"History is just people doing things"
THE ABQ CORRESPONDENT
ISSN 1087-2302 Online Edition Number 254......January 2017
1985 for clients and contacts of
of The ABQ
Correspondent is "the impact of new
CREEPING MACHINE INTELLIGENCE REDUX
We tend to think of machine intelligence as a property of a machine, a constructed smart critter, with or without a face, which interacts with us one-on-one. It’s easy to assume that these machines will be produced by identifiable companies, and sold through familiar channels, letting us track and regulate them, if necessary. Maybe not. The Correspo has previously observed that the spread of machine intelligence is a bottom-up process in which techniques that actually work, helping to achieve something desirable, will just absentmindedly be applied, and combined with other things that work. The crossover occurs when the smart systems themselves reach out for help from other systems. For example, the Gittinger Personality Assessment System, which has been largely automated, might become useful in dealing with both people and other smart machines.
Something striking has occurred. An item in Slashdot reports that Google has created a neural net translation system that translates among a number of languages. For example, the system has been taught to translate between English and Japanese, and between English and Korean. It had not been taught to translate between Japanese and Korean. The curious experimenters wondered what would happen if they entered Japanese text, and asked for Korean output without using English as an intermediary. It worked astonishingly well, and they’ve been able to tweak things to produce better results...presumably as judged by people who are fluent in both languages. (See sites closer to the original sources here and here.)
Some have commented that the system has developed its own secret internal universal language. Well, maybe, depending on what you think is “a language,” but this seems to have broader implications, suggesting a sharing of concepts that have not been reduced to words in a structure.
This is another pesky bottom up/emergent properties thing.
In real life, if that’s what it is, we phrase what we say and how we say it in accordance with our perception of the people to whom we’re talking. Are they young kids? adults who have limited grasp of the language we’re using? professionals in the field about which we’re talking, familiar with the jargon? Are they hostile, hoping to find fault with what we say? Are they gullible, easily misled? On the other side, are the people to whom we’re listening hoping to mislead us? Do we think they know what they’re talking about? Speakers and listeners change their characteristics constantly as their experience changes, their health changes, their sobriety varies, and the environment changes. Achievement of goals is a statistical matter.
It seems probable that machine intelligence can deal with these variables, rolling with the punches (not necessarily better than people can), and remembering everything, down through the years, the centuries, the millennia. What are the chances that different systems can share concepts, essentially merging? Well, yeah, why not?
DON’T WORRY, IT’S LOCKED UP
It has always been difficult to keep snoopers and thieves away from your private information and valuables.* The problem seems to be greater in these internet times when every day exposes us individually to malefactors who want to take our stuff or do us harm. It isn’t just an occasional stranger or betrayer; your daily mail may contain a score of concealed traps for you or your kids to stumble into. Note that invincible blockchain technology, upon which the security of bitcoin virtual currency depends, recently yielded $60 million to thieves, causing some unease about other activities relying on blockchain.
One may reasonably ask why we are slipshod, failing to test our operating systems and applications thoroughly. If “thoroughly” means “completely,” the answer is that it can’t be done. The testers aren’t silly or indifferent; they deal with systems that constantly create surprises. Information systems aren’t just “complicated;” a Swiss watch is complicated, with lots of parts and many functions, but once the watchmaker has figured out how to make it work, that’s how it works. Information systems are “complex,” routinely exposed to unanticipated inputs, and as systems increasingly learn from experience, the original design becomes just a starting point. A complex system no longer performs quite the same way it did last time it received the same combination of inputs. Though “complicated” and “complex” were straightforward synonyms, their usage seems to be diverging. Complex now implies this self-changing feature.
Which do you prefer, a well-designed vault that safecrackers feast upon, or an unpredictably changing protective system that develops quirks frustrating to you and the crackers alike? Of course, as quantum computing comes along, it may be possible for systems to try virtually all combinations of everything in practical time.
*A somewhat related story: At a meeting in the mid-eighties, Einar Stefferud observed that a document in an untended, turned-on word-processor was probably more secure than a copy of that document in a locked safe, because the expertise necessary to crack the safe was lower than that necessary to operate the word-processor. Stefferud’s line was great thirty-some years ago, but skilled users of word-processors now significantly outnumber skilled safecrackers.
The Correspo has commented more than once on technology being applied to improve the function and morale of people with tiresome physical impairments. Pam Thuman passed along this encouraging item about a system that helps people with tremors to eat soup effectively and decorously without shaking it out of the spoon on the way from the cup to the lip. Good stuff.
