Is AI Causing a Credibility Crisis in Networking?

The Non-Paper Session will take place at Room Auditorium.

The purpose of this session is to raise awareness about an important yet often underappreciated challenge in the application of AI/ML for solving problems in science, in general, and networking, in particular. In today’s era of AI/ML, the well-known “reproducibility crisis” in science is increasingly accompanied by a less visible but equally significant “credibility crisis.” By drawing on examples from networking, we highlight that among the relatively few reproducible studies describing AI/ML-based solutions, even fewer provide evidence that their trained models are “credible,” i.e., they can be trusted to perform well not only in the original training setting but also in new and untested environments.

We will explore the underlying factors contributing to this credibility gap, reflect on why model generalizability is vital for the practical deployment of AI/ML, and build on the arguments proposed here on how our community might address these challenges constructively. This session aims to outline an agenda for a future of networking that can fully benefit from the potential of AI and ML and is grounded in credible models that are not viewed with skepticism and suspicion.