From: Joshua Fox (joshua@joshuafox.com)
Date: Tue Nov 06 2007 - 09:51:55 MST
Under "I'm-sure-someone-must-have-done-this-before":
What about the idea of an morality simulator. Just as computer models of
weather or car crashes -- however imperfect -- allow researchers to test
their assumptions, why not do this for morality?
You would use various simulated worlds, whether a symbolic model, simple 2D
worlds, and or full Second-Life-style shared online worlds.
You could assign a closed-form morality-evaluation functions directly, or
take an implicit function from the input of users in shared online worlds.
At a first stage, we'd simply evaluable the morality of actions as done by
software agents or by human-controlled online agents. At a second stage,
we'd move on to letting agents choose actions according to moral goals. It
would be interesting to see how moral goals interact with other goals in a
given agent.
This would have many limitations, but it seems like a great tool for
experimentation, including, ultimately, for Friendly AI researchers.
Joshua
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:01:00 MDT