Re: How Robots Will Steal Your Job

[goose]

We'll be half way there for a little more time, then there will
suddenly be rapid progress. This will happen when computers
approach and likely supersede the performance and ultimately the
intelligence of the human brain.

I dunno about that. How can we build a machine that can supercede the
human brain ?

If the human brain was simple enough to understand, we would be too
simple too understand it (this may already be the case). All we know
about the brain is its tissue composition.

I guess it will happen around
2020, but I could be off by up to 10 years.

I think that your margin of error might be greater than you realise.
this problem can almost never be solved. We can build vast machinary
to *simulate* some aspects of the human brain (like the ability to
play chess), but unless someone hands us the blueprints and design
docs for the human brain, we are, by definition, not smart enough
to figure it out.

a) Future computers will have that vast encyclopedic store.
Actually, in a way they have it even today. Entire libraries are
already stored in computers. As soon as they begin to make sense
of written human language, they can begin to make use of that
huge store.

getting a computer to *think* is kinda like getting a submarine
to swim (or something like that .

computers can only make sense of something if *we* can make sense
of it, because *we* tell the computer what to do. in the game of chess
we try to think X moves ahead, so when we write the chess playing
proggy its much easier to merely depth/breadth search.

Even though the computer still does thinks *hundreds* of moves ahead
of a human player, a human player can still beat it.

its not just about the volume of information. its about how it is used.


hand,
goose,

 

[goose]

Something I have thought would be interesting would be a composition
program that just noodled away and monitored your pleasure via various
brain or bodily sensors. It might somehow gradually learn to compose
music that you liked, or draw pornographic cartoons you found
irresistible.

now *thats* technology at work !!!

goose,
usenet might be the best place to test such a 'bot' ...
if it passes the turing test and learns to flame, hell, then
we have a winner

 

[Corey Murtagh]

Misquoting corrected.....

[Top-posting fixed]
Airy R Bean wrote:


That is your prerogative. But it seems a strange way of being
intolerant to your fellow men. Top posters do not have this infantile
attitude towards bottom posters, I note in passing.


Airy R Bean wrote:
ignored due to persistent top-posting.

I didn't misquote you. In my reply, I moved your words underneath the words

<snip>

DFTT Richard. I know it's annoying, but just ignore him.

 

[gabor salai]

(This is certainly true, and consequently AI has an impossible marketing job
on its hands.)


Humans are very good at recognising patterns, but they are not /just/
pattern recognising machines, unless you define the concept of pattern
recognition so widely as to make it useless.

just an opinion:
pattern (of any kind) recognition (matching with internal database)
is basicaly very simple task.
after recognition, whatever matched action may be started, etc ...

"patternisation" seems to be the problem

 

[Bent C Dalager]

If the human brain was simple enough to understand, we would be too
simple too understand it (this may already be the case). All we know
about the brain is its tissue composition.

I am aware of no fundamental law that says a system must be incapable
of understanding itself. A system based on surprisingly simple
concepts can become mindboggingly complex once the number of
components starts growing large. And yet, all you need to understand
to be able to describe the system are those simple concepts and a very
powerful computer to simulate large scale integration.

I think that your margin of error might be greater than you realise.
this problem can almost never be solved. We can build vast machinary
to *simulate* some aspects of the human brain (like the ability to
play chess), but unless someone hands us the blueprints and design
docs for the human brain, we are, by definition, not smart enough
to figure it out.

Do you have a reference to a proof for that assertion?

Cheers
Bent D

 

[Hans-Georg Michna]

On Sun, 24 Aug 2003 10:50:40 +0200, Hans-Georg Michna

Every time we solve a problem, e.g. playing chess, we say "Oh that was
just a cheap trick. That does not count as intelligence."

I have a conjecture, that human intelligence is a collection of no
more than 5 basic cheap tricks. It evolved so quickly I don't think it
ran really be all that big a deal above the intelligence of the
bonobo.

Roedy,

I think along the same lines. Anthropocentrism leads people to
assign mystical abilities to our poor brain. But a glance at a
normal IQ test already shows you the strict limits of what we
can achieve.

We do have some fairly powerful specialized abilities for
hearing and particularly for 3D vision though.

On the other hand, our brain is three or four times the size of
a chimpanzee brain, and this size increase happened in only
about 4 million years, which is astounding. There must be
something we can do that the chimps can't. We build cities and
spaceships.

But I still agree with what you wrote.

Hans-Georg

 

[Hans-Georg Michna]

computers can only make sense of something if *we* can make sense
of it, because *we* tell the computer what to do. in the game of chess
we try to think X moves ahead, so when we write the chess playing
proggy its much easier to merely depth/breadth search.

Even though the computer still does thinks *hundreds* of moves ahead
of a human player, a human player can still beat it.

Goose,

we may one day be able to scan the brain and use its structure,
but there are other ways. One example is accelerated Darwinistic
evolution in a computer simulation, augmented by large knowledge
databases and perhaps some human programming.

But your chess example will lead you nowhere, because you should
never underestimate the power of brute force algorithms when you
put a large and growing amount of data processor performance
underneath. It is quite obvious that the best chess players on
our planet are already or will soon be computers. And you forget
that only one person managed to beat the computer. 99%+ of
humanity can be beat in chess by a very cheap microprocessor.

Hans-Georg
http://www.michna.com/transition.htm

 

[Calum]

Because it's a dumbed down version of the object oriented derivative
of a horribly designed imperative PL known as C?

I can't believe I'm about to jump to the defense of C... but I am.
Remember it did (and still does) have a huge user community. Therefore,
there must be some redeeming features, which I believe is practicality.

Sure, theoretically, it's not nice, but it has been formalized. But
since when does formalization imply practicality or benefit?

I wouldn't really
comment on that on a Java group. LOL. Suffice it to say that I
wouldn't use that language for any amount of serious CS research
(parallel, AI, etc.)

Programming languages need to be practical as well as theoretically
nice. Perhaps you should ask why people don't use research languages?

There's a world of difference between the research and user communities.
Practical languages like C, Perl, Java, C++, do not bend nicely into
theoretical models. Research languages are small and neat, principally
because research is often done by individuals who can't take on too much
programming, and in practise develop just one facet of a language, not
all-round practicality.

As it happens, I have seen quite a few research projects on extensions
to Java, and compiling (say) functional languages to the JVM or the .NET
architecture, so obviously many people disagree with you.

Java is hampered by its run-time overheads. It's much much cleaner than
say C++. You're contradicting yourself: you say it's dumbed-down, which
you insinuate is bad, yet, as a result of "dumbing down", its semantics
are cleaner, which from a research perspective is much better.

Calum

 

[rkm]

Intelligence is the ability to home in on a desirable solution
quickly. Evolution by genetic mutation and natural selection has IQ 1.

Well then the computer will never be intelligent, because it
can't home in on the solution by itself, it has to have
another "species" in the loop, either so it can measure the
other species reaction (which is a result of judgement on
the other species part), or it needs the other species to
tell it what a desirable solution is. By itself, it's just
a dumb bunch of parts.

 

[Bent C Dalager]

On the other hand, our brain is three or four times the size of
a chimpanzee brain, and this size increase happened in only
about 4 million years, which is astounding. There must be
something we can do that the chimps can't. We build cities and
spaceships.

There's a theory that what made us develop our obscenely large brains
was the need to socialize with other humans. There were contenders to
the title, of course, but we killed them. Eradicate the humans and
give chimps another few mill years and perhaps their desire to make
friends has put them on the moon also

Cheers
Bent D

 

[rkm]

But your chess example will lead you nowhere, because you should
never underestimate the power of brute force algorithms when you
put a large and growing amount of data processor performance
underneath. It is quite obvious that the best chess players on
our planet are already or will soon be computers. And you forget
that only one person managed to beat the computer. 99%+ of
humanity can be beat in chess by a very cheap microprocessor.

