VIrtual GuardIan AngeLs for the post-truth Information Age

Introduction

This project is motivated by the hypothesis that we are approaching a post-truth society, where it becomes practically impossible for anyone to distinguish fact from fiction in all but the most trivial questions. A main cause of this evolution is people’s innate dependence on cognitive heuristics to process information, which may lead to biases and poor decision making.

Cognitive Biases and Networked Society

These biases risk being amplified in a networked society, where people depend on others to form their opinions and judgments and to determine their actions. Contemporary generative Artificial Intelligence technologies may further exacerbate this risk, given their ability to fabricate and efficiently spread false but highly convincing information at an unprecedented scale.

Technological Impact

Combined with expected advances in virtual, augmented, mixed, and extended reality technologies, this may create an epistemic crisis with consequences that are hard to fully fathom: a post-truth era on steroids.

Project Goals

In the VIGILIA project, we will investigate a possible mitigation strategy. We propose to develop automated techniques for:

1. Detecting triggers of cognitive biases and heuristics in humans when facing information.
2. Assessing their effect at an interpersonal level (their effect on trust, reputation, and information propagation).
3. Evaluating their impact at a societal level (in terms of possible irrational behavior and polarization).

Methodology

We aim to achieve this by leveraging techniques from AI itself, in particular Large Language Models, as well as by building on advanced user modeling approaches from the past ERC Consolidator Grant FORSIED.

Integration and Ethics

Our results will be integrated within tools that we refer to as VIrtual GuardIan AngeLs (VIGILs), aimed at news and social media consumers, journalists, scientific researchers, and political decision makers. Ethical questions arising will be identified and dealt with as first-class questions within the research project.