SFworld 版 (精华区)
作 家: xian (去日留痕) on board 'SFworld'
题 目: The Dominus Demonstration (4)
来 源: 哈尔滨紫丁香站
日 期: Fri Sep 26 08:37:37 1997
出 处: byh.bbs@bbs.net.tsinghua.edu.cn
发信人: KingKongKang (KKK经理/裁判), 信区: SFworld
标 题: The Dominus Demonstration (4)
发信站: BBS 水木清华站 (Sun Aug 3 18:50:57 1997)
He looked around at the audience.
"I can never remember anybody's damned phone num-
ber," said a voice from the middle of the room.
"Me neither," said Armstrong cheerfully. "Go on, Jim."
"People have been trying to build a computer memory
that operates more like a human memory for forty years,
ever since the Perceptron. we weren't very successful. We
finally presented all the data we had on the problem to
Magsman Two, and then later to Magsman Three, and let
it grind away on it. Six years ago we had the elements for
a new design. Terrific, eh?" He smiled. "Just one prob-
lem: nobody could understand how it was supposed to
work. The machine had been through an enumeration
process that we couldn't hope to duplicate. We built it
anyway. It seemed to do its job, but it did a few other
things too -- I don't know if I should get into this -- so
we've only included it as an available unit in our systems
for short spells."
There was a stirring of stronger interest among the
reporters. Bevin looked questioningly at Armstrong. The
other man nodded his encouragement.
"Go ahead, Jim. I promised we'd not hold anything
back. The performance of the high-associative memory
may be a mystery, but it's not a secret."
"Very well. Let me start, then hand you over to Dr.
Chang. Most of this involves software questions, and where
the software starts, I stop. But I'll tell you one of the
hardware mysteries. The circuits for the final memory
included aleatoric components. That's just a fancy name
for pieces that introduce random elements into the sys-
tem. In the case of the high-associative memory, those are
produced from quantum level fluctuations -- completely
unpredictable. It means that the outcome of the calcula-
tions, and the logical circuits employed to make them.
can never be known in advance." He grinned at the ex-
pressions of the faces in the audience. "I know. We didn't
like it too well, either, when we realized what was going
on. Now I'll let Dr. Chang make it worse."
He paused, as a hand was raised in the audience. "Yes,
ma'am?"
"Can you show us this 'high-associative memory'? Peo-
ple will want to know what is looks like."
"Rafael?"
Chang nodded, reached beneath the table, and picked
up a glittering object the size and shape of a shoebox. He
placed it gently on the smooth wooden surface. "This is a
prototype. The one that will be used in Dominus is a little
smaller, but not much." He raised his eyebrows. "I don't
know how interested you are in cost figures, but this one
cost us sixty million dollars to develop, and almost noth-
ing to fabricate. The assembly was done completely under
computer control. I should also mention that the pro-
grams I'm going to describe were also developed largely
by Magsman Three and its predecessors.
"Let me get to questions of software performance. We
found two things happened when we began to use the
high-associative memory on real problem. I'll call the
first one the 'realism' problem. In human terms, you might
say that the system doesn't have too good a grasp on
reality. Sometimes we receive logically possible -- but prac-
tically outlandish -- solutions to the problems we enter."
"Let's have an example, Rafael." said Armstrong. "I
don't operate too well with abstractions, and maybe some
others here are the same."
"Here's a simple one. We defined a problem in micro-
circuit' design, trying to lay out a mini-microchip an
optimal way. Very standard, very important practically."
He drew in a deep breath. "All right. First few cases, the
output was conventional solutions. So then we threw that a
much harder case, where we weren't at all sure that a
feasible design solutions existed. After a few minutes'
computation -- which is large amount, for these machines
-- we had an answer. But it wasn't a useful one. The
connections were to be made in a space of more than
three dimensions. theoretically interesting, but off the
wall. Again, if you want to describe it in human terms,
you might say that the high-associative memory made the
machine smarter, but maybe it did so at the expense of
common sense."
He paused. Another hand had been raised in the audience.
"Dr. Chang, you've used words like 'smart' and 'remem-
ber.' Those are terms I'd reserve for people. I want to
know, will Dominus be able to think? I mean, the way
that we think. Wasn't there some test done to see if that
was true?"
