Artificial Intelligence

java-programmer, Artificial Intelligence Roedy Green <email***@***.com> wrote in
news:email***@***.com:

> On Wed, 21 Sep 2005 13:11:14 GMT, "JS" <james.sarjeant90
@ntlworld.com>
> wrote or quoted :
>
>>But we can fake Intelligence by using hard coded if statements,

[snip]

> I suspect living organisms are the same way. These high level
> attributes we use to describe humans and their thoughts are just
> patterns of low level flow. There is nothing in there directly that
> controls them on that level.

Exactly, there is nothing fake about hard coded if statements, unless
you want to call everything that is manmade fake.

The problem with hard coded if statements as an implementation of
intelligence isn't fakeness, it is quality. Unless there are a huge
number of if statements, you're going to notice pretty quick that this
intelligence isn't thinking very deeply on a Turing test. It is
practically impossible to make a good intelligence that way. You would
probably need billions or trillions of 'if's to make something like a
human. At the very least you would need to use a computer to write that
program for you, which is totally violating the spirit of hard coded if
statements.

java-programmer >> Artificial Intelligence Roedy Green <email***@***.com> wrote in
news:email***@***.com:

> On Wed, 21 Sep 2005 13:11:14 GMT, "JS" <james.sarjeant90
@ntlworld.com>
> wrote or quoted :
>
>>But we can fake Intelligence by using hard coded if statements,

[snip]

> I suspect living organisms are the same way. These high level
> attributes we use to describe humans and their thoughts are just
> patterns of low level flow. There is nothing in there directly that
> controls them on that level.

Exactly, there is nothing fake about hard coded if statements, unless
you want to call everything that is manmade fake.

The problem with hard coded if statements as an implementation of
intelligence isn't fakeness, it is quality. Unless there are a huge
number of if statements, you're going to notice pretty quick that this
intelligence isn't thinking very deeply on a Turing test. It is
practically impossible to make a good intelligence that way. You would
probably need billions or trillions of 'if's to make something like a
human. At the very least you would need to use a computer to write that
program for you, which is totally violating the spirit of hard coded if
statements.

java-programmer >> Artificial Intelligence On Thu, 29 Sep 2005 08:22:23 GMT, Brendan Guild <email***@***.com> wrote
or quoted :

> 'if's to make something like a
>human.

unless by human you mean "Jerry Springer guest".
--
Canadian Mind Products, Roedy Green.
http://mindprod.com Again taking new Java programming contracts.

java-programmer >> Artificial Intelligence Roedy Green wrote:
> On Thu, 29 Sep 2005 08:22:23 GMT, Brendan Guild <email***@***.com> wrote
> or quoted :
>
>
>>'if's to make something like a
>>human.
>
>
> unless by human you mean "Jerry Springer guest".

I don't think a construct of two or three if- or switch-case -statements
would qualify as a "human".

--
-Aki "Sus" Laukkanen

java-programmer >> Artificial Intelligence "Roedy Green" <email***@***.com> wrote in
message news:email***@***.com...
>
> People with a
> straight face tell me fish and turtles are not conscious and cannot
> feel pain even though the behave quite similarly to mammals when
> skewered. How the hell could they know one way or the other?

It depends on your definition of pain. If you define pain as "an
unpleasant sensation felt by humans", then of course, no other animal can
feel "pain" under this definition.

There's a "widely accepted" definition of pain which basically says
vertebrates can feel "pain", but invertebrates can't. It has to do with
nociceptors, which are free nerve endings of neurons that have their cell
bodies outside of the spinal column. It is believed that the sensation of
pain only occurs when these nociceptors are excited and send signals up the
spine. There are no nociceptors in the brain itself, to merely having a
brain is "not sufficient" to feel "pain". Since invertebrates don't have
nociceptors, they are believed to not feel "pain".

As for an argument against behaving-similarly-when-skewered, a recently
dead human body may also behave similarly when skewered via muscle
contractions from the stimuli, but most people believe that dead humans
don't feel pain (of course, how can we know for sure?)

- Oliver

java-programmer >> Artificial Intelligence
"Brendan Guild" <email***@***.com> wrote in message
news:email***@***.com...
> Roedy Green <email***@***.com> wrote in
> news:email***@***.com:
>
>> On Wed, 21 Sep 2005 13:11:14 GMT, "JS" <james.sarjeant90
> @ntlworld.com>
>> wrote or quoted :
>>
>>>But we can fake Intelligence by using hard coded if statements,
>
> [snip]
>
>> I suspect living organisms are the same way. These high level
>> attributes we use to describe humans and their thoughts are just
>> patterns of low level flow. There is nothing in there directly that
>> controls them on that level.
>
> Exactly, there is nothing fake about hard coded if statements, unless
> you want to call everything that is manmade fake.
>
> The problem with hard coded if statements as an implementation of
> intelligence isn't fakeness, it is quality. Unless there are a huge
> number of if statements, you're going to notice pretty quick that this
> intelligence isn't thinking very deeply on a Turing test. It is
> practically impossible to make a good intelligence that way. You would
> probably need billions or trillions of 'if's to make something like a
> human. At the very least you would need to use a computer to write that
> program for you, which is totally violating the spirit of hard coded if
> statements.

