An Oracle standard trick

An idea I thought I’d been men­tion­ing to ev­ery­one, but a re­cent con­ver­sa­tion re­veals I haven’t been as­si­du­ous about it.

It’s quite sim­ple: when­ever de­sign­ing an Or­a­cle, you should, as a de­fault, run it’s out­put chan­nel through a prob­a­bil­is­tic pro­cess akin to the false ther­mo­dy­namic mir­a­cle, in or­der to make the Or­a­cle act as if it be­lieved its mes­sage will never be read.

This re­duces the pos­si­bil­ity of the Or­a­cle ma­nipu­lat­ing us through mes­sage con­tent, be­cause it’s ac­tion as if that con­tent will never be seen by any­one.

Now, some Or­a­cle de­signs can’t use that (eg if ac­cu­racy is defined in terms of the re­ac­tion of peo­ple that read its out­put). But in gen­eral, if your de­sign al­lows such a pre­cau­tion, there’s no rea­son not to put it on, so it should be de­fault in the Or­a­cle de­sign.

Even if the Or­a­cle de­sign pre­cludes this di­rectly, some ver­sion of it can be of­ten be used. For in­stance, if ac­cu­racy is defined in terms of the re­ac­tion of the first per­son to read the out­put, and that per­son is iso­lated from the rest of the world, then we can get the Or­a­cle to act as if it be­lieved a nu­clear bomb was due to go off be­fore the per­son could com­mu­ni­cate with the rest of the world.

No comments.