Re: Debian-AI
From: Michael Tautschnig <mt@debian.org>
Subject: Re: Debian-AI
Date: Wed, 2 Jan 2008 09:38:34 +0100
> [...]
> >
> > I am requesting comment on this approach, review of my project, and
> > looking for guidance. Though I have tried for 7 years, I have not
> > been able to make the breakthrough into the community (most likely
> > owing to a lack of social skills). I tried to discuss this project on
> > the Debian channels, where they suggested I post to the Debian-devel
> > mailing list.
> >
> > Would there be any interest in a Debian-AI sub-project?
> >
>
> Hmmm, I kind of fail to see the exact point: Are you really proposing a
> sub-project related to AI software (like Debian-Med is for medical tools), or
> are you rather proposing an AI approach to selecting and packaging any software?
> Judging from the beginning of your post I was tempted to think of the latter,
> but suggesting an AI sub-project seems something pretty different from that.
> Further, your repository mainly contains AI-related tools, but also some
> unrelated software.
>
> Could you/somebody else clarify this a little?
>
> Best,
> Michael
>
No I'm not really proposing a sub-project related to AI, I should have
titled the subject "the FRDCSA project", not "Debian-AI" to avoid
confusion.
One would not start a sub-project without lots of interest. Andreas
Tille points out that joining the Custom Debian Distribution effort
might make more sense than a sub-project.
As far as proposing an "AI approach to selecting and packaging any
software", yes I am proposing that.
I propose to work with people on tools that:
Create a Comprehensive Software Ontology (CSO) in Semantic Web's OWL
format of existing software, including license information, project
capabilities, all sorts of information about the projects, and make
tools that use this information. A large number of projects can be
queued from the FLOSSMole data.
Create a webspider that can automatically index metasites like the
ones Isabel Drost called to attention
(http://ml-site.grantingersoll.com/index.php?title=Existing_Learning_Tools,
http://mloss.org/software/), (detecting duplicates using alias
detection and the same software they use to differentiate people), and
begin adding those repositories to the CSO.
Develop additional tools that enhance the coverage of automated,
scripted and data-driven methods of transforming from upstream sources
to source packages (a.k.a. packaging the software).
The CSO must manage software licensing information and determine which
software is appropriately licensed for packaging into different
archives, whether it is free, non-free or simply a package that
generates an installer. For software which is not licensed compliant
to DFSG, systematically petition the authors to rerelease the software
under a compatible license.
Create a separate package archive for rough quality automated packages
that users may use at their own risk, as well as begin manually
tweaking these packages for inclusion to the main Debian repository.
Regarding the difference between AI-related tools and unrelated
software, IMHO any software or dataset may be considered as a
component to a weak-AI because of the correspondance between
proof-systems and programs (Curry Howard Isomorphism). While this may
be an erroneous viewpoint, the main point is that we should package
software relevant to increasing the capabilities of the system. Right
now one great provider of improved capabilities are the unused AI
systems, but there are many other sources as well.
I hope this has gone some way towards clarifying exactly what I
propose. Thank you for taking the time to consider my proposal.
Sincerely,
Andrew Dougherty
Reply to: