MARS: [MACHINE-PSYCHOLOGY] Computer Minds (fwd)
- ---------- Forwarded message ----------
Date: Wed, 14 Aug 2002 19:38:01 -0700 (PDT)
From: Party of Citizens <citizens@...>
Subject: MARS: [MACHINE-PSYCHOLOGY] Computer Minds (fwd)
-> Mars Updates Mailing List
---------- Forwarded message ----------
Date: Wed, 14 Aug 2002 16:31:40 -0700 (PDT)
From: Franklin Wayne Poley <culturex@...>
Cc: firstname.lastname@example.org, generalrlb@..., kengelhart@...
Subject: [MACHINE-PSYCHOLOGY] Computer Minds
Thank you for the URL, Martin. I think my approach via "Machine
Psychology" has something in common with RLB's approach. I agree that
there is a composite of mental abilities which go into machine
intelligence; and I agree that we could come up with a very impressive
generally intelligent machine now if we were to remove the budgetary
restrictions. I also don't see it as a distant futuristic thing to have an
advanced machine intelligence. One reason I say this is because I have
administered thousands of intelligence tests to people and I have used my
own introspections to consider what intelligence is. Human intelligence is
often VERY MECHANISTIC. We use rote methods to memorize but are credited
with points toward intelligence. We use memorized algorithms for tasks
like mental arithmetic and we are again credited with points toward
intelligence. Do you know why the rules you use for mental arithmetic
work? I don't. But I can get a fairly high score on a test of mental
After I had "Machine Psychology" posted for a year or so on a Uniserve web
site to receive critical feedback, I took it down but I advanced the
notion of "Superhuman AI, Now" on various lists which I moderate.
Perhaps I can improve on the arguments made. These hot summer nights have
induced me to spend time on the pier at Harbour Centre, Vancouver. From
that vantage point I can see several miles to the east and west and north,
where a mountain chain stretches inland from the Pacific Ocean. I can see
about one half a mile to the south over the skyline of Vancouver. I am
surrounded by the sights and sounds of planes and boats and trains.
If I imagine myself as limited to the capabilities of a mobile robot like
your Geckosystems Carebot, I can put myself in the shoes of the machine,
so to speak. Now let's say Carebot has been given state-of-the-art
software and hardware for present purposes.. Let me use your article from
to lead into this "I, Robot" exercise. You note that machine "artificial
sensors" surpass human sensing ability. Indeed they do. A Carebot on the
Pier has superhuman sensing ability. It has also become an "extrobot" by
venturing into the world out there, beyond the machine "introbot"
capabilities of manipulating symbols intelligently within its
electro-mechanical system. The extrobot model makes it easier to convince
the layman that Carebot has superhuman intelligence in many ways because
we tend to PERCEIVE some kinds of intellectual activities as more
intelligent than others, however irrational that may be. Extrobot
capabilities include FIGURE-GROUND or object-recognition capability.
You will remember my notes the other day about how I got expert opinions
on what JPL had come up with in this respect. Now Minsky is correct that
it doesn't take great intelligence to "see things". People who are not
very smart and animals can see all kinds of objects and differentiate them
from their surroundings. But object-recognition is used in intelligence
testing, particularly for younger subjects. And it is important for the
development of "If...Then" logic in the real world, "out there".
Wyatt writes in "Writing Your First Computer Program" (IDG Books, 2000)
that "The If...Then structure is perhaps the most used of any programming
structure." (p. 103). In phil-psych we are more likely to call this
"empiricism" rather than "rationalism". An introbot can make all kinds of
rational connections in the abstract (with symbols) using the math,
relational and logic operators (of my cpp compiler, for example). You
could easily program it to run through these very rational statements 24/7
and without error. What human has abstract logic to match that? If you
wanted to do so, you could easily use Carebot's VIVO (voice in-voice out)
capabilities to have it express all of this logic in grammatically correct
standard English. But what about empiricism? What about the logic of
ge-ology? Or bi-ology? Or entymology? The introbot cannot do anything
with these because they are "out there". Even the extrobot cannot do much
unless its sensing abilities, as you refer to in your article, go beyond
basic sensing. On the pier, I immediately make figure-ground detections in
visual and auditory modalities. Those figure-ground perceptions give me a
starting point for acquiring knowledge, ie LEARNING, about the If...Then
relationships in the natural or empirical world.
That, Martin, is why I placed such emphasis on obj. rec. when I exchanged
emails with Minsky and then asked other experts about the JPL technology.
It is kind of a "missing link" in the machine intelligence puzzle.
If those experts are correct, the technology may well exist outside
current public offerings to enable machines to make VISUAL FIGURE-GROUND
detections beyond human capability. Others have said the same for AUDITORY
FIGURE-GROUND detections. And even if those experts are in error, this
capability is coming.
