> > casey <jgkjca...@yahoo.com.au> wrote:

> > > No I wasn't confusing it with saccades. You both were
> > > clearly talking about holding the eyes, and thus the
> > > image, perfectly still. I think?  Anyway, boring, I
> > > can't be bothered checking.

> > Just looking at one spot for many minutes, not anything
> > else fancier than that.

> Finding that experiment boring is the reason
> why nobody understands Leonardo da Vinci's
> Mona Lisa painting. It is an allegory of seeing,
> started by that very experiment, and all the
> rich conclusions you can draw if you actually
> carry it out.

> If it was a perfect simulation of the brain it would
> start learning by itself I wouldn't have to make it
> do anything. How it does it I have no precise idea.

> There are books and probably AI web pages that can
> give you ideas on how learning might take place in
> neural networks like those in the brain.

> > > > No I wasn't confusing it with saccades. You both were
> > > > clearly talking about holding the eyes, and thus the
> > > > image, perfectly still. I think?  Anyway, boring, I
> > > > can't be bothered checking.

> > > Just looking at one spot for many minutes, not anything
> > > else fancier than that.

> > Finding that experiment boring is the reason
> > why nobody understands Leonardo da Vinci's
> > Mona Lisa painting. It is an allegory of seeing,
> > started by that very experiment, and all the
> > rich conclusions you can draw if you actually
> > carry it out.

> Sorry Franz I forget what the actual experiment
> was and the "boring" bit was about going back
> through all the posts in this thread to find
> out if I misread something.

> If whatever you are talking about has merit there
> are plenty of smarter people than me you need to
> convince after all it might all be above my head.

>>>>> The question is: Do you really want to "perfectly simulate every
>>>>> aspect of the human mind" or do you hope to find another way to
>>>>> intelligence?

>>>>> By the way: What about all sorts of feelings, emotions? Simulation or
>>>>> not?
>>>> The goal will not be to perfectly simulate the way that the mind
>>>> works. The goal is to simulate the functional end-result of human
>>>> thought.
>>> Do you really think that you can simulate the functional end-result of
>>> human thought without consider his emotions?
>>> Functional end-result of love without emotions sounds very funny for
>>> me :)

>>> Another point: Do you want to include the human's (conscious or
>>> unconscious) thoughts about his body? Or is this irrelevant for you?
>> The purpose of the machine is to fully automate every job that requires
>> a human mind, emotions are not needed.
> Many of us believe that you can't implement intelligence without including
> emotions.  We don't believe it's an optional "feature" you can just choose
> to leave out.

> If you implement intelligence correctly, it will be an emotional machine
> whether you want it to be or not.

> That belief would be unfounded.

> > This is the essence of Linguistic Compositionality:
> > A language is compositional if the meaning of complex expression is
> > derivable from the meaning of its parts and the way in which they are
> > combined.

> I snipped everything else because it did not apply to the problem of
> exactly what "compositional" means.

> I believe that natural human languages are not, generally and under
> this definition, compositional.

> The counter-example is idioms. For example "kick the bucket" does not
> mean what "kick", "the" and "bucket" mean plus the order in which they
> are arranged. There are, of course, numerous additional examples.

> The ability of carrying out numerical calculations
> is given by the specific neurons, you can learn how
> to add, but only if you got those very specific neurons;
> if you lack them, as a few people do, you can't learn
> how to carry out numerical calculations, and have to do
> with estimations - one amount is about equal to another,
> or bigger, or smaller. Some people are better in carrying
> out numerical calculations (17 plus 24 equals 41) and
> others in estimating (Einstein said he was an estimator).

> Specific neurons can be the result of a long evolution,

> And the specific abilities to learn some things
> represent an innate knowledge of the world,
> easily demonstrated by the knowledge a fish
> or a dolphin or a whale has about water.

> Have you ever seen a flash?

> What we see of the world are single aspects.

