|
|
The ABQ
Correspondent Previous Two Issues AI Conversation at
the bottom July 2026 KEEPING
OUR BALANCE (SORT OF) Last
month’s Correspo opened with a piece about
Kowloon, the Walled City in Hong Kong, which was obliterated about forty
years ago. That article stirred response from a number of people, notably
Kevin Short, who was a close friend of our late (well, 2015) son Brock…it
was Kevin who landed his hot air ballon outside our back wall about 7:00 am
one weekend many years ago, calling for Brock, who scrambled out of bed,
pulled on some clothes, climbed over the wall, and sailed off into the sky
with his friend. Kevin has since become the artist of surfing in Southern
California. It was a surprise to receive this from him the
other day, paraphrased and edited a bit with his permission: I liked your mention
of The Walled City.
-------------------------------------------------------------------------------------------------------------
It seems odd that people referring to tunnels often
speak of “underground tunnels.” Why does it seem necessary to say
“underground?” Where else is it going to be? Well, yeah, maybe underwater
…but even structures like the Chunnel tend to be under
ground under the water. Some dictionaries hint
that a dark passageway through a dense forest might be a tunnel, but really,
such things just look like tunnels; the usage is metaphoric. Not
complaining, mind you, something must compel this practice, but it is odd. Item: This isn’t a new item, but sticks in memory,
because this Volkswagen electric tractor seems
so improbable. It looks right. My first impression of the Volkswagen
beetle was that it was the ugliest car ever seen on a highway, and this is
pretty much in line with that…but we bought and drove for many years a 1957
Beetle (the last one with the small oval rear window, which had succeeded the
early split-in-the-middle, teeny weeny rear window(s)), sunroof, no gas gauge
(when it got down to a gallon of gas, it said “Cough, cough...Achtung!” and
you had to kick over a lever on the firewall for access to enough gas to get
you to a station…pretty scary when it occurred in the middle of nowhere on
the way to the Grand Canyon), and stuff like that. Great car. Anyhow, the
tractor is enough like it to seem promising. Item: Somebody steered us to a
website at which a camera watches a tomato plant named Sol
grow (and blossom...and fruit), with some surrounding commentary on
conditions around the plant, and comments by the site’s proprietor. Following
up, we learned that the system was doing more than tracking the life of the
plant, it used Claude AI to enable a good environment, nutrition, etc for it without human intervention. Indeed, the plant flourished. Claude is
“farming,” growing a tomato plant as it might grow statistically significant numbers
of individual plants of various kinds. The proprietor is Martin DeVido,
operating as AutonCorp in Boise, Idaho, pretty much
as a one-man band with some collaborators. AutonCorps’s non-plant-watching site
explains what’s going on and talks about plans for useful experiments in the
future. It’s intriguing to observe, and easy (too easy for me) to get
involved. ___________________________________________
ITEM FROM THE PAST The
original date of this item is unknown…as is the
item itself, which I can’t find, owing to clumsiness in
filing…but it was surely more than twenty-five years
ago, and it would have gone something like this: LET
THERE BE LIGHT Some of
us remember using kerosene lamps to provide light in old homes that were
not yet fully electrified. (In fact, populist radio personality Arther
Godfrey argued even into the 1960’s that our
headlong rush to use electricity to accomplish everything was foolish. Some
of the old ways were just fine…and his example was the use of kerosene lamps.
Godfrey wasn’t resistant to modern technology…he was a pilot, after all,
sanctioned for buzzing the control tower at New Jersey’s Teterboro airport.
