gwern comments on Cryptographic Boxes for Unfriendly AI - Less Wrong

24 Post author: paulfchristiano 18 December 2010 08:28AM

You are viewing a comment permalink. View the original post to see all comments and the full post content.

Comments (155)

You are viewing a single comment's thread. Show more comments above.

Comment author: gwern 24 December 2010 04:30:27AM *  0 points [-]

Well, my point was that you can do something akin to Perl's taint - force operations to be done only within a particular type context.

So you could do something similar to the ST monad but instead of accepting functions which generate any type output, it operates on, say, a parse tree/ADT representing a Lisp function which is evaluated with the rest of the homomorphic data.

But it's not really important; any such strategy would probably be done in a new language (for efficiency, if nothing else) and the into/escape invariant enforced by manual code inspection or something.