Re: AI:This is how we do it

From: Eliezer S. Yudkowsky (
Date: Mon Feb 18 2002 - 20:29:37 MST

Harvey Newstrom wrote:
> Eliezer S. Yudkowsky wrote,
> >
> > we have determined that SIAI's current
> > priority is to place the ideas in a form where a solid project could be
> > initiated immediately on funding being available - i.e., more
> > internal design documents, not more publications, not more evangelism.
> This would be very interesting to see. Do you have any of these design
> documents done yet? I would probably pay money for a chance to review these
> to get an "inside" view of your work. I would gladly sign any
> non-disclosure agreements as required to protect your interests.
> If these designs are not done yet, do you have a prioritized list of what
> design documents need to be developed, possibly with a project-plan
> timeline?

Alas, we don't have anything human-readable enough to charge for. Right now
I'm generating all the design notes I need as a preliminary to writing up
human-readable design documents. Under the circumstances, it seems better to
try and prototype the design documents themselves before actually writing
them, as 'twere. Goals for the design notes: A complete outline of the
design subsystems of a general intelligence on the various levels of
organization; a grouping of the outline into major tasks to be divided up
among the programmers (or assigned one to a programmer, if we can somehow
afford to hire that many people); high-level walkthroughs of some example
cognitive processes; a list of milestones for the first few stages; a
breakdown of submilestones along the way to the first stage. (Actually
outlining the system modules and top-level classes is something that I'd
rather do with the rest of the programming team, once the programming team has
been pulled together.) Right now, as I said, I'm working on the formal paper,
although of course I've gone on with the design notes during that. Currently
the design notes are about 250K in length and include a very large (and
growing) subsystem outline. I hope to be done with the paper soon, after
which the primary task goes back to design. The next step in growing the
subsystem outline, when the paper is done, will be to try the walkthroughs.
When the subsystem outline is complete it'll be possible to do the
milestones. Then comes writing it all up. At least that's the current plan,
which is rather highly susceptible to being obsoleted by events.

-- -- -- -- --
Eliezer S. Yudkowsky
Research Fellow, Singularity Institute for Artificial Intelligence

This archive was generated by hypermail 2.1.5 : Fri Nov 01 2002 - 13:37:40 MST