From Newsgroup: comp.lang.cobol

It's interesting to see that what I have been doing for the last half
century is now going to change, radically...

(Where have I heard that before?)

Take a look at this: http://www.infoworld.com/article/3160088/application-development/want-to-be-a-software-developer-time-to-learn-ai-and-data-science.html

I remember doing some experiments (in COBOL) back in the last century
after I read an article on heuristic programming. The result was a
program that could (eventually) find its way through a maze. There was
no guarantee that the way it found was the BEST way, but, provided the
maze COULD be solved, it would solve it.

Heuristic programs start off with a heuristic (a "rule of thumb", if you
like; for example: "Every time you encounter a wall, turn left".)
Another part of the program monitors progress and checks if the same
junction is being reached, so it detects if it is going in circles and modifies the heuristic at that point...). After a period of time or a
number of changes of direction (decided by the programmer initially, but
could be modified by the program in an advanced incarnation), the
heuristic is modified and the process is iterated. Obviously, depending
on the power of the processor, several billion alterations and tries can
be made in a relatively "short" period of time, and, eventually a
successful solution is reached. (The final heuristic depends on the size
of the maze, but it will typically be a long string of "Right"s and
"Left"s.)

I got really excited about this and discussed it with some of my peers
over coffee at work. The general reaction was one of horror...

The only way to "know" how the solution was reached would be to log
every change made (including the discarded ones) and the "audit trail" (containing billions of lines) would require several Human lifetimes to analyze, so you'd need another computer to do that...

One of my colleagues summed up the general feeling by saying:"Pete,
don't you find it a bit frightening that a computer can reach a result
without anyone knowing how it got there?" (As I had programs under development for work, that did that every day [it's called:
"Debugging"], I wasn't in the least bit phased by it...)

I remember replying: "No, I find it exciting...".

The applications of heuristic programming were kind of limited to things
like production planning (I remember looking at an heuristic corrugator cardboard scheduler, written in Fortran, which determined the best
places to cut long runs of corrugated cardboard for packaging. One of my colleagues, a young girl who had just graduated from Auckland University
of Technology [this establishment no longer exists...] took the
corrugator scheduler "under her wing" and became a world expert in the
use of the package.)

All of these events were in the early 1970s.

They were, at least for me, the first inkling that a computer maybe did
not have to be told how to do everything... the first glimmers of
"artificial intelligence".

Subsequently, as IT gathered momentum and some of the world's best and brightest were attracted to the newly emerging "Computer Science", the
early attempts at neural networks were developed, with inconclusive
results. There was no shortage of scoffers and naysayers, and in some
cases, the sheer arrogance of "computer programmers" (how could an
inanimate machine ever be able to do what I do? Without me, that machine
is just a useless assembly of wires and metal...) made me want to punch people.

But, today we are poised on the brink of another great leap.

The deep networks and heuristics used in today's AI are quantum leaps
above the first stumbling steps discussed above.

Data CAN be used to train systems and that will just be the beginning.

I remember writing a database management system called "DIC" - Database Interface Control - (for VSAM/KSDS on IBM mainframes), back in the late
1970s. It was eventually superseded when the company standardized on the
new Relational Database technology in the mid 1980s and went to DB2. But
for around 6 years DIC managed the distribution of photo-copiers across
14 countries in Europe and it was well received by everybody except the French. (No surprises there... :-))

The point is that when I designed and built it I remember saying to my
team: "Because DIC will control the interface to the data, we could
write a module that will monitor how and where data is accessed. We
could then use that information to manually adjust and organize the
physical VSAM files for optimal access...or, we could let DIC interface
to IDCAMS and do it automatically."

A case of the actual data causing re-organization of itself, depending
on how it was being accessed. (Most modern RDB systems have modules that
can do this, at least at a rudimentary level. Good DBAs use this data to control physical organization of datasets.)

I posted the link above because there may be some people here who share
my enthusiasm for the idea of computers programming themselves. The
article is quick to point out that a Human is required to broker the
data being used for training, so they postulate that the ROLE of a
programmer will change.

