Earth System Models such as E3SM are released with single set of “tuned” values for uncertain parameters, artificially constraining the uncertainty in the model predictions. We use an automated calibration method to find multiple alternative tunings for uncertain parameters in the atmosphere model in E3SMv3. The alternative tunings are derived by: 1) creating a surrogate model that uses machine learning to emulate E3SM output as a function of uncertain parameters, and 2) using the surrogate model to estimate input parameters for desired E3SM output. We use this method to select alternative tunings that predict present-day climate with acceptable fidelity but diverge in the prediction of climate sensitivity.