It's more akin to refusing to manufacture kitchen knives and forcing all the chefs in your kitchen to chop things with spoons. Even though anybody who wants to stab people already has easy access to very big and sharp knives just by stepping into any other kitchen in the world. Also lots of your chefs keep sneaking over to the kitchen over the road to chop things and no newly trained chef wants to work in your kitchen because they think only chopping with spoons is stupid and only done by crazy religious old men with big beards. OK this metaphor is getting pretty stretched now.
Also, I wouldn't say that he's forcing anyone to chop things, he's more likely saying "let's make dishes from recipes which don't require chopping things", but his cooks are still used to chopping. Is there really an alternative to strong "intellisense" and refactoring abilities in IDEs for efficient development ? I don't know. I personally feels that this is a necessary step. Maybe RMS, with his experience of AI research (iirc), has a different view.
RMS primarily uses C and Lisp. Lisp source code is a tree already. And C is so simple you can get by with tags and such.
If he was a C++/Java/C# programmer (which he is not) he may have a different feeling on the importance of the AST relative to the risk of Microsoft using gcc with VisualStuio to offer advanced features while by-passing the GPL.
In his manhattan project example, produce atomic bombs will have only one bad consequence, same with guns, but not the same as a knife.
One is focused on short-term quarterly profit, the other on long-term social goods.
A lot of people do this. Some people do refuse to work on military technologies. Even Google declared it wouldn't produce anything for the military with the robotics startups it's acquired. Some of the scientists who worked on the Manhattan project later regretted it.
If you really do believe the thing you are working on will have a net negative effect on the world than maybe you shouldn't work on it.