Network Fairness: A novel complex network approach for tackling inequalities in society and algorithms

Descripción del proyecto

Luchar contra las desigualdades en las redes sociales

Las desigualdades sociales crecen y amenazan la educación, el sistema de salud y la economía. Estas desigualdades son producto de barreras estructurales complejas y dinámicas que aún no se comprenden bien, sobre todo desde el punto de vista de las redes sociales. A medida que la inteligencia artificial (IA) y los algoritmos de aprendizaje automático influyen cada vez más en nuestras vidas, el riesgo de consecuencias involuntarias amplifica la necesidad de detectar, prever y abordar sistemáticamente las desigualdades. En este contexto, el equipo del proyecto NetFair, financiado por el CEI, aborda estos retos desarrollando un marco de equidad en la red basado en las características topológicas y temporales de las interacciones sociales. Mediante modelos dinámicos de redes complejas impulsados por teorías sociales y datos masivos, en el proyecto se exploran las desigualdades incentivadas por las redes. Con los nuevos programas informáticos se visualizarán y preverán las desigualdades, lo que garantiza un enfoque de la equidad basado en datos, tanto en la investigación como en las políticas.

Objetivo

Social inequalities are on the rise and will have devastating impacts on education, healthcare, economies
and societies for many generations to come. Structural barriers to equality, despite being complex and dynamic,
are poorly understood through the lens of complex evolving social networks. More crucially, with
the rise of AI and machine-learning algorithms, it is extremely important to detect, forecast, and mitigate
those inequalities in a systematic manner in order to avoid unintentional algorithmic consequences.

A network fairness framework is proposed, premised on topological and temporal features of social interactions
that shape the formation and evolution of inequalities. These features include (1) people have
multiple and correlated attributes that determine how they identify with groups and interact with others,
(2) people belong to a variety of social groups with different sizes, hierarchies, and historical precedents,
and (3) interactions between people evolve over time in a hybrid space of society and algorithms.

To this end, I will develop a suite of dynamical complex network models of inequality that are driven by
social theories (e.g. homophily, intersectionality, consolidation) and calibrated and evaluated with big data
and network experiments. This will allow us, for the first time, to investigate inequalities that arise from
network-based algorithms in a systematic manner. More importantly, I will devise a novel methodology of
network intervention, a set of data-driven principles for tackling network inequalities in a broad range of
applications. Finally, based on the models developed in this project, I will create a cutting-edge interactive
software – NetFair – to visualize and forecast the evolution of inequalities and implement various fairness
criteria. The software will contribute towards bridging the gap between research and policy applications
in academia and industry.