Really good post here on the dream of automatic inference: https://statmodeling.stat.columbia.edu/2025/05/28/ecologists-endless-quest-for-automatic-inference/
The author says “ The other option to me is that you pick your few best-guess damn variables — the ones you can make predictions about and describe the functional relationship of them to your response variable(s) and you put those in your model. Maybe you fit a few models, but not endless models. In my experience, the first step in this process alone (picking those variables) gains me way more insights than any model comparison ever has. Why? Because it’s the opposite of automatic inference. It requires me to think.” which I think is really really spot on.
Something they say earlier is that ml is for out of sample prediction and we are interested in understanding is interesting to me as for real time work like we do it’s not always clear what we want. I don’t think there is great evidence these ml approaches are doing great for outbreaks etc though (interested to be proved wrong)
I also really like this statement about teaching “ They need to see how interconnected all the inference methods are and what aims each one works well on for now (and not) and be prepared that that might change.” Which I think also applies beyond just inference to a lot of things we are thinking about (I.e delays, nowcasting, forecasting , “ situational awareness”).
An example we perhaps can be fully automate or at least partially so is delay estimation if the mission is to characterise a distribution for onward use as something parametric or a mixture of parametric approaches which Oswaldo has been doing some thinking on. Even here though I think I much prefer stopping thinking and trying a few candidates as the author suggests.