No, I don’t think this is correct. See here and the link to Tristan Mahr’s explanation in my first response to my own question. In Tristan’s article, there is some explanation of what the ‘population-level effects’ in brms are as relates to these smooth terms, but I think they are the coefficient for one of the ‘natural’ parameterized basis functions. Thus, they really have no direct interpretation.
Are you trying to determine which terms should be included at all? or which terms don’t really need a smooth? You could use LOO to compare the models. I think if you are trying to decide which out of a bunch of terms to include in the models based on predictive performance, then maybe projection predictive variable selection might be the way to go, but I am not really familiar with this.
As @MilaniC said, making plots from predictions is the way to go to understand the results from models with smooth terms (at least that is what I do). If you want to understand the slopes of the smooths, you can take the first derivative and plot this with uncertainty intervals. See here for the concept, and for brms I think you would need to use posterior_smooths() for reasons described here.