| From: | "Robert A. Hayden" <hayden@*******.MANKATO.MSUS.EDU> |
|---|---|
| Subject: | Updates galore |
| Date: | Tue, 10 May 1994 02:10:11 -0500 |
>>>>>[Sorry I've been out of touch all day, but I spent most of the day
trying to restore what few brain cells I have left to working order.
MONICA packs quite a whallop!
Once I was pretty much back on my feet, I took a look at the damage
MONICA did. CLU was running in a six-system SuPAR (Super Parallel)
network in order to keep it fast and to give it the evaluation cycles.
It took me almost 2 weeks to get the 100,000+ processors talking to each
other, let alone working together. (As an interesting side-note, if this
had been a single computer, it would have been the 103rd most powerful
(known) computer in the world).
Unfortunately, all of the systems are pretty much dead. The feedback
that MONICA sent in when she forceably downloaded CLU pretty much fragged
the whole thing. I may be able to salvage around ten percent of the
electronics or so, much of it's really beyond repairable.
That's kinda a shame, though, because if it was still working, I could
fire up another CLU construct (ah the joy of backups). In any case, even
if I could get the hardware again, it would take about two weeks to burn
the chips and another week or two to run the program through some
learning simulations. We don't have time for that.
Luckily, my deck isn't dead. And the only reason for that is because I
was running hitcher on CLU. I knew that MONICA would frag my deck, but
I'm surprised she was able to hurt me as bad riding a hitcher.
Once I shoveled up the last of the dead computer, it was about
mid-afternoon. Wolf-359 came over and the two of us sat down to give the
information we downloaded a really good looking over. With Wolf's usual
speed, he was able to organize the data into a well-worded package for
distribution. I'll let him talk about what he is going to do with it.
I have PILES of data to go over that CLU outputted during the run. I
definately do not want to attempt ANY kind of a rescue until that has been
evaluated. Probably another 24 hours or so. I believe that CLU will be
able to hold his own.
When writing CLU, I instituted a failsafe protocol that will not
allow anyone to forceably analyze his code. As long an MONICA doesn't
kill it, she won't get any information. She may be able to bypass that
failsafe, but it will take a lot of work on her part. She seems to want
it alive for something, so CLU should still exist in a couple of days.
I am disturbed by much of CLU's behaviour, though. Two things in
particular are completely outside of any operating parameters I gave it.
The first was the rewriting of the various ICe constructs. Standard
encounter procedure is to try to spoof the identification precedure,
using a combination of masking processors or utilities. During
simulation, that was also the procedure that CLU adopted. Yet, when
encountering AI's ICe, he chose to rewrite the identification procedures
to ignore us. This is no simple task, and I wouldn't try it on anything
by low-end constructs. It is just somewhat perplexing, although the fact
he didn't have to supress the constructs throughout the run certainly
were a boon.
The other was that relocate utility he whipped up. Standard relocate
utilites operate by sending a ping to the CPU asking for it to assign a
new node ID to the tracing ICe. Thus it becomes confused and loses the
trace. Instead, the relocate that CLU ran stipped the node ID from US,
and then pinged the CPU for a random one. This relocated us within the
system randomly. Theoretically that is possible, but internal checksums
are supposed to make the act of STRIPPING the ID impossible, especially
on a persona. I am greatly disturbed about where CLU not only got the
idea, but also determined how to do it, because I never gave it to him.
Highlander -- You mentioned a virus attack on CLU to kill him.
Unfortunately that won't work. CLU has several inbred immunities that
will make classic viral attacks impossible. It's part of his self
adaption code. There is an internal failsafe code that can be executed
that will cause a self-destruction, but it can only be delivered in close
proximity, and not when he is in attck mode (to prevent enemy
triggering).
This is all assuming, of course, that MONICA's version of CLU is
identical to the one that I was operating. Until we know differently, we
must assume the version existing entirely in the matrix is identical to
the one that was hardware based.]<<<<<
-- Ramirez <>
