19 Feb

Sofia Silva: new Master Dissertation on “An experiment about the impact of social influence on the wisdom of the crowds’ effect”

Sofia Silva presented her Master dissertation on the 20th of December 2016, at ISCTE-IUL, entitled “An experiment about the impact of social influence on the wisdom of the crowds’ effect”. The jury was composed by Luis Antunes, Jorge Louçã, and Luis Correia (supervisor).

Abstract:
Groups have the impressive ability to perform better collectively than the best of its individuals. Galton observed this first in 1907 in his ox weight experiment, but the term wisdom of the crowds (WoC) was coined only later in 2004 by Surowiecki. Cognitive diversity at the individual level enables groups to produce differentiated solutions that ultimately cluster near the true value. By cancelling out the wrongs, the aggregation method exposes the convergence of multiple local optima solutions into one, typically an averaged value that comes incredibly close to the truth-value of what is being estimating. Some accounts suggest that social influence hinders the WoC effect because it diminishes the group diversity resulting in biased outcomes. However, social influence is a naturally occurring phenomenon and it is hardly determinable the extent to which individuals are biased or independent given the complexity of the social interactions. We investigated the impact of social influence on the WoC effect by comparing the collective predictions of 4 groups regarding the number of jellybeans in a jar. We demonstrate that the group disclosing full information performs nearly as well as the control group, where no information was shared. The aggregation method to converge the estimates was the arithmetic mean showing that both groups predicted by approximately 7% the correct number. Statistical analysis has shown that diversity is not affected significantly in the social groups. We conclude that the WoC is not affected by social influence but by the degree of aggregation of the social information shared.

19 Feb

Carlos Lemos – new PhD thesis On Agent-Based Modelling of Large Scale Conflict Against a Central Authority: from Mechanisms to Complex Behaviour


Carlos Lemos presented his PhD thesis on the 16th of December 2016, at ISCTE-IUL, entitled “On Agent-Based Modelling of Large Scale Conflict Against a Central Authority: from Mechanisms to Complex Behaviour”. The jury was composed by Ernesto Costa, Francisco Santos, Pedro Magalhães, Luis Antunes, Jorge Louçã, Rui Lopes (supervisor), and Helder Coelho (co-supervisor).

Abstract:
In this work, an Agent-Based model of large scale conflict against a central authority was developed. The model proposed herein is an extension of Epstein’s Agent-Based model of civil violence, in which new mechanisms such as deprivation-dependent hardship, generalised vanishing of the risk perception (`massive fear loss’) below a critical ratio between deterrence and `group support’, legitimacy feedback, network influences and `mass enthusiasm’ (contagion) were implemented. The model was explored a set of computer experiments and the results compared with statistical analyses of events in the “Arab Spring”.

The main contributions of the present work for understanding how mechanisms of large scale conflict lead to complex behaviour were (i ) a quantitative description of the impact of the “Arab Spring” in several countries focused on complexity issues such as peaceful vs violent, spontaneous vs organized, and patterns of size, duration and recurrence of conflict events; (ii ) the explanation of the relationship between the estimated arrest probability and the size of rebellion peaks in Epstein’s model; (iii ) a new form of the estimated arrest probability with a mechanism of `massive fear loss’; (iv ) the derivation of a relationship between the legitimacy and action threshold for complex solutions to occur with both low and high values of the legitimacy; (v) a simple representation of political vs economic deprivation with a parameter which controls the `sensitivity’ to value; (vi ) the effect of legitimacy feedback; and (vii ) the effect of network influences on the stability of the solutions.