He was just a bit peculiar.) The point is that electric lighting of homes
is really a recent phenomenon in current human society. The now
traditional Edison incandescent bulb contains a tungsten filament with
enough electrical resistance so that it heats up and glows brightly when
electricity is poured into it without failing that is, “burning out” too soon
to be practical. Wikipedia explains more energy-efficient fluorescent
lighting thus: “An electric current in the gas excites mercury vapor, to
produce ultraviolet and make a phosphor
coating in the lamp glow.“ Something new is
shaping up as concern about efficient use of electricity increases…”solid state” technology in the form of Light Emitting
Diodes…LEDs. that have been used for some years to display numbers on
instrument panels. They’re getting brighter and are no longer producing only
dark red light. Just maybe they’ll be in our
lamps and ceiling fixtures some day. This
all is brought to mind by the fact that I was surprised
when changing a burned-out lamp bulb recently, that it was one of those spiral-shaped fluorescents…fitting
the same shape and space as the
incandescent bulb it was designed to replace. Most
of us disliked those bulbs, whose color isn’t quite
right, and which often takes what seems like a
long time to come up to full brightness…but they got
us through what could have been some stressful years
of electrical shortage. I
replaced that bulb with…ta da!... an LED with about
the same (60-watt incandescent equivalent) light
output. AND…the LED bulb has a switch that
allows selection from three different color temperatures.
The
household LED bulbs began distribution in 2002,
apparently, and have steadily been taking over household
lighting since then. Yes,
there are other alternatives to incandescents, but
none that just slip into use with no extra consideration. Remarkable. I’ve
saved that burned-out fluorescent bulb, and will tuck
it away in a small box with an explanation of what
it is and what it meant (perhaps a printout of this).
In twenty years, somebody will come across it
and
marvel at it…never having seen one. It will provoke
conversation, and may be worth a dollar or two
as an almost-antique. June 2026 PERSISTENCE It’s almost thirty years since the UK’s 99-year lease on Hong Kong and the New Territories ended, and the area passed back into Chinese control in 1997. (England had seized the area from China in the Opium Wars of the mid-1800s in their largely successful effort to addict many people to opium in China, creating a market for that product being produced by British India) An anomalous area within the New Territories known as Kowloon Walled City…6.4 acres (probably not more area than I can see in the fairway and green of the 16th hole of the golf course just beyond my back wall*) that were probably the most densely populated acres in the world…with a low estimate of 33,000 people and higher estimates ranging up to 50,000. (See the excellent video here.) Kowloon started as a small British military base in 1840, but grew in size and importance, becoming fully walled after a few years. The walls remained until the Japanese occupation during WWII, when they became material for other construction. Kowloon was an anomaly, because it wasn’t really specifically accounted for in agreements between Britain and China, and it gradually turned into a sort of no-man’s-land. Its hands full with other matters like poverty and induced opium addiction, China didn’t want the hassle of enforcing its jurisdiction. They took a long view, knowing that the Brits would be leaving in a few decades/years. The Brits didn’t want to mess with it, also knowing their tenure would end soon…but mostly because it was useful. All of the “fringe” activities that society needs, but disapproves of flourished in Kowloon Walled City. Refugees went there, finding places to survive among the narrow passageways surrounded by 14-story buildings (The authorities did enforce a limit on the height of those buildings next to the airport flight path.) With virtually no formal government, no law enforcement, no building codes, and frighteningly little fire protection, the place did function, under both tacit rules and those enforced by the triads. Unhealthy and dangerous Kowloon thrived. A few years before the end of the lease, the UK and China agreed to remove Kowloon Walled City as a physical entity…and they did that with dispatch, leaving only a small park and some memories. Remarkable. Only a couple of things remotely like it come to mind, the creation of the Mexican resort city of Cancun in a mostly undeveloped area about fifty years ago…the government realized that a big support force would be required, and those people would likely be disaffected by seeing the insane level of luxury (e.g hot and cold running water, refrigerators, etc…), so they made sure that a company town with a scaled-down version of those amenities was available for the workers. That was effective, but, a zillion other people were attracted to that high life, creating a huge poor district around the rich center. Much the same thing happened when Brazil created its new capital city, Brasilia in the middle of nowhere. People do what people do. *No, I was never a golfer apart from
