This appears to just be talking about using ML to estimate heterogenous treatment effects via standardization/g-formula. It's just another kind of regression. There are other methods aimed at finding causal relationships in observational data but that doesn't appear to be what this is talking about