Re: Einstein, Edison, & IQ

From: Michael Wilson (
Date: Fri Nov 18 2005 - 04:21:52 MST

justin corwin wrote:
> SIAI contains basically Eliezer (who isn't known to have
> done any software development) and M Wilson, who has, and
> is listed as an associate researcher, and seems to be
> fairly active. I have no details, but his blog and such
> doesn't mention any other engineers on staff that I recall.

The SIAI does not currently have any paid staff doing AGI
design or implementation. It's a little under two years
since I started working full-time on AGI technology
development, but this has been necessarily been privately
funded. Besides myself, my company currently only has one
part time software engineer and two part-time sales/business
development staff. Funding permitting, I intend to hire a
couple of full time software developers next summer.
Unfortunately I rate the chances of finding anyone capable
of doing serious AI development through standard recruiting
as essentially zero, so these staff will be working only on
the commercialisation aspects and supporting software. As I
have often stated, I am not currently attempting to build a
complete AGI, but I am working on a set of components that I
intend to be a sufficient basis for such a project (such as
possibly the 'official' SIAI project) in the future. I would
certainly hire any decent AGI-qualified researchers that
are available in the UK, but I'm not aware of anyone suitable
at this time.

On the issue of project size, I'm not sure (and I don't think
anyone is) what the shape of the staff_count/researcher_
quality/completion_time/success_probability surface looks
like, particularly as I ascribe so much of the success
probability to whether the team has the right general attitude
and methodology. I expect that serious manager expertise is
needed to prevent the staff_count/success_prob curve peaking
quite early. But if I had unlimited funds, I'd aim for about
five core researcher/designers and perhaps twenty supporting
engineers, and see how that works out in practice. You can
have as many additional theorists as you like (e.g. FAI
theorists) as long as they're not interfering with development
work on a day-to-day basis.

> When the microsofts and Sun's of the world see exciting
> results, it won't take much for them to replicate and
> overtake small firms with anything but a colossal head
> start,

That depends on several factors, such as the opacity of the
design, its closeness to existing techniques, the level of
detail publically disclosed, where the code is actually
deployed (e.g. network versus local application delivery)
and critically how good the competitor's industrial espionage
and startup's security procedures are.

> unless someone manages to develop much AI theory in relative
> isolation, and keep from having to release much or any of the
> software to public attention, which seems extremely unlikely.

It's not so unlikely for researchers that have a clue and are
aware of the high stakes and dangerous players involved. Just
look at James Rogers, who insists that he's going to keep any
AGI successes quiet and refuses to reveal anything about his
design. I'd guess that Novamente and A2I2 have given this at
least some thought, if only from the IP protection angle (which
is basically essential to any IT business these days), but
right now I wouldn't give them a significent chance of avoiding
a serious attempt to steal their code (and/or have their staff
hired away).

> a theory gap (proprietary or simply advanced software theory
> that is difficult to discern by untrained, or
> reverse-engineers(also possible)),

IMHO this is highly likely for any AGI design that actually
works on current hardware. AGI is very hard to understand,
extremely difficult to explain, and I'm willing to bet damn
near impossible to reverse engineer without the full source
and development documentation (and pretty difficult and time
consuming even then).

Plus appropriate software patents /may/ slow down commercial
competitors in the US. They will of course do nothing against
rest-of-world competitors and military projects.

 * Michael Wilson


This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:53 MDT