Forecasting the US federal funds rate is important since it impacts global financial conditions, and in turn the real economy. This working paper examines whether artificial intelligence (AI) models could forecast policy rates accurately by inferring the implicit decision rules the US Federal Open Markets Committee (FOMC) follows when setting the policy rate. We examine two recurrent neural networks (RNNs), long short-term memory (LSTM) and gated recurrent unit (GRU), and a large zero-shot language model (LLM) capable of combining numerical economic data and interpreting qualitative information. The forecasting performance of the three models is good, especially during the zero lower bound and early COVID-19 periods. The policy rules implied from the RNN models suggest that the FOMC might have been overly accommodative during the COVID-19 pandemic. Since 2020, the models suggest a deviation from prior monetary policy patterns, with the FOMC adjusting rates less frequently than predicted by the model-implied policy rules. These findings suggest that incorporating AI insights could enhance our understanding of and ability to predict future Fed rate decisions.