16 Mar
1998
16 Mar
'98
4:59 a.m.
At 04:34 AM 3/16/98 -0600, you wrote:
Exactly. All of the useful ideas from AI that became
mainstream are no
longer considered AI. Today, OCR, speach recognition, machine
translation, and predictive analysis are off-the-shell apps or embedded in
products like Microsoft Word to help catch your spelling and grammar
errors.
What was once the stuff of science fiction epics is now mundane?
There are still interesting problems, though. Machines can kick your
chess-playing butt, but you won't find one nearly coordinated enough to
hit a baseball and run around a few bases. IMHO, AI researchers have
overestimated the brainstuff and underestimated the sensor and actuator
stuff. Here's my theory of how you learn to speak, for example:
I have always thought that digital computers would never allow us to
achieve the ultimate goal of replicting a learning organism. Aren't we just
simple conceptual pattern recognition machines? It seems like an analog
computer, capable of integration of raw percepts and conceptualization at
high speeds, could actually learn and become better and faster than man at
thinking and working. If a computer could search a text file for a pattern
using the same method as humans, i.e. looking for a shape as the first
indicator of a match, rather than a discreet chacter pattern, it would be
able to process text much faster than a digital machine.
I think it was Ayn Rand's "Objectivist Epistemology" that got me thinking
along these lines.
-- Doug
--
David Wollmann |
dwollmann(a)ibmhelp.com | Support for legacy IBM products.
DST ibmhelp.com Technical Support | Data, document and file conversion for
IBM http://www.ibmhelp.com/ | legacy file and media formats.
16 Mar
16 Mar
6:14 a.m.
On Mon, 16 Mar 1998, David Wollmann wrote:
I have always thought that digital computers would
never allow us to
achieve the ultimate goal of replicting a learning organism. Aren't we just
simple conceptual pattern recognition machines? It seems like an analog
computer, capable of integration of raw percepts and conceptualization at
high speeds, could actually learn and become better and faster than man at
thinking and working. If a computer could search a text file for a pattern
using the same method as humans, i.e. looking for a shape as the first
indicator of a match, rather than a discreet chacter pattern, it would be
able to process text much faster than a digital machine.
If you believe Turing, there's nothing an analog computer can compute that
a digital one can't. A brain is many things: it's wet, it's analog, and
it's massively parallel. I don't think anybody believes that it's wetness
or analogness that matters, but clearly a high degree of parallel
processing seems important to solving perception problems quickly. This
is the basic inspiration that drove Danny Hillis to create the Connection
Machine, with 64,000 simple processors working in parallel.
But if you ignore the issue of speed, a single CPU should be able to
simulate the parallel processing done by eye balls, the visual cortex, and
whatever associative memory comes into play.
The early AI folk thought they could do brainstuff using straight-forward
algorithms. Maybe you can, but there's no biological analog that anybody
can find to support the idea that humans work that way.
So, the connectionists tried to create machines and structures modeled
after the brain, but they didn't get too far. Let's say that you build
an OS and a programming language that allows you to accurately model a
brain. If you stick a human brain in front of a newspaper, you get
nothing. So add some input devices and actuators. Now you stick a baby
in front of a newspaper -- still nothing. So let the baby run for a
while, experience a variety of sensations, make a whole bunch of
associations, stick it in front of many newpapers and many non-newspapers
for many years, and finally you get a pretty good character recognizer.
We're still a long way from being able to put a multi-billion node
connection machine onto a sturdy frame with millions of intricate wires,
actuators, and sensors, let it experience and manipulate the environment
for many years, and then get the thing to demonstrate emergent properties
that make the whole greater than the sum of the parts. But I don't see
any reason why it won't or can't happen.
And if you think that the prospect of human cloning is causing moral
pangs, wait until somebody creates the first artificial life form!
Ob classiccmp: Perceptron.
-- Doug
10:56 a.m.
In article <Pine.LNX.3.95.980316052219.18567G-
100000(a)behemoth.host4u.net>gt;, Doug Yowza <yowza(a)yowza.com> writes
If you believe Turing, there's nothing an analog
computer can compute that
a digital one can't. A brain is many things: it's wet, it's analog, and
it's massively parallel. I don't think anybody believes that it's wetness
or analogness that matters, but clearly a high degree of parallel
processing seems important to solving perception problems quickly. This
is the basic inspiration that drove Danny Hillis to create the Connection
Machine, with 64,000 simple processors working in parallel.
Perhaps incredibly,
Turing _did_ believe that there was something
special about the brain (in particular he could/would not rule out ESP)
and so I don't think he would ever have claimed that a Turing Machine
could do anything that a human brain could. The TM was designed to
solve a specific problem in mathematical theory, rather than as a
theoretical ultimate brain.
But now you've got me trying to think of something that an analog(ue)
computer can do that a digital one can't. Reversibility might be one
thing. I guess it's reasonable to argue that digital computers are a
subset of analogue computers, as transistors are analogue.
I'm going to stop thinking about this before I recurse.
