Study examines risks companies face when relying too heavily on AI systems

A new framework called Counterfactual Resilience Framework (CReF) aims to help organizations assess the risks associated with over-relying on artificial intelligence systems. Developed by Vedant Das Swain from NYU Tandon and Koustuv Saha from the University of Illinois Urbana-Champaign, CReF encourages companies to consider the potential consequences of AI becoming unavailable.

The framework presents a hypothetical scenario with three components: a workplace where AI is used, an event that causes AI to be inaccessible, and the aftermath of this disruption. By treating the absence of AI as a stress test, CReF encourages organizations to identify dependencies and vulnerabilities that may not be immediately apparent during normal operation.

Three key vulnerabilities are explored through CReF: the uneven distribution of AI benefits across different jobs, the potential erosion of skills due to reduced practice, and the difficulty in restoring disrupted workflows. The researchers argue that while AI adoption should continue, it is equally important to plan for the worst-case scenario.

The framework does not predict specific outcomes or assign resilience scores to organizations. Instead, it serves as a generative analysis tool to help stakeholders understand hidden dependencies and potential safeguards needed for a future where AI may not always be available.