The more major problem with hardcoded if statements, I think, is that
they are not capable of learning. The human mind has memory, and is
self-modifying (i.e. new neural connections can be made, and old ones can be
disconnected). With a finite number of if statements, you could "fake"
learning for any finite number of years (80 years? 800 years? 8 trillion
years?), but with an actual neural network, in principle, it could go on
learning forever, and eventually differ from its "equivalent" if-statement
construct.

It's like the difference of expressive power between a finite state
automata and an infinite-state automata.

- Oliver

java-programmer >> Artificial Intelligence On Thu, 29 Sep 2005 14:25:14 GMT, "Oliver Wong" <email***@***.com>
wrote or quoted :

> As for an argument against behaving-similarly-when-skewered, a recently
>dead human body may also behave similarly when skewered via muscle
>contractions from the stimuli, but most people believe that dead humans
>don't feel pain (of course, how can we know for sure?)

That is as similar argument to the ones that define humans as the only
intelligent species by positing human intelligence as the only valid
kind.

From an Darwinian point of view, humans are way down there. We will be
one of the shortest lived experiments. Even a chimp knows better than
to shit in its nest. We take huge risks daily with survival of the
entire species.

My great hope is that AI will be more sensible than humans are, and
will tame us before we kill ourselves off. Once AI reaches the point
it starts controlling its own evolution it will leave us in the dust
literally overnight. We will awake to a whole new world, hopefully
free of XML and JavaScript.

--
Canadian Mind Products, Roedy Green.
http://mindprod.com Again taking new Java programming contracts.

java-programmer >> Artificial Intelligence
"Roedy Green" <email***@***.com> wrote in
message news:email***@***.com...
> On Thu, 29 Sep 2005 14:25:14 GMT, "Oliver Wong" <email***@***.com>
> wrote or quoted :
>
>> As for an argument against behaving-similarly-when-skewered, a
>> recently
>>dead human body may also behave similarly when skewered via muscle
>>contractions from the stimuli, but most people believe that dead humans
>>don't feel pain (of course, how can we know for sure?)
>
> That is as similar argument to the ones that define humans as the only
> intelligent species by positing human intelligence as the only valid
> kind.

I don't see the similarities. The argument I gave can be abstracted to:

* It is widely believed that recently deceased humans do not feel pain.
* Recently deceased humans also react about when skewered (this being
explained as a responce of muscle tissue to stimuli)
* Therefore, if a given entity reacts when skewered, it is not enough to
prove that that entity is feeling pain.

An alternative argument which is similar to the above one:

* It is widely believed that rocks and other minerals do not feel pain.
* It is possible to construct a object using only rocks and other
minerals such that when it is skewered, it thrashes about and makes loud
noises, and then slowly stops, gradually losing energy.
* Therefore, if a given object thrashes about and makes loud noises when
skewered, and gradually stops, that is not sufficient to prove that said
object felt pain.

I don't see how these arguments resemble "proof by definition", which is
what it sounds like you are comparing them to. They're more like "proof by
counter-example" (first one) or "proof by construction" (second one).

- Oliver

java-programmer >> Artificial Intelligence On Thu, 29 Sep 2005 21:20:24 GMT, "Oliver Wong" <email***@***.com>
wrote or quoted :

>I don't see the similarities.

Here is another way to approach what I am trying to say. I understand
the eye has independently evolved at least 8 different times in
evolution.

If you study the eye's fine structure and mechanisms in just one
species, you can't necessarily extrapolate your findings to all the
others.

In a similar way I suspect "pain" could have evolved multiple times in
evolution. There may be a great many different mechanisms that all
have the same basic purpose. So again a discovery about how pain
works in only one species should not be blanket extrapolated to all
other species, especially when doing so seems primarily to justify
cruelty and exploitation.

There is a penalty to be paid for being too kind to a rock and a
penalty to be paid for being too cruel to a whale. I am far more
concerned about making the second error than the first.

