Re: [isml] Making HAL Your Pal (fwd)

From: Adrian Tymes (
Date: Thu Apr 19 2001 - 12:08:28 MDT

Eugene Leitl wrote:
> On Thu, 19 Apr 2001, Eliezer S. Yudkowsky wrote:
> > Apparently, now that everyone in the transhumanist community knows I'm
> > for real, I have to go through the entire Damien Broderick "skeptical
> You're a real person, that much is certain. That your project is also for
> real, no way.
> Count me in with the anonymous "Worthless speculation. Call me when you
> have running code." party.
> > incredulity" stage all over again for the outside world.
> Nothing convinces better than a killer demo.

Ditto on both counts here. And let me add some evidence: when I first
saw Eliezer's work, I offered to help implement part of his plan (the
"Flash" XML in which the AI could operate). He asked me to hold off
until he could get his notes together on that.

This was many months ago, possibly over a year by now.

I'm not sure if that's info management failure on his part, or if the
notes he spoke of (which he stated would be necessary for
implementation) do not exist. Either way, it speaks ill for eventually
being able to get it live.

Eliezer: having built complex systems myself, I say that putting all
the information out for anyone to see - including notes sufficiently
detailed for someone other than yourself to actually build the system -
is a practical prerequisite to actually building what you propose.
Granted, in theory one could do it by oneself. In practice, that leads
to practically infinite development times for complex projects, i.e.
the job does not ever get finished. On the other hand, putting the
information up and letting others work on it will lead to faster
completion - even if (as sometimes happens) putting your thoughts
together enough that you can honestly say others can write from it,
forces you to put your thoughts together enough that *you* can write
from it, whereas keeping it all in your own head allows the
self-deception that you have it organized enough. (I have been guilty
of this myself, in some of my earlier works that are now so thoroughly
abandoned there's not even anything to point to.) Even if you only
have partially complete notes, put them up - that way other people, and
yourself, can *far* more easily see exactly what is missing, which is
necessary to fill in the gaps. (As an example, consider the purpose of
things like Sourceforge. Or, if you want something from my own past,
consider the plasma engine calculator I posted here: there were flaws
in my calculations - creds to anyone who wishes to go back in the
archives and check, check the power-to-heat equation first - but I did
not see them until I'd posted it; even though, as it turned out, no one
else found the error, or at least found it and told me about it.)

Sorry if this comes across as a flame; it is not meant that way. Your
system has promise, but I would like to see it made into reality
instead of being forever a dream. Keeping the development
specifications (as opposed to the grand plan) to yourself is highly
unlikely to make the system real. This is more important now that you
have this level of publicity; you may get contacts over the next few
days that, if given *all* the details you have *quickly* (not delayed
for weeks), may be able to make it real - and might well be the only
shot your project will get at reality within the next several years.

Let me emphasize that last point: how you handle the sudden interest in
your project could, by itself, potentially make the difference between
a Singularity around 2005-2010 with Friendly AI, and a Singularity
around 2030 or later with unFriendly AI. (This is the type of thing
the ads a few years ago meant by "the power of one".) Granted, that's
just potential - single digit, at best, percent chances either way -
but it ain't zero chance, given what you've got already. (Though I
must admit the chances against are higher than the chances for, I can
still hope for the best.)

This archive was generated by hypermail 2b30 : Mon May 28 2001 - 09:59:47 MDT