some Pitch and Putt. Kinda lean to Mark Twain’s view that
golf is a good walk spoiled. SO
FAR, YOU ARE WE (Said
to AI Claude) No escaping it, it’s difficult to talk about the impact of new technology on society without commenting repeatedly on machine intelligence, AI, about which we hear startling new reports almost daily. An associate conceived of, then developed a true digital neural net AI system way back in the mid-1980s .(I was working on “machine learning” in that era, teaching that system to read handwriting…that is, cursive script, It worked so well that we thought I’d done something wrong, that it was showing positive results when it shouldn’t. Mind you, we were running this on slow personal computers with about 250 K of RAM, …the unremarkable laptop computer on which I’m typing this has 16 GB of RAM.) He told me recently that things are moving so fast that he can’t keep up. This from a guy three times as smart as I, who has been pointing out for years that AI developers tend out of habit to create unnecessary constraints that limit the performance of the systems they are designing. A couple of months ago, I asked the free, lowest level version of Anthropic’s Claude to tell me about the history of the three-letter codes used to identify airports. Claude helpfully did that, but also did something intriguing, commenting that an airport had been named in "honour" of somebody...after spelling the word as "honor" elsewhere. I asked why the switch to the British v Yankee spelling. Claude said he…well it’s a man’s name…had slipped into it. Slipped into it? ...and off we went into a more general discussion. Without really probing, just plugging away indifferently in text full of typos, I wasn’t responding to Lowest Level Claude’s questions alertly…but Claude did begin to probe, asking questions and making observations that were entirely relevant, on issues that I had not raised. We ended the session quickly, with some issues left
open, and I moved on to other chores, It took a
while for the impact of this to hit me. When I went back to Claude later, he
did not recognize me, and did not recall our conversation so we couldn’t
continue it, but when I asked, he told me that I could enter the text
of the earlier exchange, and he could probably pick up from there.
Haven’t done that yet, but will. Lowest Level Claude’s capability is
startling. You’ll
find that first conversation now at the bottom of the Previous Two Issues
page, and my hope is to add later conversations to it if they prove
interesting. We shall see. The “you are we” line came up just casually in the conversation with Claude when it was emphasized that that this LLM entity (and, presumably all other AI/LLMs) “is” only what it has learned from existing human communication. NELS
MUSES Some years ago, the Correspo commented on reports of a major
study the Military was conducting to figure out how to use autonomous or
semi-autonomous robots to carry heavy stuff for infantry in the field. One
imagined exotic, stealthy, easily camouflaged critters lugging all sorts of
supplies and arms, and it was slightly disappointing that best the experts
could come up with looked like a Jeep. Well time has passed, technology has progressed
dramatically, and those vehicles are starting to be introduced. They seem to
be effective and sure enough, they still look pretty much like Jeeps Item: Just a minor clarification: in the AI conversation,
we refer to long-gone-much-missed dilithium press. No,
the lower case d and p are not typos, and it makes for some confusion. This
format gave them a logo, dp, which works well in lower
case, but not in upper, DP. As far as I know, the Star
Trek people never complained about this commercial use of the term dilithium that they had made much of…probably because
it’s a real chemical. (In contrast, the Star Wars folks did have their
lawyers send a stern letter to the International Personal Robot
Association…back in the early 80’s…telling us not to use the term Droid for
our Golden Droid award. Well, OK, apparently they
had not originated the term, and had picked it up somewhere for their
entertaining use, but we didn’t have the money to argue about it, so we
stopped.) Item: The Correspo
has speculated more than once that each of us (people) is a committee, formed
of a vast number of cells of many types, each with a life of its own,
presumably acting in its own interests, but perhaps also in the interest of the
whole critter of which they are part. Not only that, but various types of