Baumhertzig
There was a German Christian who lived in the 1900s in Stadtsteinach
in Germany named Gottfried M黮ler. Some called him a saint. He
arranged homes for orphans around the world. He suggested the key to
transforming earth from hell to heaven lay in tiny acts of compassion,
for the lowest of the low, e.g. even earthworms, done in secret.
揧ou must behave as if your every act, even the smallest, impacted a
thousand people for a hundred generations. Because it does.?
~ Gottfried M黮ler

--
Canadian Mind Products, Roedy Green.
http://mindprod.com Again taking new Java programming contracts.

java-programmer >> Artificial Intelligence You might want to take a look at the WEKA project:
http://www.cs.waikato.ac.nz/~ml/

It has, among other machine learning algorithms, a neural network
implementation and it's in Java.

java-programmer >> Artificial Intelligence > My next challenge is where to start.

Here is a great place to start:
http://www.gamasutra.com/features/19970801/pathfinding.htm

You need to register, but it's free.

Another great URL is:
http://www.policyalmanac.org/games/aStarTutorial.htm

This handles the A* algorithm, which is truly the star of path-finding
algorithms. But if you're new to this, you'll probably want to tangle
with simpler path-finding algorithms first, which the first URL offers.

Then, when path-finding is second nature to you, head over to:
http://www.cs.waikato.ac.nz/ml/weka/

which offers all sorts of machine learning algorithms, in Java
(including neural nets).

Kind regards, Jonck

java-programmer >> Artificial Intelligence
"Roedy Green" <email***@***.com> wrote in
message news:email***@***.com...
> On Tue, 20 Sep 2005 22:13:42 GMT, "Oliver Wong" <email***@***.com>
> wrote or quoted :
>
> > I believe a fruit fly has about 250,000 neurons, and even this will
be
> >quite a struggle to simulate.
>
> I met a guy named Bernie Till who did his master's thesis back in the
> early 90s doing a simulation of a nematode -- a tiny white worm. It
> had something like 149 neurons. What blew me away is the simulation
> was able to demonstrate all known activities of the nematode from
> finding mates to food to avoiding chemicals, to the way it moved. This
> is amazingly compact coding -- all this with a mere 149 neurons!
> He said the secret was in the S function each neuron implements.
> --
> Canadian Mind Products, Roedy Green.
> http://mindprod.com Again taking new Java programming contracts.

What was in the S function?

java-programmer >> Artificial Intelligence <email***@***.com> wrote in message
news:email***@***.com...
>> My next challenge is where to start.
>
> Another great URL is:
> http://www.policyalmanac.org/games/aStarTutorial.htm
>
> This handles the A* algorithm, which is truly the star of path-finding
> algorithms. But if you're new to this, you'll probably want to tangle
> with simpler path-finding algorithms first, which the first URL offers.

I actually wanted the OP to try and solve the problem him/herself first
before introduce him/her to A*. For the benefit of those who don't know, A*
is considered the "best" general purpose path finding algorithm; so much so
that very little further research is done in path finding after A*'s
discovery.

- Oliver

java-programmer >> Artificial Intelligence On Thu, 22 Sep 2005 09:06:36 GMT, "JS" <email***@***.com>
wrote or quoted :

>
>So to solve the problem in computing terms we need to write a computer which
>has a few hard coded rules to get it started.

Look how we train humans. Mom first tells you "never go in the
street". Or we tell newbie coders "never use public members".

When the student eventually feels confident he understand the REASONS
for the rule of thumb, then he is in a position to discover instances
where it might make sense to violate it.
--
Canadian Mind Products, Roedy Green.
http://mindprod.com Again taking new Java programming contracts.

java-programmer >> Artificial Intelligence On Wed, 05 Oct 2005 08:00:21 GMT, "JS" <email***@***.com>
wrote or quoted :

>What was in the S function?

The paper was over my head. But I do recall him saying that the
secret of the nematode's intelligence is not the wiring of the
neurons, but the parameters of the S functions that control their
firing.

I remember the diagrams of them looking like S curves you get in
titration.

Think about it. Imagine how much Java code you would have to write to
simulate ALL the behaviours of a small worm. Yet the worm does it
with 140 odd neurons, a hard-wired set of connections, and
parameters to some S functions for the neurons.

Bernie's simulated S-function worm mimics the worm and exhibited all
the behaviours. That demonstrates the power of worm programming skill.

It also suggests that we are playing with only a small subset of
reasonable coding techniques.
--
Canadian Mind Products, Roedy Green.
http://mindprod.com Again taking new Java programming contracts.

java-programmer >> Artificial Intelligence
"Roedy Green" <email***@***.com> wrote in
message news:email***@***.com...
> On Wed, 05 Oct 2005 08:00:21 GMT, "JS" <email***@***.com>
> wrote or quoted :
>
>>What was in the S function?
>
> The paper was over my head. But I do recall him saying that the
> secret of the nematode's intelligence is not the wiring of the
> neurons, but the parameters of the S functions that control their
> firing.
>
> I remember the diagrams of them looking like S curves you get in
> titration.

