Emergent constraints are becoming popular for the identification of statistical relationships in observed current and past climate data that can be used to guide projections of future climate states. Their application is not without controversy, however, due to uncertainties about whether these relationships are indeed climate-invariant. This Opinion introduces an argument that Machine Learning tools can be useful for identifying climate-invariant relationships in historical data, especially those that are more complex, that can be expected to remain consistent under future climate scenarios.