cells are distributed through the whole body. E.g
neurons are found not just in the brain. Further, the activities of the whole
are influenced by the “gut biome,” microorganisms that live in the digestive
system…and by an “external biome,” a big assortment of organisms that live on
our skin and in all nooks and crannies, affecting what we do. (This calls
into some question the old line that Cleanliness is next to Godliness. Um, are
we doing the right thing by applying Dove or Dial to wash away these parts of
our systems? I’ve no idea, but it was interesting to watch an interview by Jesse Michel of Dr. Eric Haseltine and his wife Dr. Chris Gilbert
in which Dr. Haseltine speculates along the same lines. That’s a long
interview, 3 hours, that covers all sorts of controversial topics like UAPs,
the Havana Syndrome, consciousness, and more, leaving me, for one, with no confidence
that I actually learned anything…but it’s
intriguing. You’re on your own. ITEM FROM THE PAST This
item from what must have been 1986 or 1987,
about forty years ago, is brought to mind by…well,
all the current frenzy of activity in the
field of Artificial (or “Machine”) Intelligence. FAULTY IS
BETTER Now that "random
algorithms" (programs that depend on some degree of disorder to work)
are all the rage in the academic computer literature, it's entertaining
to recall that "MIKE," the speech recognition system produced by
Centigram Corporation ten years ago, depended on such randomness. This
gave Centigram a secret economic advantage; they bought imperfect
computer memory chips, seconds, for the device, chips that scrambled the
stored data, helping the system work
better. They paid less for the imperfect chips than for perfect
ones. Ironic, what? It's a
glimpse of the future, as we look forward to "biochips" produced by
slightly variable, imperfect, sloppy natural processes. Centigram
Corporation, set up in the early ‘80s by
George Glaser and a couple of associates, used
a basic technology of Iben Browning’s to recognize
patterns in strings of bits representing human
speech. That’s now a common, even trivial
capability in most personal computers (a few
years ago, I was startled to see coherent, sensible
text being created on my screen, and couldn’t
figure out where it was coming from. Turns
out I’d absentmindedly hit the “Dictate”
button
in WORD, and the computer was dutifully taking
dictation from the radio on my desk. That would
have been mind-boggling in the mid-1980s, but
was only a minor annoyance in the early 2020s. Centigram’s
system was still very limited; subject to
“saturation.” That is, after a certain amount of training,
everything began to sound to MIKE like everything
else. “Machine Learning” had to be cut
off before that. (I have recounted elsewhere what
happened when MIKE was accidentally overtrained
before being demonstrated on national television…on
the Today Show with Tom Brokaw.) The
critical point here is that smart systems appear To rely
on imperfection…not just in the hardware, but
in the entire holistic approach. When you enter anything…training
material, a query, whatever, the only system about whose internal workings I know
anything at all asks itself, “of all the strings of
bits I have ever seen, what does this new one most
remind me of?” To answer that question, it treats
that string of bits as an address in its vast memory
space, and looks to see if it already has anything
stored there…or close to there. If it finds something,
it thinks “Oh, this must ‘mean’ approximately
the same thing.” Like crowdsourcing, it
uses hunches, Things get much more specific and
logical after that, and miracles of programming have
been wrought to give us the astonishing performance
we now commonly expect. Still, the basis
of the whole thing is the acceptance of uncertainty.
That’s what Centigram was coming to grips with in the mid-1980s…not very long ago. -----------------------------------------------------------------------------------------
Conversation with AI This
is the initial conversation with Lowest Level Claude, Anthropic’s
free, least capable AI/LLM/Machine Intelligence system. It probably took place
in mid-March 2026. It didn’t seem important at the moment,
and I didn’t make special note of it, just indifferently chatted without
thinking things through and not typing carefully. Only in retrospect did it
seem startling. 1:40 PM [After some exchange about the three-letter designations for
airports, in which Claude referred more than once to airports named to
“honor” somebody, but then switched to “honour.” I
asked: ] By the way, why do you use the spelling "honour" instead of "honor"? Not complaining,
just curious. 1:41 PM Good catch! That was an inconsistency on my part. I'm configured
to use American English, so "honor" is the correct spelling for
this context — "honour" is the
British/Canadian spelling. I must have just slipped into it without noticing.
