using different original teams. This process leverages the benefits of within-team cooperative and between-team competition, two natural tendencies that can motivate people to do their best.
References
Zoé Christoff Logical foundations of social influence in networks.
We give an introduction to the use of logical tools in understanding social influence and social networks phenomena. Individuals often form their opinions by interpreting the behavior of others around them, and by reasoning about how those others have formed their opinions. This leads to several well-known herd phenomena, such as informational cascades, bystander effect, pluralistic ignorance, bubbles, and polarization. For instance, in the case of informational cascades, agents in a sequence imitate each other's choices despite having diverging private evidence, sometimes leading the whole community to make the worst possible choice. Similar cascading mechanisms are at the heart of social networks diffusion phenomena.
We first show how an epistemic logic modeling allows to understand the conditions for such cascades to form, as well as their inescapability. We then turn to what logical tools can do for understanding information flow and influence in social networks. We illustrate how extremely simplified models might yield surprising new results, for instance about stabilization conditions of diffusion processes.
Mirta Galesic Wisdom of small, slow, and biased crowds
It is often thought that group decision making improves when groups are larger, communicate and decide faster, and have members who are as accurate as possible. However, interaction of group decision rules, network structure, and task characteristics can often produce situations in which these conclusions do not hold. I will present results suggesting that small groups can be better than larger groups and individuals in real-world situations. In addition, slower decision strategies or slower communication can improve group performance on complex tasks. Finally, I will show that election winner expectations of biased forecasters can be quite accurate when their social circles are sufficiently diverse.
References:
Davide Grossi Deliberation and Collective Decision-Making in the Blockchain Consensus Problem
In this talk I will provide a social choice theory perspective on blockchain consensus protocols. After discussing the basic workings of such protocols, I will try to showcase them as a novel and promising application domain for economic research into processes of information diffusion, deliberation and collective decision-making.
To illustrate this perspective I will present a simple model of binary opinion diffusion on networks and show how it can be used as an abstraction to analyse specific types of consensus protocols (specifically, those deployed by successful international businesses such as 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 limitations of those protocols and illustrates how economic theory can shed light on fundamental issues concerning consensus in blockchains.
Ali Jadbabaie On Bayesian and non-Bayesian Social Learning and Opinion Exchange: Algorithms and Complexity
In this talk, I will present the latest results on my group’s decade long study of social learning and opinion dynamics where we study the behavioral foundations of non-Bayesian models of learning over social networks and present a taxonomy of conditions for information aggregation in a very general framework. I will compare and contrast such models with fully rational models of Bayesian social learning, and present our very recent results on computational complexity Bayesian decision making. Next, I will present a model of information sharing on social media according to which agents make forwarding decisions based on whether new information can persuade others to think more like them. I will show how such persuasion motives might result in misinformation cascades.
References:
Christian List The Will of the People Revisited
Emiliano Lorini A Logic of Collective Belief
I will present a logical analysis of the connection between agents’ individual beliefs of both explicit and implicit type, expressed beliefs and collective beliefs. Individual explicit beliefs are an agent’s private beliefs, i.e., the body of information in the agent’s belief base. An agent's expressed belief is an information that the agent reveals to the others. Individual implicit beliefs are facts that an agent can infer from what she explicitly believes and from the public information. Finally, collective belief is the result of the aggregation of the agents’ expressed beliefs. Different kinds of aggregation operators will be discussed ranging from shared belief to distributed belief.
Manuel Mueller-Frank Naïve Learning Through Probability Matching
We analyze boundedly rational updating in a repeated interaction network model with binary states and actions. We decompose the updating procedure into a deterministic stationary Markov belief updating component inspired by DeGroot updating and pair it with a random probability matching strategy that assigns probabilities to the actions given the underlying boundedly rational belief. This approach allows overcoming the impediments to consensus and naive learning inherent in deterministic updating functions in coarse action environments. We show that if a sequence of growing networks satisfies vanishing influence, then the eventual consensus action equals the realized state with a probability converging to one.
Klaus Nehring Weighing Experts, Weighing Sources: The Diversity Value (joint work with Ani Guerdjikova)
A decision maker has to come up with a probability, preference or other judgment based on the judgments of a number of different information sources. To do so, he needs to assign weights to each source reflecting their assessed reliability. We argue that, crucially, reliability is to be understood as an attribute of sets of sources, not of sources in isolation. Specifically, we propose to view reliability as "valued diversity", reflecting both individual source quality and similarity among sources. Intuitively, larger weight should be assigned to sources of greater quality and greater dissimilarity from the others. The main contribution of this paper is to propose and axiomatize a particular weighting rule, the Diversity Value, that captures these desiderata. The Diversity Value is defined by a logarithmic scoring criterion and can be characterized as a weighted Shapley value, in which source weights are determined endogenously. Due to the central role of source similarity, the Diversity Value frequently violates Reinforcement and exhibits the No-Show Paradox.