--
Lawrence Wilkinson ljw(a)formula1.demon.co.uk
The GirlFrendo homepage: http://www.formula1.demon.co.uk/girlfrendo/
"You've got the brains, or so you say, maybe you see things another
way"-bis
1:41 p.m.
On Mon, 16 Mar 1998, Lawrence Wilkinson wrote:
But now you've got me trying to think of something
that an analog(ue)
computer can do that a digital one can't. Reversibility might be one
thing. I guess it's reasonable to argue that digital computers are a
subset of analogue computers, as transistors are analogue.
To be fair, analog computers can do things digital computers can't. For
example, a digital computer can only approximate 1.0/3.0 whereas an
analog box has no trouble with this. Certain ops would also be much
faster with analog vs. digital, but I'd have to guess that these are
implementation issues that get lost in the noise.
It's understandable that Turing (or anybody else) would have to say "and
then a miracle happens" in any discussion of machine intelligence.
Emulation of human intelligence is so very far off -- we're simply not
smart enough to create it from scratch. If we ever get to the point where
we can emulate, say, bee or ant intelligence/behavior pretty well in my
lifetime, I'd be absolutely astounded -- there's still *that* much to do.
-- Doug
3:21 p.m.
In article <Pine.LNX.3.95.980316132720.19589A-
100000(a)behemoth.host4u.net>gt;, Doug Yowza <yowza(a)yowza.com> writes
To be fair, analog computers can do things digital
computers can't. For
example, a digital computer can only approximate 1.0/3.0 whereas an
analog box has no trouble with this. Certain ops would also be much
faster with analog vs. digital, but I'd have to guess that these are
implementation issues that get lost in the noise.
But a digital computer can
represent rational numbers exactly as you
have (e.g. Smalltalk has a rational data type which behaves just like
any other number) but irrational numbers cause problems. But then can
an analogue machine represent irrational numbers exactly?
--
Lawrence Wilkinson ljw(a)formula1.demon.co.uk
The GirlFrendo homepage: http://www.formula1.demon.co.uk/girlfrendo/
"You've got the brains, or so you say, maybe you see things another
way"-bis
17 Mar
17 Mar
3:39 a.m.
On Mon, 16 Mar 1998, Lawrence Wilkinson wrote:
[1.0/3.0]
But a digital computer can represent rational numbers
exactly as you
have (e.g. Smalltalk has a rational data type which behaves just like
any other number) but irrational numbers cause problems.
That's interesting. How does SmallTalk deal with complex expressions?
Does it postpone evaluation until forced?
But then can
an analogue machine represent irrational numbers exactly?
Sure, values like pi and e are real-world analog constants, but "analog
machine" is pretty open-ended; at the lowest level, even digital machines
use analog processes.
It's been 10 years since I played with artificial neural nets, but the
basic goal was to define a neuron-like computing element that:
1) Could be arranged in networks that mapped arbitrary input sets to
arbitrary output sets.
2) Was simple enough to analyze mathematically.
3) Was trainable.
All of the ANNs ignored the distinctions between analog and digital,
as well as more obvious characteristics of real neurons.
The first try was the single-layer perceptron. A neuron simply summed
it's imputs and generated an output if the value exceeded a threshold.
You can make AND and OR maps with these, but not XOR, for example.
So, Minsky and Papert added a second layer, and showed that you could now
make XOR maps, but they couldn't find a way to train the net to build such
a map on its own.
Then Rummelhart (at UCSD) showed that by adding non-linearities (a
sigmoidal threshold function) and "back-propagating" errors through the
net, you could train a net to generate an XOR map (and many other maps).
For more complex mappings, there were lots of problems related to the time
it would take a net to converge, how to best layout the net, scalability
issues, how to model temporal associations, etc. Around this time, I had
to get a job, so I don't know the state of affairs today, but since today
there's more interest in Java than in neural nets, I assume they haven't
solved the problems yet.
-- Doug
7:54 a.m.
On 16 Mar 98 at 21:21, Lawrence Wilkinson wrote:
In article <Pine.LNX.3.95.980316132720.19589A-
100000(a)behemoth.host4u.net>gt;, Doug Yowza <yowza(a)yowza.com> writes
When I took my Electronics course , one of my greatest bafflements
and frustrations was dealing with the "approximations" of analogue
circuits. Needless to say it was a great relief when we moved on to
the exactitude of digital circuits.
ciao larry
lwalkerN0spaM(a)interlog.com
To be fair, analog computers can do things digital
computers can't. For
example, a digital computer can only approximate 1.0/3.0 whereas an
analog box has no trouble with this. Certain ops would also be much
faster with analog vs. digital, but I'd have to guess that these are
implementation issues that get lost in the noise.
But a digital computer can
represent rational numbers exactly as you
have (e.g. Smalltalk has a rational data type which behaves just like
any other number) but irrational numbers cause problems. But then can
an analogue machine represent irrational numbers exactly?
--
Lawrence Wilkinson ljw(a)formula1.demon.co.uk 
5:06 p.m.
A non-text attachment was scrubbed...