It just occured to me that what your friend calls an "S function" might
be what I call a "sigmoid function". As can be seen in this random diagram I
found in google, the graph of a sigmoid function does vaguely look like an
S: http://www.gisdevelopment.net/aars/acrs/2000/ts9/images/imgp0001a.jpg

If my assumption that S refers to sigmoid is indeed correct, then when
the friend said that the "secret was in the S function", what they meant was
that the "secret" was the fact that an S function was being used at all.
It's a "secret" in the sense that people who haven't studied much AI are
astonished when they find out how simple neural networks are.

The neurons in a neural net can essentially be connected randomly; it
really doesn't matter. As long as each neuron implements a sigmoid function
with some sort of feedback so that it can learn to adjust the the parameters
to the function, the neuron by itself can learn some surprisingly
sophisticated relationships between input and output. Put 149 of those
neurons together, and you get something about as smart as a worm. It's
really that simple.

- Oliver

java-programmer >> Artificial Intelligence "Roedy Green" <email***@***.com> wrote in
message news:email***@***.com...
> On Thu, 22 Sep 2005 09:06:36 GMT, "JS" <email***@***.com>
> wrote or quoted :
>
> >
> >So to solve the problem in computing terms we need to write a computer
which
> >has a few hard coded rules to get it started.
>
> Look how we train humans. Mom first tells you "never go in the
> street". Or we tell newbie coders "never use public members".
>
> When the student eventually feels confident he understand the REASONS
> for the rule of thumb, then he is in a position to discover instances
> where it might make sense to violate it.
> --
> Canadian Mind Products, Roedy Green.
> http://mindprod.com Again taking new Java programming contracts.

So are you saying that writing hard coded if statments isnt a good way to go
about it unless we can decide which situations require which actions?because
if we only have one statement that was written when we were told not to go
into the street then we would never cross the road so we have to have two
actions and decide which one is sensible to use in that situation

java-programmer >> Artificial Intelligence On Thu, 06 Oct 2005 15:54:26 GMT, "Oliver Wong" <email***@***.com>
wrote or quoted :

>The neurons in a neural net can essentially be connected randomly; it
>really doesn't matter. As long as each neuron implements a sigmoid function
>with some sort of feedback so that it can learn to adjust the the parameters
>to the function, the neuron by itself can learn some surprisingly
>sophisticated relationships between input and output. Put 149 of those
>neurons together, and you get something about as smart as a worm. It's
>really that simple.

That sounds like what he said. He might even have used the term
sigmoid. I just remember seeing the graphs. This was back in the
early 90s.
--
Canadian Mind Products, Roedy Green.
http://mindprod.com Again taking new Java programming contracts.

java-programmer >> Artificial Intelligence On Thu, 06 Oct 2005 18:29:35 GMT, "JS" <email***@***.com>
wrote or quoted :

>So are you saying that writing hard coded if statments isnt a good way to go
>about it unless we can decide which situations require which actions?because
>if we only have one statement that was written when we were told not to go
>into the street then we would never cross the road so we have to have two
>actions and decide which one is sensible to use in that situation

All I am saying is if statement style rules of thumb can be used
without any understanding WHY you should follow the rules. As a child
you don't need to have a traffic model to apply them. Later, once you
understand acceleration, mass, traffic, drunk drivers etc etc. you can
dispense with the rules of thumb and use your own internal modeling
which will sometimes tell you it is safe when the rules of thumb
predict danger and vice versa.
--
Canadian Mind Products, Roedy Green.
http://mindprod.com Again taking new Java programming contracts.

Posted: 2005-10-24 17:01:00

java-programmer >> Artificial Intelligence Hi all.
Does anyone happen to have the thread about Artificial Intelligence saved
anywhere or can they tell me where I can get it because I was saving it for
a rainy day so to speak but last night I had to reinstall my OS and lost it.
It was the one that started with a simple question about Java and AIML and
ended up as a deep, almost philosophical, chat about AI and ways to
implement it etc.
Thanks in advance
JS

Posted: 2005-10-24 17:33:00

java-programmer >> Artificial Intelligence On Mon, 24 Oct 2005 09:01:29 GMT, "JS" <email***@***.com>
wrote, quoted or indirectly quoted someone who said :

>Does anyone happen to have the thread about Artificial Intelligence saved
>anywhere or can they tell me where I can get it because I was saving it for
>a rainy day so to speak but last night I had to reinstall my OS and lost it.
>It was the one that started with a simple question about Java and AIML and
>ended up as a deep, almost philosophical, chat about AI and ways to
>implement it etc.
>Thanks in advance

if you remember the group it was in or some keywords, you can go to
groups.google.ca which archives old postings.

group:comp.lang.java.help

selects the group.
--
Canadian Mind Products, Roedy Green.
http://mindprod.com Again taking new Java programming contracts.