References:
Drazen Prelec Finding truth even if most people are wrong
The question whether to trust the judgments of a few experts or the wisdom-of-the-crowd is not just of scientific but also of political and philosophical interest. Crowd wisdom is usually defined as consensus — the majority vote or the median estimate or forecast. This principle seems fair and simple, but it has a blind spot for information that is new or unfamiliar. The cost of wrong collective decisions can be high in terms of environmental risks underestimated, or promising ideas ignored. The challenge is to combine the virtues of a ‘democratic’ procedure, which allows anyone, irrespective of credentials, to register an opinion, with an 'elitist’ outcome that associates truth with the judgments of a select few. I will describe a simple alternative to democratic averaging by a panel or online crowd. The alternative principle is to select judgments that receive more support than predicted by those same people. I will review some recent evidence bearing on this approach, and discuss extensions to forecasting.
Anton V. Proskurnikov Mathematics of Social Influence Network Theory: Centralities, Opinions, Belief Systems.
Social influence is a term commonly adopted in social psychology. An individual tends to change their attitudes and behaviors, intentionally or unintentionally, participating in relations and interactions with other people. Quantitative analysis of social influence effects in real social networks however remains a challenging problem lying at the borderline of mathematics and sociology. The social influence network theory (SINT) started in the works by N.E. Friedkin and E.C. Johnsen more than 20 years ago stipulates that social influence mechanisms are temporal rather than static. These mechanisms do not reduce to sporadic interactions in pairs of actors but should be considered as dynamical systems over the networks, describing evolution of some characteristics (opinions, attitudes, beliefs). The lecture presents some mathematical models of SINT and relevant results, obtained in the literature. Open problems and challenges of SINT, as well as recent experiments aimed at validation of its main concepts will be considered. We discuss, in particular, the relation between SINT, the theory of centrality measures and game theory. New models of SINT that have recently been proposed to describe the evolution of belief systems will also be considered.
References
Agnieszka Rusinowska The aggregation function approach to diffusion, opinion formation and (anti-)conformism in networks
Studying diffusion in networks has a very strong and established position in the literature, with a variety of approaches and different ways of modeling diffusion. Adoption of a new technology, product purchasing and marketing, opinion formation, influence, social learning, information cascades and transmission, fashions, contagion and disease infection, financial contagion, among many other related phenomena, they are all relevant to network diffusion. The present talk concerns diffusion in networks, opinion formation, conformism and anti-conformism, with a particular focus on the approach based on aggregation functions.
Grabisch and Rusinowska (2013) introduce a general framework of influence based on aggregation functions, where every individual updates his ‘yes’ or ‘no’ opinion by aggregating the others’ opinions. This determines the probability that the agent’s opinion will be ‘yes’ in the next period. Instead of allowing for arbitrary aggregation functions, Foerster et al. (2013) consider anonymous aggregation, i.e., anonymous social influence which is merely due to the number of agents having a certain opinion, not their identity. Both frameworks cover only positive influence (imitation), since by definition aggregation functions are non-decreasing, and hence can model only conformism. The aim of Grabisch, Poindron and Rusinowska (2018) is to study opinion formation under anonymous influence in societies with conformist and anti-conformist individuals. Three classes of aggregation rules that can be used by the agents when revising their opinions are examined: purely conformism, purely anti-conformism, and mixed aggregation rules.
Morris (2000) studies contagion defined as a diffusion phenomenon occurring when one of two actions can spread from a finite set of individuals to the whole (countably infinite) population. The basic mechanism used in his work is a deterministic process, known under the name of threshold model, where an agent becomes infected (or active) if the number of infected agents in its neighborhood exceeds a given threshold. Grabisch, Rusinowska and Venel (2019) depart from Morris (2000) and investigate diffusion in a countably infinite society of individuals interacting with their neighbors. The diffusion mechanism is based on an aggregation function, which leads to a Markov process with an uncountable set of states. In particular, when considering Boolean aggregation functions, the diffusion process becomes deterministic, and the contagion model of Morris (2000) can be seen as a particular case of the framework with aggregation functions.
References:
People often believe falsehoods, a fact that is hard to reconcile with standard models of choice under uncertainty. I introduce an alternative framework in which an agent sequentially learns propositions and must decide which to believe. A first question is: are there “good” update rules? I study two families of axioms, “willingness-to-learn” axioms and “non-manipulability” axioms, obtaining an impossibility result. Selected combinations of axioms lead to rules that capture skepticism, wishful thinking, and stubbornness. I subsequently discuss the relationship between this framework and Savage's “small worlds.” Two examples illustrate applications to bounded rationality.
Katarzyna Sznajd-Weron Statistical Physics Of Opinion Formation: is it a SPOOF?
I will present a short review based on the nonlinear q-voter model about problems and methods raised within statistical physics of opinion formation (SPOOF). I will talk about relations between models of opinion formation, developed by physicists, and theoretical models of social response, known in social psychology. Finally, I will show examples of studies directly inspired by social psychology like: “independence vs. anticonformity” or “personality vs. situation”.
References: