AI is not ready for central bank policymaking

In one speech At a workshop this week, Edward Robinson, deputy chief executive of MAS, acknowledged that central banks had to answer “difficult questions about our collective failure to predict the persistence of inflation post-pandemic, which in turn has called into question the usefulness of our models.” . .

While AI is touted as a potential tool to improve modeling, Robinson advises caution, pointing out that LLMs are “not yet capable of providing credible explanations for their own predictions.”

“By and large, AI models currently lack the “clarity of structure” that makes structural models useful to policymakers. To quote former Fed Governor Laurence Meyer, model-based forecasts must “start with a paradigm and end with a story.” No The ability to articulate a vision of how the economy works or distinguish between competing narratives, AI models cannot yet replace structural models at central banks.”

For Robinson, the best way to integrate AI techniques into central banks’ modeling toolkits is to use them in satellite models that complement core structural models.

However, while central banks need to prepare for the further development of GenAI as a “general-purpose technology,” it may require “some prudent interventions in advance to influence the transition path.”