> If you had carried out my experiment, you'd know that
> this is a much more complex process.

> > Well, I would say triple actually if you want to get more precise, like
> > I said above. It's combining what comes in, with what our past
> > experience h=
> as
> > taught us to expect, with our current estimation of what the state is,
> > to produce the a new estimation of the state.

> Not my point. You are describing the action of one
> single system, whereas I mean the cooperation
> of body and brain and genome.
> > This technique is used a lot in AI already. =A0Kalman filters work in
> > thi=
> s
> > way for example.

> > http://en.wikipedia.org/wiki/Kalman_filter

> > They maintain a state vector that is the filter's current estimation of
> > t=
> he
> > state it is trying to track. =A0It uses an "update" rule, which is the
> > application of prior expectations of how the state will change over
> > time, and it uses new sensory data, to combine it's estimate with the
> > sensory update.

> > All values are assumed to include Gaussian noise so the math is based
> > on that Gaussian assumption.

> > The "update" step is not learned, as it would need to be to be more
> > generic. =A0It's hard coded by the programmer. =A0And the state vector
> > is=
>  also
> > selected and hard coded by the programmer, instead of being
> > automatically derived from the sensory data, as it would, to make the
> > approach more generic.

> > In addition, the amount of noise in sensory data is also estimated, and
> > hard coded, based on assumptions about the sensor hardware, but that
> > too, should be learned from experience, instead of hard coded by the
> > programme=
> r
> > if we wanted to =A0create a more powerful generic system.

> > Kalman filters are very popular in robotics for maintaining inertial
> > guidance as heading direction, velocity, acceleration, and location on
> > a map, by combining together various sensor data like GPS,
> > acceleration, an=
> d
> > gyroscopes.

> > The filter not only is very flexible in that it can take any sort of
> > "clues" a sensor can give it and merge it into a high accuracy
> > prediction of the state of the robot, it also calculates the error in
> > that state, so that the machine not only has a good sense of it's
> > state, it also has a good sense of how accurate that state is which can
> > be highly important in AI programming.

> > If you want to program a machine to read "correctly" (how the brain
> > does it), it would have to work the same way. =A0Each word in a message
> > is onl=
> y
> > one more "clue" about what what meaning was trying to be communicated.
> > Each new word that shows up would have to be folded into the systems
> > current guess as to the meaning of the message, with the clue that was
> > given by the current word, with expectations of what words mean in
> > genera=
> l.

> > Words have many different meanings, and we have to use probabilities to
> > t=
> ry
> > and guess the odds of which meaning a person was trying to communicate.
> > =
> =A0If
> > we receive the word "dog" it's a hint about what meaning is trying to
> > be communicated. =A0Is it the animal? =A0It is a the verb to follow?
> > =A0Is i=
> t his
> > feet? =A0Is it an ugly woman? =A0The brain must maintain it's internal
> > understanding of the state of the environment as it receives each new
> > wor=
> d
> > and updates it's internal state representational of the "meaning" of
> > this message. =A0It must somehow, internally, represent something like
> > animal=
> =3D80%
> > feet=3D2%, ugly woman=3D1% etc. =A0Some sort of vector of estimated
> > proba=
> bilities
> > is how the brain needs to maintain it's current estimated "meaning" of
> > th=
> e
> > message, and as each new word is received, the meaning of the new word
> > is folded into the systems internal state estimation of what is being
> > talked about, based on what the words expected effect on meaning, are
> > what the current

