Ah! Creativity! In that case, rather than just picking random bits of memory, why not open a (random?) file? You can list the files in your favourite directory, pick one of them randomly, open it as a binary file, load lumps of it to memory that does belong to you (hence no segmentation faults or complaints of any kind) and interpret the data in any way you want.
Of course the other thing you can do is allocate yourself a chunk of memory and merely look at it without initialising it, but if you're unlucky it might just be a load of zeroes.
If used data for an artwork in any way in which the data might be recognised (e.g. ascii dumps etc.) you must make sure that you aren't putting on public display information that might upset someone; you might, for instance, accidentally come across passwords, fragments of e-mails, e-mail addresses etc.; [for any voyeurs reading this, you probably won't find anything interesting at all. Murphy's law predicts that if you are, you won't, and if you aren't, you will.]
If you're audiolising you could just tape a dial-up modem dialling... Incidentally there is a long history of this. The first ever computerised payroll was that of the Lyons corner tea-shops, and I believe their (valve-and-mercury-tank) machine also contributed noises to the BBC, I think it was a program involved in weather forecasting, but might be wrong.
I'd avoid any file that's been compressed. Compression (by definition) makes the data look more random, and most of us know what random noise looks like and sounds like.