Dino Pedreschi

Responsible AI for social media governance – the GPAI experience

Social media platforms are one of the main channels through which AI systems influence people’s lives, and therefore influence countries and cultures. The number of social media users worldwide is estimated at around 5 billion at the current moment, or 60% of the world’s population. Crucially, the experience of a social media user, on any given platform, is heavily influenced by AI systems that run on that platform.

• The central component of any social media system is a content feed of items presented to the user. The feed for each individual user is curated by a recommender algorithm: an AI system that monitors each individual user’s interaction with platform content, learns what they like to engage with, and then gives them ‘more of the same’. This ability to tailor content to users is what gives social media systems their vast appeal. But the mechanisms through which recommender algorithms learn raise important questions that remain to be answered. Our project has proposed concrete measures for the governance of recommender algorithms, through the study of their effects on platform users.
• Another pervasive AI influence on social media platforms is in their content moderation processes. Content moderation has to be supported by automated tools, given the huge volume of content that is posted, and AI content classifiers are the key tools that companies deploy. Again there are concerns about the processes through which these classifiers are trained. Our project is trialling a mechanism for training classifiers outside of companies, in a semi-public domain, which we think may offer a better model for their governance.
• A final way in which social media companies channel AI influences is through their role as content disseminators. The revolutionary advances in generative AI that have taken place over the last years allow ordinary citizens to create, and distribute, AI-generated content, through text generation tools like ChatGPT, and image generation tools like MidJourney. Alongside the many productive uses of such tools, there are concerns that they will facilitate the production of disinformation, potentially destabilising political processes, and other information ecosystems. Our project has made a proposal for the governance of generative AI tools, which has had considerable traction within EU policymaking bodies, and also within the US Senate.

In this talk I will briefly account for the work and results of GPAI “Responsible AI for Social Media Governance” project, that I co-lead together with Alistair Knott (Victoria University Wellington, NZ),  along the three mentioned lines.