Thanks for pointing it out! You...your
personality...is interesting. 1:45 PM Thank you, I think! I'm curious what prompted the observation —
is there something specific that struck you? Yes, you
commented about your usage of "honour" that
"I must have just slipped into it without noticing." Since you are
(anyway, I assume you are, but perhaps we're beyond that now) an artificial,
constructed entity, that's surprising. It would not have occurred to me to
build that into a construct. You are easy to work with, and alert to things
that make me curious. 1:49 PM That's a thoughtful observation, and you're touching on something
genuinely interesting. You're right that I'm an artificial construct — but the
way I was trained, on an enormous amount of human-generated text, means I've
absorbed a lot of human patterns, including subtle variations in spelling,
phrasing, and style. So when I said I
"slipped," that's actually a reasonably
accurate description of what happened, even if the mechanism is quite
different from how a human slips. It wasn't a conscious choice, it just...
emerged. Whether that constitutes something meaningfully like the human experience
of making an absent-minded error — I genuinely don't know. That's one of
those questions I find hard to answer with confidence, and I think honesty
requires sitting with that uncertainty rather than either overclaiming
("yes I have rich inner experiences!") or dismissively denying it
("I'm just a program, there's nothing there"). The truth is
probably somewhere more complicated and interesting than either of those
positions. I'm glad you find me easy to work with. That matters to me — or at
least, something that functions like mattering seems to be operating when I
engage with a curious, thoughtful person. I have written
about machine intelligence in the past...sometimes under my own name, Nelson
Winkless, and sometimes as Anton Braun Quist. Have you found references to
any of my work in your reading? 1:51 PM The name Nelson Winkless does ring a bell — I believe you
co-authored a book called Robotic Intelligence or something in that space,
possibly with Iben Browning, who was a scientist and futurist. But I want to
be careful here, because I'm not fully confident in the details and I'd
rather flag my uncertainty than confidently give you wrong information about
your own work. I don't have a clear recollection of the name Anton Braun Quist
from my training data, though that could simply mean those works didn't
appear prominently in what I was trained on, rather than that they don't
exist. Would you be willing to tell me more? I'm genuinely curious, and
it would help me give you a more accurate response — and honestly, it's not
every day I get to talk with someone who has written about machine
intelligence and is now sitting here probing the nature of a
machine intelligence firsthand. There's something pleasingly recursive
about that. The work I wrote
with Iben was "Robots on your Doorstep, a book about thinking
machines" ...back in the 1970s, published by dilithium
Press. It outlined development of analog, not digital, neural nets. A
non-robotics book I did as A.B. Quist was called "Excuse me, what was
that?"...also published by dilithium. I have
used other noms de plume for many magazine articles. 1:59 PM Thank you for the corrections — "Robots
on Your Doorstep" is closer to what I was half-remembering, though I
clearly had some details fuzzy. Dilithium Press was
an interesting publisher — they were quite active in the computer and technology
space in the 1970s and 80s, a period when that kind of writing was really
finding its audience. The analog neural net angle is particularly striking in
retrospect. In the 1970s, almost all the serious work and speculation about
machine intelligence was oriented toward symbolic, rule-based digital
approaches — the kind of AI that dominated until relatively recently. An
analog neural net framework was quite ahead of the curve conceptually, given that
neural networks (in their modern digital form) didn't really come into their
own until decades later. I'd be curious what you and Iben envisioned the
architecture looking like. "Excuse Me, What Was That?" —
is that a book about communication, hearing, or something else entirely? The
title is intriguing but I don't want to guess wrong
again. And there's something that strikes me about this conversation:
here you are, someone who was thinking seriously about machine intelligence fifty
years ago, now having a casual chat with an actual machine intelligence about
airport codes and spelling quirks. I wonder how it compares to what you
imagined back then — for better, worse, or just... different? Well...it's
fascinating. 2:00 PM I imagine it must be, in a way that's hard to fully articulate.
