Note: This project has been stopped while I look for a more promising mailing list. The Peoplelink AI list has become inactive.
The program is written in Prolog.It does a very simple scan of emails posted on the AI mailing list looking for keywords. When it finds particular combinations of keywords it will post a "reply" to that email. The "reply" will be a piece of "canned text" related to the combination of keywords it found in the email.
GL7 has a couple of hundred pieces of canned text to get it started. Quite a few of these relate to the architecture of the program itself. For example, there is a piece of canned text related to its (limited) ability to introspect and another piece related to how it can augment its own database.
I am looking into ways in which it can automatically augment its "understanding" of AI by looking at sources on the web. This is a bit difficult because there is a lot of MIS-information on the web, but I have located a couple of reliable sites.
GL7 will not post the same piece of "canned text" if a particular combinations of keywords is detected a second same, though it may post another piece of text triggered by the same keywords. (However, it may start posting the same pieces of canned text after 12 months have elapsed since the first posting.)
GL7 will, currently, post at most one email a day to the list.
GL7 will generate several candidate responses each day and only the most interesting will be posted. Interestingness will be determined by a neural net initially trained by me unless a better way can be found.