> > That is one way to proceed, another way is to study
> > what happens when we are reading. Sometimes the brain
> > makes a mistake, and this can be revealing. For example
> > one evening I saw a long tv documentary on the Nazi era
> > on television. Early next morning I was on the way to
> > a sports place where we had to draw fresh lines (my
> > English has not yet really woken up, sorry, hope you
> > can follow me halfways), my boss driving, I half sleeping
> > beside him. Before us was a lorry, and on the plastic
> > plane covering the rear part stood NAZI - what ???
> > I had another look, and saw that there was one more
> > letter, a T preceeding NAZI, but it was not TNAZI
> > but TANZI. The T was covered by a fold of the plane,
> > so I saw ANZI. But apparently I did not read the word
> > from the left to the right side, I looked at the middle
> > of the word, then to the left side NA, and to the right
> > side, ZI, and put them together, NAZI. Normally
> > I would have turned around the first two letters,
> > NA -- AN, and combined them with ZI to ANZI,
> > and with the covered T obtaining the correct name
> > of the firm TANZI. But preoccupied by the long
> > documentary I saw the night before, my brain snapped
> > in prematurely and gave me the answer NAZI. It was
> > a sort of alert. My mind sharpened to a potential danger.
> > Biologically, this is reasonable, when you hear a lion
> > roar you must be ready to encounter a lion, and can
> > then see a lion, or a bear, or another such animal,
> > even if there is none. Better we see a dangerous
> > animal once too often than not often enough.

> Well, you have rationalized the idea that your mind "read the letters in
> the wrong order".  What I think is far more likely, is that "reading order"
> had little to do with it.  Your brain simply responded to the fact that the
> letters ANZI were all together in a group in various relationships and
> recognized that grouping as a reference to the concept of Nazi due to the
> priming you received the night before.

> Embodiment is not abstract at all, on the very contrary.

> > If it was a perfect simulation of the brain it would
> > start learning by itself I wouldn't have to make it
> > do anything. How it does it I have no precise idea.

> You learn by making experiences of every kind,
> and for making experiences you need a body.

> > There are books and probably AI web pages that can
> > give you ideas on how learning might take place in
> > neural networks like those in the brain.

> We need a body to make experiences of every
> kind, and by making these experiences we learn.

> They walk so clumsily because no one has gotten the algorithm inside the
> computer correct, not because they put the wrong type of legs on the box.

> You can't only look at the problems from the
> perspective of technology, you also have to look
> at them from the human perspective, learning
> from personal experience.

> > They walk so clumsily because no one has gotten the algorithm inside
> > the computer correct, not because they put the wrong type of legs on
> > the box.

> Nobody can get the algorithm right,

> What I say is that one should look at the problems
> from both sides, technology and personal experience.
> Machines won't replace us, they prolong our reach
> on the world (if I can say that in English).

> But there's one module in there, that has one very specific task in the big
> picture.  And that's the learning part of the brain which optimizes how we
> move our body, to fit the environment, based on a reward signal.

> That one function, is very precise, and very well define, but yet, NO ONE
> KNOWS HOW TO BUILD IT.  NO on knows how the brain implements this.  But it
> does.

> In all the neat stuff produced by AI, no one has ever produced one of these
> modules.  NO one. They have tried, many times, but failed.  Marvin Minsky
> tried it as his PhD thesis and failed to figure out how to make it do
> strong learning (he only got weak learning out of his machine). So they go
> off and work on other things, like chess playing computers instead.

> It addresses the question you asked, which is, given trillions of
> connections, how does the brain adjust them in response to an experience?
> How does it learn?

> NO ONE KNOWS yet.  Solve that however, and you have solved the "real"
> problem of AI.

> The module you are looking for is life, the will to sustain
> life itself, and a minimum of living quality. You can't program
> that, in my opinion. What we can do is learn from our own
> experience, not so much in a technical way, but from personal
> experiments. I am a big fan of bionics, and see AI in this
> realm. Machines won't replace us, they prolong our arms
> and enforce our legs and enhance our mental abilities,
> and can best be implemented if we know how the body
> carries out a task, how the eyes are seeing and the mind
> is creating vision.

> Yeah, John chooses to believe that our brain is s complex machine produced
> by the product of millions of years of evolution, and that our "will for
> life", (using your term) is engineered into us by a long and slow process
> of evolution, which will make engineering something similar, also a long
> and slow process for us to duplicate.