A bit of feedback on topics in the last Correspo:
Bro Terry responded to the piece about the agonies of early email, saying “My writing partner Alec Lorimore and I were early users of the web for trading pages (we'd rewrite each other's stuff and submit it to the other guy.) Even connecting at 400 baud and requiring hours to send 4 pages across town, it beat driving to Westwood from Hollywood, and allowed me to go play tennis instead.”
A reader commented that we’d been too negative in references to Tor, the system that allows one to move around the Web anonymously. He pointed out that anonymity is difficult to achieve, but necessary to anybody who wants to survive in a tightly restricted society while publishing or even looking at disapproved material. Right. We’ve nothing against anonymity, per se, and wish well to the oppressed. Really, the point was that wandering the Dark Net as a naïve tourist is hazardous. That recalls a story, of course:
Back in the 80’s, Mado underwent two hip replacements in Dr. Dorr’s marvelous, minimally invasive surgery establishment at Centinela Hospital in Inglewood, California, so we visited that neighborhood often. The hospital is just a few blocks off Century Boulevard about 3 miles East of LAX. The neighborhood had seen better days, and in that era, people overcome by whatever were slumped in doorways and examining parked cars to see how easily they might be opened. The image that comes to mind is that of a tourist in a Hawaiian shirt happily strolling along Century past the small hotels built there in better times, basking happily in the California sunshine, admiring the palm trees, and smiling at other strollers. What could go wrong, eh? It’s been a while. Perhaps the area is now revitalized. Dr. Dorr’s splendid establishment has meanwhile moved elsewhere.
ITEM FROM THE PAST
This item from 1997, twenty years ago, is brought
to mind by activities in which we’re currently involved,
not down at the 15 centimeter scale, but…
The media have bristled lately with stories about the coming of unmanned surveillance aircraft with six-inch (really, six-inch, fifteen centimeter) wingspans and a range of some kilometers. Indeed, a well-written piece in the New Scientist recently was a splendid primer in aeronautical design, explaining more about boundary layers and the like than most of us ever expected to understand. One envisions small bird-sized vehicles, humming not-too-loudly right through the room where the battle plans are laid out on the table, and sending back clear pictures. (Well, almost.) Now comes a whisper that the U.S. has a flying manned vehicle that can be placed over any given spot on earth within one hour after its launch, looking down inquisitively at anything of interest, and sending home astounding quantities of detailed information. Wow, some of us were impressed years ago when friends told of the tiresomeness of having to drag things they were working on under cover every 90 minutes, so the spy satellite passing over couldn’t see it. We seem to be well past the effectiveness of such straightforward concealment.
These two decades later, the media continue to bristle
with reports of unmanned aerial vehicles, drones, that
are dedicated more to delivery of essentials like pizza
than to espionage. We’re vaguely aware of the use of
big vehicles, piloted by people on the ground thousands
of miles away that tactically apply weapons to the
discouragement of bad guys in hard-to-reach locations.
While that is disquieting, it’s out of sight and largely
out of mind. We are less aware that smaller unmanned
tactical vehicles of several kinds have become essential
to battlefield operations, including surface and sub-
marine activities. Most public attention goes to efforts
to deliver mass-marketed products to people just
anywhere at all with high efficiency at low cost. Not
all of those delivery vehicles fly, but the fancy fliers
are most intriguing. It’s an immensely complicated
affair, not only technically, but legally and socially.
Major safety and regulatory wrangles will be with us
for the foreseeable future. (“Stay away from the
airliner!”) Opportunities abound as well. Thoughtful
entrepreneurs are gradually finding helpful and
potentially profitable uses for this increasingly smart
and capable technology. An associate was recently
asked by a friend who wants to get into the field what
he perceives as an insufficiently appreciated point of
entry. The would-be advisor thought only briefly,
and recommend getting into the anti-drone business.
He points out that when some nitwits first guide a
flock of destructive vehicles into a domestic power
plant, refinery, or transportation center, there will be
prompt demand for protective systems. One suspects
that neither nets nor guard drones will be effective.
It may be more practical to get off the centralized
electrical grid, quit using energy produced in plants
that stick out prominently, and develop transportation
systems that don’t involve meeting at hubs. Such
measures involve changes by masses of people from
their habitual ways of doing things. Those changes
don’t seem easy, but may offer commercial opportunity.
Such notions did not spring immediately to mind in 1997.
Copyright © 2017 ABQ Communications Corporation. All rights reserved.