"Releases will be based on the software being trained to another level; updates will be based on new data sets and experiences."

This conjures the image of a programmer as a "lion tamer", whip in one
hand, template in the other... :-)

with that happy thought I have to get back to writing some particularly complex code that has had my head spinning for the last 2 days... I'm
only sorry that I probably won't see the day when THAT code could write itself...

Pete.
--
I used to write COBOL; now I can do anything...

SEEN-BY: 154/30 2320/100 0 1 227/0

From Newsgroup: comp.lang.cobol

On Wednesday, January 25, 2017 at 4:30:42 PM UTC+13, pete dashwood wrote:

Auckland University
of Technology [this establishment no longer exists...]

You have that the wrong way around. AUT is still around. In 2000 the Auckland Technical Institute (ATI) was granted university status and renamed itself to AUT.

So, in the 70s she would have graduated from ATI, but that still exists just renamed to AUT, and still uses the Wellesley St campus but has several others as well.

SEEN-BY: 154/30 2320/100 0 1 227/0

From Newsgroup: comp.lang.cobol

On 25/01/2017 5:22 p.m., Richard wrote:

On Wednesday, January 25, 2017 at 4:30:42 PM UTC+13, pete dashwood wrote:

Auckland University
of Technology [this establishment no longer exists...]

You have that the wrong way around. AUT is still around. In 2000 the Auckland

Technical Institute (ATI) was granted university status and renamed itself to AUT.

So, in the 70s she would have graduated from ATI, but that still exists just

renamed to AUT, and still uses the Wellesley St campus but has several others as well.

Thanks Richard.

I got confused from seeing "AUT" on University Challenge... :-)

It was, of course, "ATI".

And yes, I remember attending some COBOL-related meetings at the
Wellesley Street campus of ATI back around that time.

Cheers,

Pete.

--
I used to write COBOL; now I can do anything...

SEEN-BY: 154/30 2320/100 0 1 227/0

From Newsgroup: comp.lang.cobol

In article <eeqkevF17rfU1@mid.individual.net>,
pete dashwood <dashwood@enternet.co.nz> wrote:

It's interesting to see that what I have been doing for the last half >century is now going to change, radically...

(Where have I heard that before?)

Take a look at this: >http://www.infoworld.com/article/3160088/application-development/want-to-be-a-software-developer-time-to-learn-ai-and-data-science.html

[snip]

One of my colleagues summed up the general feeling by saying:"Pete,
don't you find it a bit frightening that a computer can reach a result >without anyone knowing how it got there?" (As I had programs under >development for work, that did that every day [it's called:
"Debugging"], I wasn't in the least bit phased by it...)

I remember replying: "No, I find it exciting...".

There should be another half-century of experience under your belt, Mr Dashwood. How many times in those decades have you been excited by being
told 'The computer tells us (blatant and obvious representation of an alternate reality which bears no relation on the one in which you believe
to find yourself and causes you inconvenience) so it must be so'?

For example... being required to register for school at age 108. Being notified that one is in default of a fully-paid loan. Receiving magazine subscription-notices cheerfully declaring 'Now That You're Dead, You're Eligible for a Discount!'

In the article 'Jim McHugh, vice president and general manager for
Nvidia's DGX-1 supercomputer', asserts 'We're using data to train the
software to make it more intelligent.'

The assumption that intelligence is the result of training causes me to
grin, albeit wearily.

There was no shortage of scoffers and naysayers, and in some
cases, the sheer arrogance of "computer programmers" (how could an
inanimate machine ever be able to do what I do? Without me, that machine
is just a useless assembly of wires and metal...) made me want to punch >people.

Wanting to punch people for arrogance is a sure sign of humility... and I
am the King of England, God save the Me!

DD

SEEN-BY: 154/30 2320/100 0 1 227/0

From Newsgroup: comp.lang.cobol

On Wednesday, 25 January 2017 13:30:42 UTC+10, pete dashwood wrote:

It's interesting to see that what I have been doing for the last half
century is now going to change, radically...

