Final Report Summary - THINKBIG (Patterns in Big Data: Methods, Applications and Implications)
The ThinkBIG project was aimed at exploring three different aspects of the growing field of Data Science: new methodologies, new application areas, and possible implications for society. Its fundamental idea was that none of these three directions should be developed in isolation, and they should instead be pursued in combination. The approach paid off. We developed novel data science methods to address questions in fields as distant as history, social sciences and medicine, learning about the potential of this field to contribute but also disrupt. We identified critical challenges for society that might come from data technologies, we investigated them with philosophers and legal scholars, and then we engaged colleagues, the public and policy makers with the outcomes. All this while also proposing technical solutions to various problems that emerged along the way.
For example: AI methods trained on media content are known to absorb gender bias from their very training data. We have measured this bias in media content, we have measured it in the model that are created by AI algorithms, and then we have proposed methods to remove or mitigate it. Then we have engaged the public with outreach videos and communicated this to policy makers.
Interdisciplinary work included collaborations with neuroscientists to interpret various periodic patterns found in social media content; collaborations with historians to understand macroscopic trends in newspaper content; and with philosophers, to frame important questions that emerge from the deployment of AI infrastructures into society, for example persuasion, psychometrics and social regulation issues.
Dissemination work included workshops for researchers, public lectures and online videos for the general public, and direct engagement with policy makers both at UK and at EU level.
For example: AI methods trained on media content are known to absorb gender bias from their very training data. We have measured this bias in media content, we have measured it in the model that are created by AI algorithms, and then we have proposed methods to remove or mitigate it. Then we have engaged the public with outreach videos and communicated this to policy makers.
Interdisciplinary work included collaborations with neuroscientists to interpret various periodic patterns found in social media content; collaborations with historians to understand macroscopic trends in newspaper content; and with philosophers, to frame important questions that emerge from the deployment of AI infrastructures into society, for example persuasion, psychometrics and social regulation issues.
Dissemination work included workshops for researchers, public lectures and online videos for the general public, and direct engagement with policy makers both at UK and at EU level.