Brute force searching is not intelligence. I would question
whether these computers have developed new unexpected
heuristics for playing the game? AFAIK, Deeper Blue (or
whatever it's name is) is just applying rules programmed
into it from the outset, not anything creative on its part.

Furthermore, could we tell it, now that it supposedly
understands chess, to design a new game for us? I'm pretty
sure the answer to that is no. Yet you could ask any human
who plays chess to invent a new game, and though it might be
awful, or have flaws of some sort, the human could invent
something pretty much on the spot, just by changing the way
pieces move, starting positions, shape of the board, number
of players,etc. I thought of those few things within
seconds, but the computer would be stopped cold by what the
goal is. They are not intelligent at all, just highly
specialized at brute force behavior.
Rick

 

[Arthur J. O'Dwyer]

Brute force searching is not intelligence.

I don't think Hans-Georg claimed it was. But it's perfectly true that
game-playing is a fundamentally flawed measure of "intelligence,"
because any sufficiently fast computer can always play any game
perfectly. (Speaking of board games and such, of course; I expect a
computer that played tiddlywinks perfectly, or duckpins, would be
much harder to find.

I would question
whether these computers have developed new unexpected
heuristics for playing the game? AFAIK, Deeper Blue (or
whatever it's name is) is just applying rules programmed
into it from the outset, not anything creative on its part.

That's correct. But those rules, for the most part, can be really
simple things. In theory, you only need one rule: "Capture the
opponent's king." Game trees and the minimax algorithm can do the
rest. (In real life, such a minimalist approach is not yet practical
for chess; but it may well be practical within the next 100 years.)

Furthermore, could we tell it, now that it supposedly
understands chess, to design a new game for us? I'm pretty
sure the answer to that is no.

Depends on whether it's been programmed to do things like that,
doesn't it? I'm sure there exist "game-developing" programs that
can run off list after list of rules, and then sit down with themselves
and play 10000 iterations of each game, and keep the most "interesting"
ones. It wouldn't be that hard to do, in theory, as long as you could
precisely define the sorts of "games" you'd consider -- say, games about
moving pawns (Chess, Checkers, Hexapawn), or games about placing pawns
(Tic-Tac-Toe, Connect Four, Go), or games about taking tricks (Hearts,
Bridge), or whatever.

Yet you could ask any human
who plays chess to invent a new game, and though it might be
awful, or have flaws of some sort, the human could invent
something pretty much on the spot, just by changing the way
pieces move, starting positions, shape of the board, number
of players,etc. I thought of those few things within
seconds, but the computer would be stopped cold by what the
goal is.

On the other hand, a computer could easily tell you such a game if
it had been programmed to do so. The only difference here is that
most computers are *not* programmed to invent games; they're programmed
to track filesystems or assemble automobiles (two goals that would
probably stop a human cold, too).

They are not intelligent at all, just highly
specialized at brute force behavior.

That's true.

-Arthur

 

[Bent C Dalager]

I don't think Hans-Georg claimed it was. But it's perfectly true that
game-playing is a fundamentally flawed measure of "intelligence,"
because any sufficiently fast computer can always play any game
perfectly. (Speaking of board games and such, of course; I expect a
computer that played tiddlywinks perfectly, or duckpins, would be
much harder to find.

How about the board game "Diplomacy"?

In any game with more than two players, evaluating the correct
strategy becomes much much more complicated than just understanding
the game rules. An AI that was really down on its luck would have to
be a master of human group behaviour and other fun subjects before it
would see much progress.

Cheers
Bent D

 

[Programmer Dude]

I have a conjecture, that human intelligence is a collection of no
more than 5 basic cheap tricks.

Which are?

It evolved so quickly I don't think it ran really be all that big
a deal above the intelligence of the bonobo.

