Meta tries to explain its AI with an Instagram algorithm decoder

Everybody wants Facebook and Instagram to explain “the algorithm.”

But their parent company, Meta, doesn’t use just one algorithmic system or one machine learning model or one piece of technology to rank or moderate or label Facebook and Instagram posts. There are a variety of AI and non-AI based tools at play, sometimes in collaboration with actual human moderators or decision-makers. It means some standard approaches to explaining AI don’t necessarily work to explain Meta’s ever-elusive algorithms.

It’s why the company said today it is using a new approach to explaining its algorithmic systems to people. Meta’s Responsible AI team developed “system cards” and a related prototype tool they said, “has the potential to provide insight into underlying AI system architecture and help better explain how these systems operate.” A pilot system card is intended to show how technologies used to determine Instagram feed rankings work, and states: "When you open or refresh the Instagram app, the feed system ranks posts you haven’t seen yet from accounts you follow, based on how likely you are to be interested in each post.”

System cards might apply for other algorithmic processes such as those used by the company to translate languages, detect fashion items in images, flag harmful content, or for speech recognition. The system card concept was developed as a hybrid of other more commonly-used approaches to explaining AI including model cards, which detail how models were built and how they are intended to be used, and Datasheets for Datasets, a framework for showing information about the data used to train machine learning models.

But don’t expect Meta to give away the keys to its prized (yet also maligned) algorithms. The company made a point in its post about system cards to say the approach might not be able to illuminate how highly complex systems work. And, the company said might not reveal information in system cards that could be a security risk or expose systems in ways that allow people to reverse-engineer them.