I like this analysis! Makes the classification and pros and cons clear.
Just to add to it: I think steelmanning is useful as being 2 steps downstream of ITT. This is because ITT is a good model of someone else’s views, not of their underlying logic or motives or desires, mostly because most of us are not very good at introspection. ITT reproduces this surface-level reaction without necessarily analyzing the reasons for it, but just based on the aggregation of the collected data. The intermediate step between ITT and steelmanning would be a “gears-level” model of someone. Once you have that model, you can analyze it and extract some useful information from it, by refining the model to be more self-consistent and reflecting the parts of the territory you may have missed.
I like this analysis! Makes the classification and pros and cons clear.
Just to add to it: I think steelmanning is useful as being 2 steps downstream of ITT. This is because ITT is a good model of someone else’s views, not of their underlying logic or motives or desires, mostly because most of us are not very good at introspection. ITT reproduces this surface-level reaction without necessarily analyzing the reasons for it, but just based on the aggregation of the collected data. The intermediate step between ITT and steelmanning would be a “gears-level” model of someone. Once you have that model, you can analyze it and extract some useful information from it, by refining the model to be more self-consistent and reflecting the parts of the territory you may have missed.