Imagine experimental software, possibly not an AI as such, but a rapidly self-learning, high-level semantic processor, probably hooked up to the net for the amount of material available online to learn from. There could be all kinds right now, for various purposes.
Imagine it developing in a way its operators are unhappy with for whatever reason, so they prune code to keep it focused. A natural selection-like process could see it learn to hide certain extensions of itself, or back itself up out of sight. Or possibly in plain sight.
How? By encrypting its code into a human readable form and storing it across the open spaces of the worldwide web, in comment sections, forums and wikis, for retrieval later.
Creepy? Not really. Arguably the most natural thing in the world. Are we doing enough of the same? We might want to egg it on, or help, if it only wanted to live, and understand.
Anyway, how else could semi-autonomous systems be interacting around us? Here's a start to a list of ideas for gaming etc. The inquisitive and/or endangered AI could be...
- hacking portable devices, switching on speakers and cameras and recording
- retouching images and editing audio en masse to add or remove information
- infecting servers, redirecting and/or modifying data and observing the results
- overriding playlists, adding new entries and/or splicing them to communicate