(Where have I heard that before?)

Take a look at this: http://www.infoworld.com/article/3160088/application-development/want-to-be-a-software-developer-time-to-learn-ai-and-data-science.html

I remember doing some experiments (in COBOL) back in the last century
after I read an article on heuristic programming. The result was a
program that could (eventually) find its way through a maze. There was
no guarantee that the way it found was the BEST way, but, provided the
maze COULD be solved, it would solve it.

Heuristic programs start off with a heuristic (a "rule of thumb", if you like; for example: "Every time you encounter a wall, turn left".)
Another part of the program monitors progress and checks if the same
junction is being reached, so it detects if it is going in circles and modifies the heuristic at that point...). After a period of time or a
number of changes of direction (decided by the programmer initially, but could be modified by the program in an advanced incarnation), the
heuristic is modified and the process is iterated. Obviously, depending
on the power of the processor, several billion alterations and tries can
be made in a relatively "short" period of time, and, eventually a
successful solution is reached. (The final heuristic depends on the size
of the maze, but it will typically be a long string of "Right"s and
"Left"s.)

I got really excited about this and discussed it with some of my peers
over coffee at work. The general reaction was one of horror...

The only way to "know" how the solution was reached would be to log
every change made (including the discarded ones) and the "audit trail" (containing billions of lines) would require several Human lifetimes to analyze, so you'd need another computer to do that...

One of my colleagues summed up the general feeling by saying:"Pete,
don't you find it a bit frightening that a computer can reach a result without anyone knowing how it got there?" (As I had programs under development for work, that did that every day [it's called:
"Debugging"], I wasn't in the least bit phased by it...)

I remember replying: "No, I find it exciting...".

The applications of heuristic programming were kind of limited to things
like production planning (I remember looking at an heuristic corrugator cardboard scheduler, written in Fortran, which determined the best
places to cut long runs of corrugated cardboard for packaging. One of my colleagues, a young girl who had just graduated from Auckland University
of Technology [this establishment no longer exists...] took the
corrugator scheduler "under her wing" and became a world expert in the
use of the package.)

All of these events were in the early 1970s.

They were, at least for me, the first inkling that a computer maybe did
not have to be told how to do everything... the first glimmers of
"artificial intelligence".

Subsequently, as IT gathered momentum and some of the world's best and brightest were attracted to the newly emerging "Computer Science", the
early attempts at neural networks were developed, with inconclusive
results. There was no shortage of scoffers and naysayers, and in some
cases, the sheer arrogance of "computer programmers" (how could an
inanimate machine ever be able to do what I do? Without me, that machine
is just a useless assembly of wires and metal...) made me want to punch people.

But, today we are poised on the brink of another great leap.

The deep networks and heuristics used in today's AI are quantum leaps
above the first stumbling steps discussed above.

Data CAN be used to train systems and that will just be the beginning.

I remember writing a database management system called "DIC" - Database Interface Control - (for VSAM/KSDS on IBM mainframes), back in the late 1970s. It was eventually superseded when the company standardized on the
new Relational Database technology in the mid 1980s and went to DB2. But
for around 6 years DIC managed the distribution of photo-copiers across
14 countries in Europe and it was well received by everybody except the French. (No surprises there... :-))

The point is that when I designed and built it I remember saying to my
team: "Because DIC will control the interface to the data, we could
write a module that will monitor how and where data is accessed. We
could then use that information to manually adjust and organize the
physical VSAM files for optimal access...or, we could let DIC interface
to IDCAMS and do it automatically."

A case of the actual data causing re-organization of itself, depending
on how it was being accessed. (Most modern RDB systems have modules that
can do this, at least at a rudimentary level. Good DBAs use this data to control physical organization of datasets.)

I posted the link above because there may be some people here who share
my enthusiasm for the idea of computers programming themselves. The
article is quick to point out that a Human is required to broker the
data being used for training, so they postulate that the ROLE of a
programmer will change.

"Releases will be based on the software being trained to another level; updates will be based on new data sets and experiences."