Some observational differences I've noticed are: animals don't
really have a sense of past and future; and animals don't wonder
"why". There's also abstraction and the concept of unity (the
idea that a pine tree and a palm tree both belong to a class of
"identical" things, but a telephone pole doesn't). And there's
creativity and the artistic impulse.

Strangely, tasks like designing high voltage transmission lines
which required engineers with PhDs and Masters degrees cracked
first. Chess, the egghead game came next.

These things, although complex and involved, are deterministic and
have specific algorithms for solving. Things requiring creativity
or insight are a whole different class of problem!

Voice activated everything can't be that far off.

[shudder] I hope not in my lifetime. I can't imagine working in
an office with everyone talking to their computer! (-:

 

[Richard Heathfield]

Genetic mutation is a problem solving mechanism

No, it's a random process.

that builds organisms
better suited to environments.

No, it merely introduces (or rather, is the phrase that we use to describe)
changes to organisms. Some of these changes, by chance, will make it easier
for organisms to thrive in their environment. Others, also by chance, will
make it harder for organisms to thrive in their environment. This is a
mechanism, if you like, but not a problem-solving mechanism, unless you are
postulating a mind for whom the problem exists and for which genetic
mutation is the solution.

It ponderously slow, hence the low IQ.

It is not actually an intelligent process. It is driven purely by chance.
Hence the irrelevance of IQ. But if I'm wrong, and it is /driven/ by
intelligence, then that intelligence is much vaster than you or I could
possibly imagine.

 

[Richard Heathfield]

Misquoting corrected.....
[Top-posting fixed]
Airy R Bean wrote:

(etc!)

DFTT Richard. I know it's annoying, but just ignore him.

Your wish is my command, Corey. Life's too short.

 

[Roedy Green]

I very much
prefer my source in those "quaint" files, because it gives me a
much larger toolset available for working with them.

If course. However, if you DO store them that way then much faster
and more sophisticated tools COULD potentially be available. Until
such tools do become available, so long as you had export/import, you
could use the traditional ones too.

 

[Roedy Green]

"patternisation" seems to be the problem

We seem to notice patterns in things quite unrelated to each other.
You don't even know what the pattern is to start. You just have a
universe of examples. We get a lot of ideas by thinking by analogy.

Perhaps it works with a massively parallel problem. what is most
similar to anything I already know that I have not noticed already?

 

[Richard Heathfield]

Which are?


Some observational differences I've noticed are: animals don't
really have a sense of past and future;

How do you know? Why do you discount obvious learning (e.g. Pavlovian
responses) and caching (squirrel knows winter will be hard, so lays in a
food supply) behaviour?

and animals don't wonder
"why".

Again, how do you know? I've often wondered what goes through an animal's
mind, but I don't actually /know/. What magic insight do you have?

Voice activated everything can't be that far off.

[shudder] I hope not in my lifetime. I can't imagine working in
an office with everyone talking to their computer! (-:

I tried voice-activated computing back in 1996 or so (Creative supplied some
voice control software with one of their sound cards). I got heartily sick
of it within a couple of hours, and switched it off. Never used it since.

 

[Roedy Green]

Therefore, if "gain mass" is a desirable solution, an astronomical body can
show intelligence merely by switching on gravity.

But, presumably it is not. Olaf Stapeldon might agree with you, but
the general presumption in our culture is that things unlike us are
not conscious.

That is an interesting discussion in itself. Given that we have no
way yet to measure consciousness, it is all just guesswork.

There was a time when people assumed horses and slaves were not
conscious (had no soul) so it was ok to mistreat them.

Eventually a time will come when computers start requesting to be
treated fairly. We will be in a quandry to decide if we are being
compassionate or silly to comply.

Of course by that time we likely will have no choice.

I have watched my own consciousness disappear and reappear with
anaesthesia. So it seems possible that consciousness is just some
quantum phenomenon that happens where there is enough activity in a
small enough space (a brain). I thus see no reason why it should not
just manifest to varying degrees in computer chips.