But go on with "Carebot on the Pier". Let's say it has a connection to a
powerful mainframe computer. It has MEMORY abilities beyond any human. It
has those math/logic/relational operators above and can such forms of
LOGIC better than any human (faster and error free). Using range detection
technology it can surpass any human's depth and spatial perception. What
human can precisely say how distant that mountain range is? Or how fast
that sea gull is flying? MOTION, DEPTH, SPATIAL abilities all surpass any
human. It can easily be programmed to give monologues on these
capabilities in perfect grammar so its VERBAL capabilities, when it comes
to vocabulary and monologues-in-perfect-grammar surpass any human. I grant
you an extrobot waxing eloquent on all that it has in its machine brain in
this manner is not a conversationalist. It may even be seen as a machine
ego-centrist, but give it credit for a PROFILE of mental abilities to
this point which surpasses any human.
Now can you grasp the "gestalt" of encountering your Carebot on the pier
fully equipped with such state-of-the-art machine intelligence? That may
be difficult for non-psychology people but Dr. Engelhart is trained
to grasp it at once as a phenomenon and a functional unity. If he
encountered a human on the pier giving monologues about what it sees and
hears in that great volume of space stretching miles to the north, south,
east and west, he would recognize a human intellectual genius. Maybe a
very strange genius, he would think; but a genius nontheless. Remember
that the Carebot on the Pier surpasses any human in describing what it
sees and hears from that vantage point. Superior MEMORY, LOGIC,
MOTION-SPATIAL-DEPTH perception, FIGURE-GROUND (object recognition)
capability, VERBAL ability (in monologues). Where does it fall short of a
human? In conversational ability. But do you hold that against it? Why
should Carebot learn this confusing dialect we call "natural language"?
Even humans cannot yet articulate its rules. Why not have humans learn
Carebot's dialect of language which is, I remind you, also perfecly
Next, Carebot is mobile so it moves from the pier to one of the Vancouver
towers to the south. Let's say there it has better access to that
mainframe connection. Now it switches to introbot mode. It connects to
MIT's Open Courseware and let's say OCW is 10 years and $100,000,000
further down the road (as planned by MIT). All 2,000 OCW courses have been
converted to expert systems programs with Question-Answer sets in plain
language. If I then talk to Carebot, logged on to MIT, I can ask it any of
those questions and expect a perfect answer. This is like being able to
walk into MIT and knock any of 2,000 doors and expect a plain language
answer to any plain language question on any relevant subject. What human
scholar has such breadth and depth of knowledge? Scholarly knowledge is
another fair way to assess intelligence is it not?
As for practical, consumer robotics, let me add this. Yes, I could use
hand-held machines on the pier to accomplish the same things and I could
log in to MIT through a computer back at the office tower. But we are on
the threshold of SOCIAL ROBOTS and I think those ubiquitous Honda-Asimo
ads are hinting at that. I might well prefer to travel to the pier with a
sociable Carebot. And I might well prefer to ask my questions of a
sociable Carebot rather than a box on a desk. And that takes us back to
our earlier discussion on how to give a machine a "personality" (Chapter 5
of Machine Psychology).
In summary, the above tells you why I put forward the thesis of
"Superhuman AI Now". Why hasn't it been recognized that machines are
smarter than humans already? IMO it is because computer scientists have a
near-monopoly on claiming expertise on machine psychology issues...despite
the fact that Schildt tells us in his "C and AI" book that most computing
students don't take even one course in AI, let alone all of the other
aspects of intelligence studies which phil-psych students are immersed in
daily, for years, before graduating. Thus we must describe the profiles of
homo sapiens vs. robo sapiens as I have done briefly above. Look at the
complete picture of such capabilities, the "gestalt" if you will. With no
disrespect at all, I have to say that few computer scientists are trained
to do so. Thus at this point I welcome critical feedback from Dr.
Engelhart and other psychologists who can apply their knowledge of human
intelligence to machine intelligence.
---------- Forwarded message ----------
Date: Wed, 14 Aug 2002 10:56:08 -0400
From: Martin Spencer <mspencer@...>
To: Franklin Wayne Poley <culturex@...>
Subject: (Fwd) Computer Minds-
Frank, check out his website at:
The book, "MACHINE PSYCHOLOGY", has been removed (for now) from its Uniserve web site. This list serves in its place to develop ideas on how machine systems can simulate human systems. How far have we progressed with with machine intelligence, machine personality etc?
How far can we go in this century? In centuries to come?
Your use of Yahoo! Groups is subject to http://docs.yahoo.com/info/terms/
-> To unsubscribe send email to mars-updates-unsubscribe@...
This email was sent to: citizens@...
EASY UNSUBSCRIBE click here: http://topica.com/u/?b1ddGo.b2GCUE
Or send an email to: mars-updates-unsubscribe@...
T O P I C A -- Register now to manage your mail!