This conjures the image of a programmer as a "lion tamer", whip in one
hand, template in the other... :-)

with that happy thought I have to get back to writing some particularly complex code that has had my head spinning for the last 2 days... I'm
only sorry that I probably won't see the day when THAT code could write itself...

Pete.
--
I used to write COBOL; now I can do anything...

Hi Pete
Interesting article.
I like the idea that Data CAN be used to train systems and that will just be the beginning.
I am a bit of a skeptic about media views of AI but can see it advancing.
I think that Google is doing some of this today with the way a browser remembers your searches and anticipates keystrokes. For instance all I have to do is type "you" and Google anticipates www.youtube.com. This is intelligence by keeping and using data. It is even specific to my logon. There is also worldwide sharing of data so one can place a photo in Google Earth for all the world to see or find by search. I realize this is short of what you are talking
about.
I have a challenge for you.
Which data structure does Google use in its search engine or does it have its own data structure?
Is it a conventional database or discrete ISAM files or other?
In the late 1970's, Digital Equipment Corporation (DEC) (remember them?) produced a promotional video and within they said something like the computer is the evolution of the human mind. So we humans inject our intelligence into machines. 40 years later we are still doing it as a slow process with acceleration.
Today, I think me best internet friends are Google Chrome and YouTube and throw
in email. I can find and learn so much because of this worldwide sharing of "searchable" data.
Advertising is trying very hard to use AI by examining your searches and buying
habits and targeting ads. For instance, if Coles finds you bought disposable nappies, their AI may assume there is a baby and therefore target ads for other
baby products. The problem is linking a credit card purchase to an online user. Greg

SEEN-BY: 154/30 2320/100 0 1 227/0

From Newsgroup: comp.lang.cobol

On 26/01/2017 1:47 a.m., docdwarf@panix.com wrote:

In article <eeqkevF17rfU1@mid.individual.net>,
pete dashwood <dashwood@enternet.co.nz> wrote:

It's interesting to see that what I have been doing for the last half
century is now going to change, radically...

(Where have I heard that before?)

Take a look at this:
http://www.infoworld.com/article/3160088/application-development/want-to-be-a-software-developer-time-to-learn-ai-and-data-science.html

[snip]

One of my colleagues summed up the general feeling by saying:"Pete,
don't you find it a bit frightening that a computer can reach a result
without anyone knowing how it got there?" (As I had programs under
development for work, that did that every day [it's called:
"Debugging"], I wasn't in the least bit phased by it...)

I remember replying: "No, I find it exciting...".

There should be another half-century of experience under your belt, Mr Dashwood.

Please don't wish that on me. I have enjoyed my career and I love what I
do, but I don't think another 50 years would be appealing. The IT
workplace is a demanding one and, as time has passed, I have come to
value the tranquility of coding in an armchair (or, indeed, by a river bank...) over the frenetic stress of the cubicle and bullpen.

How many times in those decades have you been excited by being

told 'The computer tells us (blatant and obvious representation of an alternate reality which bears no relation on the one in which you believe
to find yourself and causes you inconvenience) so it must be so'?

Many, many, times, Doc. Although that is not quite the same excitement
because it means the tedium of debugging must be undergone...

However, having studied information theory (and read Marshall McLuhan) I
am well aware that "the media is the message" so if it's on green
lineflo (or, these days, a computer screen...), it must be true... :-)

When I first encountered this idea I was sceptical... surely Managers
couldn't be so dumb as to believe everything that came from a computer?
Could they...?

I changed a Management Reporting program so that it gave completely
untrue and quite ridiculous numbers, just to see if anybody noticed...

All of the Upper Management, with the sole exception of the CEO,
believed the Company was facing collapse and disaster.

The only reason I didn't get fired was because the CEO had a sense of
humour and also realized the point I was trying to make. (To be fair, my
Boss, after giving me a good dressing down, sprang to my defence and
said I was a valuable employee...) I was 23 years old at the time...

I lost an increment and some bonuses but it was worth it for the
experience.

