After reading Benito's rephrasing of the point...

Your identity/role, in near mode, seems to tell your System 1 what to be thinking, and your System 2 can change what identity/role you're running at a given time.

I immediately had the same reaction as Lumifer:

There is a trade-off, of course: your reaction is fixed by training and it's not necessarily appropriate to the specific situation.

A carefully crafted role (or set of roles) that allows your System 1 to function the way you want it to in most circumstances, while being a huge improvement for most, will still have problems if it is restricted only to System 1 responses. The appropriate reaction is sometimes to switch to System 2 where one can go through all the methods of rationality, not to just perform certain habits or responses that rationality has shown to be better than the default.

At first, this seemed contradictory to me, as many of the high-value things on LW seem to be System 2 processes. But then I realized that many of the habits of rationality involve training your System 1 to fire up your System 2 in the right circumstances and it seems that this could be crafted into the Roles that we try to adopt. For example, not "flinching away" from an Ugh Field. Noticing when you are about to do it and kicking over to System 2 to do something about it. If you see yourself as the kind of person who does this (and practice of course) then I don't see why that can't be a part of the Roles that you train.

Excellent article Eneasz!