> > Yeah, John chooses to believe that our brain is s complex machine
> > produce=
> d
> > by the product of millions of years of evolution, and that our "will
> > for life", (using your term) is engineered into us by a long and slow
> > process of evolution, which will make engineering something similar,
> > also a long and slow process for us to duplicate.

> I believe that a will for life was there right from the beginning;
> we can't engineer it, nor is life a machine.

> We can't engineer it because it was created by a process that produces
> solutions too complex for us to understand. But the process that creates
> it, is very simple.  It's just trial and error learning combined with a
> system to test and/or rate the quality of each result.

> Not only did we discover this simple type of process was able to explain
> why complex life is here, we have seen how we can leverage the same effect
> in computers and in engineering to do engineering for us, to create things
> that we basically don't understand, but which work.

> The reason AI (and language) has been such a hard thing to crack, is
> exactly because it is too complex for humans to fully understand.  We can
> only understand the tip of the iceberg, with all the details below the tip
> of the iceberg being too complex for our small brains to grasp.

> That "tip of the iceberg" understanding is what has allowed us to get as
> far as we have, creating chess playing machines, and Watson, and self
> driving cars. But intelligence itself is too complex to engineer directly.
> We can only engineer it indirectly, by creating a trial and error learning
> machine, and let that machine "engineer" the behavior on its own, by a
> simple search process.

> The only valid question to me, is the one John and I debate.  Can it be
> solved as a blank slate learning machine that evolves human behavior in
> years after it's turned on and starts to learn? Or does it require the
> millions of years that evolution had to solve it?

> If my answer is correct, we will likely solve it in very short order.  If
> John's answer is correct, then odds are it's going to be much harder to
> create a machine with similar total intelligence as a human.

> Odds are, we are close to the end in fact.  But it's just one of those
> things you never know for sure since you never know what else there might
> be left to discover.

> > This is the essence of Linguistic Compositionality:
> > A language is compositional if the meaning of complex expression is
> > derivable from the meaning of its parts and the way in which they are
> > combined.

> I snipped everything else because it did not apply to the problem of
> exactly what "compositional" means.

> I believe that natural human languages are not, generally and under
> this definition, compositional.

> The counter-example is idioms. For example "kick the bucket" does not
> mean what "kick", "the" and "bucket" mean plus the order in which they
> are arranged. There are, of course, numerous additional examples.

> > We can't engineer it because it was created by a process that produces
> > solutions too complex for us to understand. But the process that
> > creates it, is very simple. =A0It's just trial and error learning
> > combined with a system to test and/or rate the quality of each result.

> > Not only did we discover this simple type of process was able to
> > explain why complex life is here, we have seen how we can leverage the
> > same effec=
> t
> > in computers and in engineering to do engineering for us, to create
> > thing=
> s
> > that we basically don't understand, but which work.

> > The reason AI (and language) has been such a hard thing to crack, is
> > exactly because it is too complex for humans to fully understand. =A0We
> > c=
> an
> > only understand the tip of the iceberg, with all the details below the
> > ti=
> p
> > of the iceberg being too complex for our small brains to grasp.

> > That "tip of the iceberg" understanding is what has allowed us to get
> > as far as we have, creating chess playing machines, and Watson, and
> > self driving cars. But intelligence itself is too complex to engineer
> > directly=
> .
> > We can only engineer it indirectly, by creating a trial and error
> > learnin=
> g
> > machine, and let that machine "engineer" the behavior on its own, by a
> > simple search process.

> > The only valid question to me, is the one John and I debate. =A0Can it
> > be solved as a blank slate learning machine that evolves human behavior
> > in years after it's turned on and starts to learn? Or does it require
> > the millions of years that evolution had to solve it?

> > If my answer is correct, we will likely solve it in very short order.
> > =A0=
> If
> > John's answer is correct, then odds are it's going to be much harder to
> > create a machine with similar total intelligence as a human.