For example... being required to register for school at age 108. Being notified that one is in default of a fully-paid loan. Receiving magazine subscription-notices cheerfully declaring 'Now That You're Dead, You're Eligible for a Discount!'

I never suggested that computers are always right... :-)

I just like the idea that it may be possible for them to evolve into
something that can program itself and develop a completely new kind of "intelligence".

In the article 'Jim McHugh, vice president and general manager for
Nvidia's DGX-1 supercomputer', asserts 'We're using data to train the software to make it more intelligent.'

The assumption that intelligence is the result of training causes me to
grin, albeit wearily.

I take your point. However, in this case, if the training DOESN'T result
in "intelligence" then the algorithms used are modified and the process iterates until an intelligent result DOES eventuate.

One way that Humans learn is by "trial and error". I see no reason why software can't do so too. The advantage that the machines have over us
is that they can make billions of trials in a very short time and
quickly discard what doesn't work. The advantage we have over them is
that we perceive more than just the data being presented for the trial,
so we can learn from fewer attempts, but there is no reason I can see,
to assert that one approach is "better" than the other...

There was no shortage of scoffers and naysayers, and in some
cases, the sheer arrogance of "computer programmers" (how could an
inanimate machine ever be able to do what I do? Without me, that machine
is just a useless assembly of wires and metal...) made me want to punch
people.

Wanting to punch people for arrogance is a sure sign of humility... and I
am the King of England, God save the Me!

I never indulged the impulse... OK, maybe once, but that was outside
working hours so it doesn't count... :-)

(Besides, warnings were given and ignored, so he had it coming. He
agreed as much the following day when I apologised, but I don't
recommend violence as a preferred tool for Project Managers.)

Pete.
--
I used to write COBOL; now I can do anything...

SEEN-BY: 154/30 2320/100 0 1 227/0

From Newsgroup: comp.lang.cobol

On 26/01/2017 2:42 p.m., Greg Wallace wrote:

On Wednesday, 25 January 2017 13:30:42 UTC+10, pete dashwood wrote:

It's interesting to see that what I have been doing for the last half
century is now going to change, radically...

(Where have I heard that before?)

Take a look at this:
http://www.infoworld.com/article/3160088/application-development/want-to-be-a-software-developer-time-to-learn-ai-and-data-science.html

<snipped>

Hi Pete

Interesting article.

Thanks, I'm glad you enjoyed it. I debated whether to post it to my
account on LinkedIN but it's really more oriented to software developers
and CLC is a place I really value as one of the last uncensored forums
on the planet. I'd hate to see it disappear, even though the original
reasons I came here have long been overtaken by events.

I like the idea that Data CAN be used to train systems and that will just be

the beginning.

I am a bit of a skeptic about media views of AI but can see it advancing.

Sure, we need to balance what we get through media, against experience
and common sense...

I think that Google is doing some of this today with the way a browser

remembers your searches and anticipates keystrokes. For instance all I have to do is type "you" and Google anticipates www.youtube.com. This is intelligence by keeping and using data. It is even specific to my logon. There is also worldwide sharing of data so one can place a photo in Google Earth for all the world to see or find by search. I realize this is short of what you are talking
about.

It is a fair, if basic, example.

I have a challenge for you.
Which data structure does Google use in its search engine or does it have its

own data structure?

It is a proprietary file structure distributed across millions of servers.

Obviously, in the highly competitive world of SEO, GOOGLE play their
cards close to their chest and details of their algorithms are closely guarded. (The main search algorithms change quite frequently, partly
because they need to and partly to prevent reverse engineering and
people gaining unfair advantage.)

What we can say with a fair amount of certainty is that the engine falls
into two parts: Crawlers and Indexers.

There are MANY instances of each of these components, working 24/7. (The GOOGLE engines were originally developed by Larry Page and Sergey Brin
and they called it "Back Rub" because it was designed to explore back
links on the web. The programming languages were Java and Python. Python
is still used today for Crawlers, but the indexers are now in C/C++. The
use of OO languages facilitates the instancing of MANY objects
(components) from a single Class template.)