Name: not available
Type: text
Size: 1286 bytes
Desc: not available
Url :
http://www.classiccmp.org/pipermail/cctalk/attachments/19980317/0d1fcea4/at…
10:42 a.m.
At 09:21 PM 3/16/98 +0000, you wrote:
In article <Pine.LNX.3.95.980316132720.19589A-
100000(a)behemoth.host4u.net>gt;, Doug Yowza <yowza(a)yowza.com> writes
The obvious answer, from the non-PhD perspective, an analog computer tied
together with a digital computer. Let the analog computer decide which
problems to hand over to the digital computer, just like humans do when
they're working at the office.
To be fair, analog computers can do things digital
computers can't. For
example, a digital computer can only approximate 1.0/3.0 whereas an
analog box has no trouble with this. Certain ops would also be much
faster with analog vs. digital, but I'd have to guess that these are
implementation issues that get lost in the noise.
But a digital computer can
represent rational numbers exactly as you
have (e.g. Smalltalk has a rational data type which behaves just like
any other number) but irrational numbers cause problems. But then can
an analogue machine represent irrational numbers exactly? --
Lawrence Wilkinson ljw(a)formula1.demon.co.uk
--
David Wollmann
dwollmann(a)ibmhelp.com
11:08 a.m.
A non-text attachment was scrubbed...
Name: not available
Type: text
Size: 637 bytes
Desc: not available
Url :
http://www.classiccmp.org/pipermail/cctalk/attachments/19980317/aaa80ad2/at…
16 Mar
16 Mar
11:44 a.m.
On Mon, 16 Mar 1998, Doug Yowza wrote:
On Mon, 16 Mar 1998, David Wollmann wrote:
If you believe Turing, there's nothing an analog computer can compute that
a digital one can't. A brain is many things: it's wet, it's analog, and
it's massively parallel. I don't think anybody believes that it's wetness
or analogness that matters, but clearly a high degree of parallel
There was one engineer on the program that *did* believe just that. He was
making insects, 100% analog, and insisted that we could never emulate that
which is analog (brain activity) in the digital realm. His focus is on
creating a will-to-survive instinct in the machines, and then tricking
them into working for us. He likened it to putting blinders on an ox to
make it plow.
The early AI folk thought they could do brainstuff
using straight-forward
algorithms. Maybe you can, but there's no biological analog that anybody
can find to support the idea that humans work that way.
Who was it that mentioned Hofstadter's GEB the other day? He talks a lot
about the rules of reasoning in there, illuminating the seeming paradox of
writing a structured program that allows a machine to think freely. His
explanation of the language/metalanguage/metametalanguage hierarchies
makes one leap the chasm of the technology and just assume that it will
happen. I have read that book three times and can still find new things in
it.
So, the connectionists tried to create machines and
structures modeled
after the brain, but they didn't get too far. Let's say that you build
an OS and a programming language that allows you to accurately model a
brain. If you stick a human brain in front of a newspaper, you get
nothing. So add some input devices and actuators. Now you stick a baby
in front of a newspaper -- still nothing. So let the baby run for a
while, experience a variety of sensations, make a whole bunch of
associations, stick it in front of many newpapers and many non-newspapers
for many years, and finally you get a pretty good character recognizer.
Well, sure. That's the learning curve. But isn't that the idea of
replicating a machine? That it could duplicate it's RAM/ROM contents as
well? In the human world it would be like having direct access to a
million years of evolutionary experience! What couldn't we do if we new
everything that came before? What won't these machines be able to do?
We're still a long way from being able to put a
multi-billion node
connection machine onto a sturdy frame with millions of intricate wires,
actuators, and sensors, let it experience and manipulate the environment
for many years, and then get the thing to demonstrate emergent properties
that make the whole greater than the sum of the parts. But I don't see
any reason why it won't or can't happen.
I don't think we're that long off, in terms of evolutionary speed. If
computer evolution is occuring, by some estimates, at 5-10 million times
the speed of human evolution, we could expect a fully-evolved human
replicant within our lifetime! The technology is coming around, with
microelectronics and the promising prospect of nano-technology being
fulfilled.
And if you think that the prospect of human cloning is
causing moral
pangs, wait until somebody creates the first artificial life form!
One of the things I really liked about the Discovery program
is the quote about the state of technology, something to the effect of:
"We are a lot closer to being able to create an artificial human than we
are to being able to comprehend the consequences of creating an
artificial human."
I realise that this is not *directly* about +10 year old computer systems,
but it does directly relate to the them and their role in the history of
this field (which is what I originally asked the list about). Does anyone
on the list want to take it outside to a temporary list to discuss the
moral/ethical/probability issues of artificial life? Let me know by
email and I'll set one up.
Aaron
9795
days inactive
9796
days old
10 comments
7 participants
participants (7)
-
aaron＠wfi-inc.com -
ard＠p850ug1.demon.co.uk -
dwollmann＠ibmhelp.com -
ljw＠formula1.demon.co.uk -
lwalker＠mail.interlog.com -
shoppa＠alph02.triumf.ca -
yowza＠yowza.com