"Let me handle this," said Armstrong quietly. He looked
up, blinking in the bright lights. "I'm sure that what you
are referring to is the Turing test. It's a classical test,
proposed by Alan Turing in the 1940s, to see if a machine
is 'intelligent.' But it has problems. The idea is that you
communicate with something -- a machine or a human --
over a remote connection, so you can't see, hear, smell,
touch, or taste what's at the other end. Then you are
allowed to ask any questions you want. If you can't tell
from the replies whether you're in touch with a man or a
machine, then the machine is said to be intelligent. Sounds
good, right? The initial partial tests of Dominus included
a test of the Turing type. And Dominus failed -- the first
time. The machine had inhuman powers of memory, su-
perhuman computation speed, and never made an error.
The testers knew damned well it wasn't a human. On that
test, Dominus was too intelligent to be human. On the
second test, the machine passed. A heuristic analysis of
the desired objective, based on the record of the first test,
allowed it to simulate lapses of memory, slowness of
thought, and all the thousand failings that flesh is heir
to."
He smiled at the perplexed expressions on the faces of
the audience. "Can Dominus think, ladies and gentlemen?
I don't know. I suspect that it can do at least as well as
any of us. I don't think that was the answer you perhaps
wanted, but it's the best one I can offer. Rafael, would you
like to finish what you were saying about your worries
over the high-associative memory? Then we ought to call
it a day. Our audience has deadlines to meet."
Chang nodded. "It won't take a minute. The second anom-
aly we've noted I call the 'sense of humor' problem. It's
not easy to put in concrete terms, but when there are
several solutions permitted to a problem, the one rated
highest by the machine is often the least likely. There
appears to be a preference for the most surprising answer.
Surprise is central to most humor. Having said that, I
should point out that I'm stating this in very anthropo-
centric terms. I shouldn't do so. These are purely techni-
cal issues. The group that are opposed to activating
Dominus do not seem to distinguish human emotion from
computer operations."
"How are you responding to the civil suits that those
groups have prepared?" The question was interjected by a
reporter in the back row.
"Mine again, I think," said Armstrong. He stood up. "I
don't have a good answer for you. I wish I did, and I
wonder what I'm doing here at all, in all this snow and
lousy weather. It's snowing again outside now. That's the
bad news. The good news is that we won't have to worry
about the civil suits to reach out decision on activation of
Dominus. As you know, there are two suits involved. The
Church of Christ Ascending wish to force the program to
proceed. The Citizens for Appropriate Technology seek an
injunction to prevent it. Shortly before we began this
meeting, the judge called the lawyers for both groups. He
will neither halt the final assembly of Dominus nor will
he require that the final hook-up take place. He will leave
the responsibility here."
Armstrong shielded his eyes against the lights and
scanned the roomful of attentive reporters. "I will offer
you one guarantee. Some of you have been here for a long
time. I think I see shirts that have been worn for more
than one or two days. None of us wants to spend the
winter here. I promise you a decision within twenty-four
hours. One more day, ladies and gentlemen."
"Have you already made your decision, but don't want
to announce it?"
"We have not. We need more time to think, and want
your permission we will go and do it."
"One last question, General. Did you know that the
members of the Church of Christ Ascending are still roll-
ing in here? Our estimate is that there will be more than
twenty thousand of them tomorrow. They plan a mass
demonstration."
"I know." Armstrong shrugged. "The judge refuses to
block that demonstration. We will go along with that. But
let me assure you the decision will not be forced by pre-
sure from any groups."
* * *
--
KK KK KK KK KK KK
KK KK KK KK KK KK
KKKK KKKK KKKK
KK KK KK KK KK KK
KK KK KK KK KK KK
※ 来源:·BBS 水木清华站 bbs.net.tsinghua.edu.cn·[FROM: 166.111.5.3]
--
※ 来源:·哈尔滨紫丁香站 bbs1.hit.edu.cn·[FROM: byh.bbs@bbs.net.tsin]
Powered by KBS BBS 2.0 (http://dev.kcn.cn)
页面执行时间:2.743毫秒