Is it a conventional database or discrete ISAM files or other?

It is "other". The structure used by the indexers is proprietary to
GOOGLE. These structures are held on the server where the page resides
and that is part of what makes the searches so fast.

In the late 1970's, Digital Equipment Corporation (DEC) (remember them?)

Yes, I remember DEC VAX and even back to PDP 8... did you know that in
the 70s the US could not export computer technology to countries that
were considered "unfriendly"? After closing a multi-million dollar deal
with a couple of Eastern European nations, this particular Law
threatened to be a show-stopper. So, they said: "It's NOT a computer;
it's a "Programmable Data Processor"... " The sale went through.

I'm writing this on an HP laptop; Compaq acquired DEC, who in turn are
owned by HP... the wheel keeps turning.

produced a promotional video and within they said something like the computer is the evolution of the human mind. So we humans inject our intelligence into machines. 40 years later we are still doing it as a
slow process with acceleration.

I'm not sure that that is a fair way to put it, because I believe the
machines will have a different "intelligence" to that which we have, but
it is certainly true that, whatever they end up with, will have been
initially sparked by Humans.

Today, I think me best internet friends are Google Chrome and YouTube and

throw in email. I can find and learn so much because of this worldwide sharing of "searchable" data.

In the latter part of the Middle Ages gentlemen went to a "University"
where they acquired a universal education... everything from Latin &
Greek, through history and Mathematics, to things like Fencing, Manners
and Drawing. The sum total of Human knowledge could fit into a few
University libraries... If you studied for a lifetime you could end up
with a fair handle on just about everything that was known. Graduates
could be considered to be "spherical" in their education.

By the beginning of the 19th century, the "knowledge explosion" was
beginning and it is continuing even as I write this. The Sum total of
Human knowledge is now so vast that all the libraries in the world can't contain it, because books cannot be written and printed fast enough to
keep up with it. It is no longer possible to get a "spherical"
education; you have to specialize. Graduates now are generally
"conical", pointed to a particular specialization... :-)

The only thing that even gives us a chance of not losing everything we
have learned and reverting back to the Dark Ages, is the storage and distribution of information on electronic media as soon as it becomes available.

The Internet (a network of networks) may well be a two-edged sword but
life as we currently enjoy it would be impossible without it.

Advertising is trying very hard to use AI by examining your searches and

buying habits and targeting ads. For instance, if Coles finds you bought disposable nappies, their AI may assume there is a baby and therefore target ads for other baby products.

Exactly. Although we all bitch about ads, they are one of the things
that finances and drives the advancement of the Internet.

The problem is linking a credit card purchase to an online user.

Greg

Thanks for your post, Greg.

Pete.
--
I used to write COBOL; now I can do anything...

SEEN-BY: 154/30 2320/100 0 1 227/0

From Newsgroup: comp.lang.cobol


"Greg Wallace" <gregwebace@gmail.com> wrote in message news:366cacfb-5c40-4470-8b4a-225e9a03e04a@googlegroups.com...
On Wednesday, 25 January 2017 13:30:42 UTC+10, pete dashwood wrote:

It's interesting to see that what I have been doing for the last half
century is now going to change, radically...

(Where have I heard that before?)

Take a look at this: http://www.infoworld.com/article/3160088/application-development/want-to-be-a-software-developer-time-to-learn-ai-and-data-science.html

I remember doing some experiments (in COBOL) back in the last century
after I read an article on heuristic programming. The result was a
program that could (eventually) find its way through a maze. There was
no guarantee that the way it found was the BEST way, but, provided the
maze COULD be solved, it would solve it.

Heuristic programs start off with a heuristic (a "rule of thumb", if you like; for example: "Every time you encounter a wall, turn left".)
Another part of the program monitors progress and checks if the same
junction is being reached, so it detects if it is going in circles and modifies the heuristic at that point...). After a period of time or a
number of changes of direction (decided by the programmer initially, but could be modified by the program in an advanced incarnation), the
heuristic is modified and the process is iterated. Obviously, depending
on the power of the processor, several billion alterations and tries can
be made in a relatively "short" period of time, and, eventually a
successful solution is reached. (The final heuristic depends on the size
of the maze, but it will typically be a long string of "Right"s and
"Left"s.)