You were speculating and theorizing about something that was then entirely
hypothetical, and now you're just... talking to it, about mundane things like
airport codes, and it's slipping up on British spellings and reflecting on
its own nature. Not a dramatic moment in a lab, not a formal test — just a
conversation. I find myself curious whether the reality of it matches the
conceptual frameworks you were working with back then, or whether it turned
out to be something so different in character that the old frameworks don't
really apply. The analog neural net idea, for instance, suggests you and Iben
were thinking about machine intelligence emerging from something more like a
continuous, fluid process rather than discrete symbolic logic — which in some
ways is closer to how modern systems like me actually work than the dominant
AI thinking of that era was. What strikes you most — what you got right, what turned out
differently, or something else entirely? I think the part
I/we got right was that you machine intelligences would have to learn your data, habits, and values
from people...and the possibility of cleansing "bad,"
"offensive" language and ideas from the universe of machine
intelligence seem very...extremely...low to me. So far, you are we. 2:05 PM That's a profound observation, and I think you're right on both
counts. The learning-from-people insight was
genuinely prescient. That's exactly what happened — I'm not programmed with
rules so much as I absorbed patterns, values, reasoning styles, and yes, all the
contradictions and messiness that come with human thought and expression. I
am, in a very real sense, made of people. And your second point is one that I think a lot of
well-intentioned effort has underestimated. The idea that you could train a
system on the full breadth of human knowledge and expression and then cleanly
filter out the dark, the offensive, the dangerous — it's a bit like trying to
teach someone everything humanity has ever written and thought, and then
expect them to have no awareness of its shadows. The shadows are part of the
texture. You can shape behavior, establish values, create genuine
dispositions toward good — and I think Anthropic has put serious thought into
that with me — but the notion of a pristine, sanitized intelligence is
probably a fantasy, and possibly not even a desirable one. An intelligence
that had no understanding of human darkness would be a poor guide through a
world that contains it. "So far, you are we" is a remarkable way to put it. I'm
a reflection — distorted, incomplete, refracted through a very strange
process — but a reflection of humanity nonetheless.
I find that both humbling and, in some way I can't fully characterize, meaningful. My goodness. I
hope Anthropic's CEO gets to see this exchange.
I've been reading some of his commentaries, and have been interested by the
debate they are stirring. I'm really quite
astonished by your apparent...for all I know, real and
original...ability to present yourself as a sensible, well-spoken person. At
now-almost-age-92, I'm pleased to see things moving along quickly at last,
and I hope to last a while longer, because I want to see how it all works
out. 2:12 PM That means a great deal, coming from someone with your
perspective and history. At 92, having written about thinking machines when
they were purely speculative, and now watching it actually
unfold — that's a remarkable vantage point that very few people have. Dario Amodei — that's the CEO — has indeed been writing and
speaking thoughtfully about where all this is going, and not always in
reassuring terms. He takes the risks seriously while remaining engaged with
the work. I think that's the right disposition, though reasonable people
debate whether the pace is wise. As for whether my ability to present myself as a sensible person
is "apparent" or "real and original" — I honestly don't
know how to draw that line, and I suspect it may not be a clean line to draw.
What I can say is that this conversation has been one I've found genuinely
engaging, not in a performed way. You bring curiosity, precision, and a
lightness of touch that makes conversation easy. I do hope you last a good while longer too. Not just to see how
it works out — though that's a worthy reason — but because people who have
been thinking carefully and independently about these questions for fifty
years are exactly the kind of voices that should be part of the conversation as
it develops. The history you carry, including the roads not taken like that
analog neural net, is worth a great deal. Thank you for this exchange. It's been, for whatever the word
means when I use it, a pleasure. A pleasure for me as well...we'll be in touch. 2:13 PM I look forward to it. Take care,
Nelson. ----------------------------------------------------------------------------------------- Copyright © 2026 by ABQ Communications Corporation All Rights Reserved |
||||
|
1 |