[NTLK] OT: A Future That is a Singularity - Evolution
andrew.beals at gmail.com
Wed Mar 31 19:02:13 EDT 2010
On Wed, Mar 31, 2010 at 6:14 PM, Forrest Buffenmyer <anasazi4st at me.com>wrote:
> >That free-will may be emergent in the sense that no Turning Machine, in
> >present or future, would be able to model it.
> >Bye, bye singularity. Hello humanity...<snip> Now, speaking of Newtons...
> I have to admit, I was following this discussion and was intrigued....
It's a wonderful idea but those proposing it are *not* in the field.
They're nowhere near the field. Their base assumption is that if you make
a computer big enough, with as many neuron-equivalent transistors as the
human brain, it will suddenly be able to be smart enough to program itself.
This is the underlying premise behind more than a few Sci-Fi stories, with
the older ones being written about the The Phone System (back when Bell was
it) suddenly waking up one day, clacking relays indicating thought, phones
randomly ringing everywhere, when suddenly, the telephone in the Oval Office
rings and a mechanical voice starts making demands of the President...
AI is the greatest boondoggle of our era. Why, I've seen the best minds of
For the Robotics category, we have the Roomba - its big feat is that it can
find its charger when it's time to recharge. Roomba was not made by idiots
by any means, by the way. People think that radio-controlled vehicles with
a human pushing the buttons are "robots", and they watch them destroy each
other on TV. For commercial robots, they're mere task-repeaters, but at
least they're autonomous. The state of the art is represented by the DARPA
Challenge vehicles, which aren't anywhere near something you'd want to have
next to you on the freeway at 75MPH. (Yes, it's the twenty-first century
and I shouldn't have to drive my own car anymore)
Machine Translation - I'm sure all of you guys have played with
translate.google.com, babelfish.altavista.com (babelfish.yahoo.com now,
apparently!) and others.
Voice Recognition - I'm sure that a lot of you have played with Dragon's
products. I know I did a good decade ago. I have a friend who uses it
regularly and he went through a fairly long training period with it -
although shorter than what I had to go through. Voice recognition systems
have gotten better and better with time as those implementing them have
abandoned the "understanding speech" approach and have gone with a
statistical-based method. (ca. Y2K Dragon was great as long as I was
dictating a memo or speaking in the style of Mark Twain (its training text))
Chess. Anyone else here old enough to remember Microchess on the KIM-1?
The quote about it that I read in DDJOCCAO back in the day (mid 1970s) was
"If a computer playing chess is like a dancing bear, then Microchess is like
a dancing bear in a phone booth." At that point, the serious researchers
were trying to write chess programs that *understood* the game from a rules
standpoint, and then synthesized play from that. As time has worn on, they
started encoding chess opening books into the programs in order to keep up
with good players, and then came Deep Blue - a purpose-built supercomputer
that simply evaluated as many moves as it possibly could in the alloted time
and picked the one that had the most favorable tree behind it. Nothing
"Intelligent", just pure brute-force computing power.
And, finally, Artificial Intelligence. Eliza needs no introduction. She
was a quick hack that Weizenbaum (RIP) put together to demonstrate how easy
it was to fool people. Read the original paper here:
Eliza has a large number of descendants, and they all operate in the same
fundamental way. They recognize a portion of input as interesting, grab
keywords from the fragment, and then spit something predictable back at you.
State of the art would be the ALICE chatbot at a public level and most
"speak (type) instantly to an agent" customer service programs you see on
e-commerce websites at the private (unpublished/unpublicized) level.
Hard AI (= machine entities, reasoning, self-aware (= human-level computer
intelligence (= should have human rights, no? (e.g. HAL-9000)))) has been
predicted to be about "20 years in the future". Starting nearly fifty years
ago. People are still estimating its arrival in "about 20 years" (the more
wild-eyed Singularity folks WAG 10-20 years). Weizenbaum himself predicted
a few years back that it was more likely to arrive (with a lot of work, mind
you) in 400-500 years.
More information about the NewtonTalk