I got really excited about this and discussed it with some of my peers
over coffee at work. The general reaction was one of horror...

The only way to "know" how the solution was reached would be to log
every change made (including the discarded ones) and the "audit trail" (containing billions of lines) would require several Human lifetimes to analyze, so you'd need another computer to do that...

One of my colleagues summed up the general feeling by saying:"Pete,
don't you find it a bit frightening that a computer can reach a result without anyone knowing how it got there?" (As I had programs under development for work, that did that every day [it's called:
"Debugging"], I wasn't in the least bit phased by it...)

I remember replying: "No, I find it exciting...".

The applications of heuristic programming were kind of limited to things
like production planning (I remember looking at an heuristic corrugator cardboard scheduler, written in Fortran, which determined the best
places to cut long runs of corrugated cardboard for packaging. One of my colleagues, a young girl who had just graduated from Auckland University
of Technology [this establishment no longer exists...] took the
corrugator scheduler "under her wing" and became a world expert in the
use of the package.)

All of these events were in the early 1970s.

They were, at least for me, the first inkling that a computer maybe did
not have to be told how to do everything... the first glimmers of
"artificial intelligence".

Subsequently, as IT gathered momentum and some of the world's best and brightest were attracted to the newly emerging "Computer Science", the
early attempts at neural networks were developed, with inconclusive
results. There was no shortage of scoffers and naysayers, and in some
cases, the sheer arrogance of "computer programmers" (how could an
inanimate machine ever be able to do what I do? Without me, that machine
is just a useless assembly of wires and metal...) made me want to punch people.

But, today we are poised on the brink of another great leap.

The deep networks and heuristics used in today's AI are quantum leaps
above the first stumbling steps discussed above.

Data CAN be used to train systems and that will just be the beginning.

I remember writing a database management system called "DIC" - Database Interface Control - (for VSAM/KSDS on IBM mainframes), back in the late 1970s. It was eventually superseded when the company standardized on the
new Relational Database technology in the mid 1980s and went to DB2. But
for around 6 years DIC managed the distribution of photo-copiers across
14 countries in Europe and it was well received by everybody except the French. (No surprises there... :-))

The point is that when I designed and built it I remember saying to my
team: "Because DIC will control the interface to the data, we could
write a module that will monitor how and where data is accessed. We
could then use that information to manually adjust and organize the
physical VSAM files for optimal access...or, we could let DIC interface
to IDCAMS and do it automatically."

A case of the actual data causing re-organization of itself, depending
on how it was being accessed. (Most modern RDB systems have modules that
can do this, at least at a rudimentary level. Good DBAs use this data to control physical organization of datasets.)

I posted the link above because there may be some people here who share
my enthusiasm for the idea of computers programming themselves. The
article is quick to point out that a Human is required to broker the
data being used for training, so they postulate that the ROLE of a
programmer will change.

"Releases will be based on the software being trained to another level; updates will be based on new data sets and experiences."

This conjures the image of a programmer as a "lion tamer", whip in one
hand, template in the other... :-)

with that happy thought I have to get back to writing some particularly complex code that has had my head spinning for the last 2 days... I'm
only sorry that I probably won't see the day when THAT code could write itself...

Pete.
--
I used to write COBOL; now I can do anything...

Hi Pete

Interesting article.

I like the idea that Data CAN be used to train systems and that will just be the beginning.

I am a bit of a skeptic about media views of AI but can see it advancing.

I think that Google is doing some of this today with the way a browser remembers your searches and anticipates keystrokes. For instance all I have
to do is type "you" and Google anticipates www.youtube.com. This is intelligence by keeping and using data. It is even specific to my logon.
There is also worldwide sharing of data so one can place a photo in Google Earth for all the world to see or find by search. I realize this is short of what you are talking about.