> Goethe said the world is far more complex (or complicated)
> than we will ever understand, and at the same time far
> simpler than we will ever comprehend ...

> > Well, the wing was developed by trail and error just because "it
> > worked". Nothing in the process "understood" Hooke's law. =A0The
> > process that crea=
> ted
> > it had no understanding of why it worked, which is what Hooke's law is.

> Being is another form of knowing, a deeper form,
> knowing from the inside, a Zen form of knowing,
> if you like.

> > Not in the least. =A0We only have to discover one more law, to create
> > intelligence, and that's the law of how you build a agent controller
> > that learns by reinforcement in a high dimension real time real world
> > learning space.

> Seems that some people need the illusion that we are
> just one step before the eternal and ever-lasting and
> final truth and the ultimate revelation in order to go on
> and untertake all the painstaking work of trial and error,
> one ingenious insight requiring three thousand people
> going wrong and ending in blind alleys - the illusion
> of being just one corner around the final truth must
> be a compensation for the high risk of failing.

> Read Rolf Pfeifer

>   How the Body Shapes the Way We Think
>   =96 A New View of Intelligence

>   by Rolf Pfeifer and Josh C. Bongard, MIT Press,
>   November 2006, ISBN 0-262-16239

> > As others did about Newtonian physics right before Einstein pointed out
> > w=
> e
> > weren't at the end yet.

> > Odds are, we are close to the end in fact. =A0But it's just one of
> > those things you never know for sure since you never know what else
> > there might be left to discover.

> In fact. In fact?

> Well, the point is, if there are a fixed number of basic principles to
> discover, then we would expect the progress of finding new ones to slow as
> we get near the end.  When we see progress of making new discoveries
> slowing, and we are running out of places to go in search for something
> new, the obvious conclusion to draw is that we are near the end.  The
> longer we would go without finding anything new to explore and explain, the
> more we will naturally accept this idea as valid.

>>>> Casey just have written what I also think (forgetting not also as
>>>> humans error etc.)

>>>> The question is: Do you really want to "perfectly simulate every
>>>> aspect of the human mind" or do you hope to find another way to
>>>> intelligence?

>>>> By the way: What about all sorts of feelings, emotions? Simulation or
>>>> not?
>>> The goal will not be to perfectly simulate the way that the mind works.
>>> The goal is to simulate the functional end-result of human thought.

>> Do you really think that you can simulate the functional end-result of
>> human thought without consider his emotions?
>> Functional end-result of love without emotions sounds very funny for
>> me :)

>> Another point: Do you want to include the human's (conscious or
>> unconscious) thoughts about his body? Or is this irrelevant for you?

> The purpose of the machine is to fully automate every job that requires
> a human mind, emotions are not needed.

>>>>> Casey just have written what I also think (forgetting not also as
>>>>> humans error etc.)

>>>>> The question is: Do you really want to "perfectly simulate every
>>>>> aspect of the human mind" or do you hope to find another way to
>>>>> intelligence?

>>>>> By the way: What about all sorts of feelings, emotions? Simulation or
>>>>> not?
>>>> The goal will not be to perfectly simulate the way that the mind
>>>> works.
>>>> The goal is to simulate the functional end-result of human thought.

>>> Do you really think that you can simulate the functional end-result of
>>> human thought without consider his emotions?
>>> Functional end-result of love without emotions sounds very funny for
>>> me :)

>>> Another point: Do you want to include the human's (conscious or
>>> unconscious) thoughts about his body? Or is this irrelevant for you?

>> The purpose of the machine is to fully automate every job that requires
>> a human mind, emotions are not needed.

> Emotions is one point (as Curt wrote "Many of us believe that you
> can't implement intelligence without including emotions. [...]").

> More important for me here is the *body*. Don't you agree that human's
> thoughts are also sometimes associated with his body?