Not far from Elsinore, adolescents are pondering the messages of ghosts again. Here is a Flash site for the project (somewhat unhelpfully laid out, in my opinion), and here is an overview of the project in .pdf form. These "ghosts" are interactive agents that learn and compete for users' ITUs ("interest token units") in order to increase their fitness, measured in terms of "vialence" ("viability valence").
A key quote from the paper:
The notion of ghosts has been chosen because of traditional characteristics of ghosts found in the popular literature and in folk tales:
1. Ghosts are mostly invisible or only vaguely visually manifested
2. Ghosts are often bound to a specific location which often has a very special relation to the ghost
3. Ghost owe their twilight status to some unfinished business and they are therefore active and striving
4. Ghosts only appear when called upon or if they feel an urge to manifest themselves
These features are heuristically very interesting for developing functionally satisfactory agent based assistance while keeping the technical requirements at a minimum.
After a brief scan of their site, I can add some other characteristics of ghosts that are useful from the point of view of system designers:
1. Ghosts are not always helpful or well intentioned.
2. Ghosts don't always tell the truth (or even know what it is).
3. Ghosts are often annoying even when they are entertaining.
If you're not familiar with this literature, you may enjoy reading about Cobot, an adaptive conversational agent who was purportedly artificial rather than metaphysical.
It's not really a surprise that Danish AI researchers didn't think to include Hamlet's ghost among their cast of meta-characters, but we can hope that some clever playwright will re-present Hamlet in a setting modeled on this new campus environment. In particular, I feel that Thin Lizzy would be able to provide helpful commentary on Shakespeare's original.
[via slashdot]
Posted by Mark Liberman at March 3, 2004 01:22 PM