I have a challenge for you.
Which data structure does Google use in its search engine or does it have
its own data structure?
Is it a conventional database or discrete ISAM files or other?


In the late 1970's, Digital Equipment Corporation (DEC) (remember them?) produced a promotional video and within they said something like the
computer is the evolution of the human mind. So we humans inject our intelligence into machines. 40 years later we are still doing it as a slow process with acceleration.

Today, I think me best internet friends are Google Chrome and YouTube and throw in email. I can find and learn so much because of this worldwide
sharing of "searchable" data.

Advertising is trying very hard to use AI by examining your searches and buying habits and targeting ads. For instance, if Coles finds you bought disposable nappies, their AI may assume there is a baby and therefore target ads for other baby products. The problem is linking a credit card purchase
to an online user.

Greg


In the U.S. we have a store named Target and they will send you specific ads and coupons based on what you have previously purchased. They have a
program that predicts what you may be interested in buying. One father
became irate when his household received ads targeted for someone who was going to have a baby. He called up and complained because his wife was too old and his daughter was too young and no one in his household could
possibly be pregnant. Later when he talked to his wife about it he
discovered that his daughter was pregnant and they had been hiding that fact from him. So Target deduced it before the father even knew about it.

SEEN-BY: 154/30 2320/100 0 1 227/0

From Newsgroup: comp.lang.cobol

On 26/01/2017 5:31 p.m., Charles Hottel wrote:
<snipped>

In the U.S. we have a store named Target and they will send you specific ads and coupons based on what you have previously purchased. They have a
program that predicts what you may be interested in buying. One father
became irate when his household received ads targeted for someone who was going to have a baby. He called up and complained because his wife was too old and his daughter was too young and no one in his household could
possibly be pregnant. Later when he talked to his wife about it he discovered that his daughter was pregnant and they had been hiding that fact from him. So Target deduced it before the father even knew about it.

Is it insecurity or spiteful shadenfreude that makes most of us happy
when a machine screws up? A little reminder that Humans are still
smarter... even though there are some VERY dumb Humans about (see the
Darwin awards...). We conveniently forget that, currently, most of the programming is done by Humans...

The day will come when the machines WON'T screw up (it might well
coincide with the time when they no longer need Humans to program
them...) and you then have a choice of accepting you have been
superseded (and facing the consequent pointlessness of your existence),
or throwing a party for them... :-)

Personally, I like the party idea, but I doubt I'll be here when it happens.

Pete.
--
I used to write COBOL; now I can do anything...

SEEN-BY: 154/30 2320/100 0 1 227/0

From Newsgroup: comp.lang.cobol

In article <s4KdnYf01coa5hTFnZ2dnUU7-QXNnZ2d@earthlink.com>,
Charles Hottel <chottel@earthlink.net> wrote:

[snip]

In the U.S. we have a store named Target and they will send you specific ads >and coupons based on what you have previously purchased. They have a >program that predicts what you may be interested in buying. One father >became irate when his household received ads targeted for someone who was >going to have a baby. He called up and complained because his wife was too >old and his daughter was too young and no one in his household could >possibly be pregnant. Later when he talked to his wife about it he >discovered that his daughter was pregnant and they had been hiding that fact >from him. So Target deduced it before the father even knew about it.

When this story came out last year my thought was 'if this is true then
the father or the daughter stands to make a Great Deal of Money by
exploiting these circumstance. 'Hi, I'm Barbara/Bobby and Target knows me
SO well that...'

DD

SEEN-BY: 154/30 2320/100 0 1 227/0

From Newsgroup: comp.lang.cobol

In article <eet3eeF8ikpU1@mid.individual.net>,
pete dashwood <dashwood@enternet.co.nz> wrote:

[snip]

I just like the idea that it may be possible for them to evolve into >something that can program itself and develop a completely new kind of >"intelligence".

Many likeable ideas are problematic in practise; 'always act justly' seems downright fine until a situation needs mercy.

DD

SEEN-BY: 154/30 2320/100 0 1 227/0