| View previous topic :: View next topic |
| Author |
Message |
Curt Welch
Guest
|
 Posted: Wed Nov 19, 2008 9:29 am Post subject: Re: A Critique of AI research |
 |
|
turtoni <turtoni@fastmail.net> wrote:
| Quote: | I think we would be more successful in continuing to work at some
"good" questions and perhaps create as a by-product some kind of
artificial "intelligence". After all that would appear to be how
"intelligence" got started.
So for example; how do we create an *actual* unlimited supply of
renewable energy?
|
Well, that's a silly question. There's no such thing as "renewable
energy".
The best energy source around here is the sun, so the real energy problem
is simply a question of how do we best tap the energy flow from the sun.
Answering that gives as a few billion years of energy (not renewable).
| Quote: | I imagine that from these types of problems a form of artificial
intelligence will arise.
|
Well, the better question along that line is more like:
How do we build a machine that can survive on its own without human help?
People working on real world robotics questions are already trying to solve
this problem and their work is already pushing AI towards higher levels of
intelligence.
--
Curt Welch http://CurtWelch.Com/
curt@kcwc.com http://NewsReader.Com/ |
|
| |
|
 Back to top |
Curt Welch
Guest
|
| Posted: Wed Nov 19, 2008 9:29 am Post subject: Re: A Critique of Prof. Hubert Dreyfus' "Why Heideggerian AI |
|
|
"Isaac" <groups@sonic.net> wrote:
| Quote: | "Publius" <m.publius@nospam.comcast.net> wrote in message
news:Xns9B57F131CB7B0mpubliusnospamcomcas@69.16.185.250...
"Isaac" <groups@sonic.net> wrote in
news:491f9f87$0$33506$742ec2ed@news.sonic.net:
Minsky, unaware of Heidegger's critique, was convinced that
representing a few million facts about objects including their
functions, would solve what had come to be called the commonsense
knowledge problem. It seemed to me, however, that the deep problem
wasn't storing millions of facts; it was knowing which facts were
relevant in any given situation. One version of this relevance
problem was called "the frame problem." If the computer is running a
representation of the current state of the world and something in the
world changes, how does the program determine which of its represented
facts can be assumed to have stayed the same, and which would have to
be updated?
Dreyfus is pointing out one consequence of the lack of a useful
definition of "intelligence."
Actually, he is doing much more than that in his paper. I just posted a
portion of its background section, but his paper sets forth what he
believes is why AI fails and how he (and certain philosophers/researchers
he relies on) thinks intelligence works. I will email you the paper for
your reference. Let me know if you are interested in me posting my
critiques of his paper for your response. Maybe you will defend his
positions?
It is problem which plagues most programs for producing
AI (which is not to deny that much progress has been made in that
endeavor).
We may define "intelligence" as, "The capacity of a system to generate
solutions to novel problems," and "problems" as, "Obstacles or
impediments preventing the system from attaining a goal."
I defy you to contrive a definition of Intelligence that works. For
example, using your current definition above, the Earth would be
intelligent because it is a system with the capacity to generate
solutions (e.g., extremely complex, yet stable atmospheric weather, ocean
currents, etc.) to solve novel problems of, for example, maintaining a
stable global temperature in the face of many (thousands) changing
(novel) variables that are constant obstacles preventing the Earth (Gia?)
from attaining her goal of minimizing temperature differences globally.
|
The earth is intelligent. So is the universe as a whole.
Life looks like it was designed by intelligence because
it was. The process of evolution is just one more example of the many
intelligent processes at work in the universe. Evolution is an example of
a reinforcement learning process and I basically consider all reinforcement
learning processes to be examples of intelligence.
| Quote: | There are many similar examples that use your language but are not
considered to be intelligent to anyone reasonable in science. Care to
update your definition or defend it?
|
Many people in science have no clue what they are talking about when they
use the word "intelligence". As such, they define what is, and what isn't
intelligent based on total nonsense and ungrounded speculation - as I've
said before - without using any empirical evidence to argue from.
Of course that doesn't stop them, because they like to claim things such as
"subjective experience is outside the scope of empirical evidence".
And then they tell us what _their_ subjective experience is like and use
their beliefs about their own subjective experience to "prove" an endless
list of nonsense ideas about the universe.
The typical argument and thought path starts with the belief that human
consciousness is something that exists only in humans. Then from there,
they make the argument that since humans have this magical attribute called
consciousness and other things like the Earth doesn't, that intelligence
requires consciousness. But since they don't have any clue what creates
human consciousness, they also don't have any clue what creates
intelligence and don't really have any way to determine if the earth is
intelligent or not.
And when asked to explain what evidence they have to suggest this attribute
exists only in humans, they use the self serving argument that since they
"known" it exists in them, and that other humans are physically similar to
them, that this stuff they known exists in them must also exist in others.
But all that argument and the arguments that grow from it are based on a
belief that has no support. The belief that "consciousness" is something
other than simple brain function. That consciousness is not an identity
with physical brain function.
However, all the empirical evidence we have tells us that assumption is
wrong. And if we choose to believe what the empirical evidence shows us
(materialism) - then we know that there is nothing here to explain, other
than the physical signal processing that happens in the brain which
produces human behavior.
Once you grasp the significance of what the empirical evidence is telling
us, all the need of defining intelligence as some sort of link with "being
conscious" goes away. We are left with defining intelligence is some class
of signal processing algorithm that describes how the brain works. And
though there are multiple options there, none of them make intelligence
hard to understand. It's no harder to understand than any typical machine
learning algorithm for example.
I choose to use the fairly broad and generic definition of intelligence
being a reinforcement learning system which allows the concept to include
many processes other than just what the brain does - such as the process of
evolution.
You could easily restrict the definition to something closer to what he
brain does, which would be something more like a real time distributed
parallel signal processing network trained by reinforcement instead of the
far broader "all reinforcement learning processes" I like to use.
| Quote: | Introducing goals into the definition gives us a handle on the "frame
problem": the problem is framed by the current goal.
Goals are really related to the frame problem in that the "frame" that
matters is the one that reflect "reality" in the context of your
priorities, experience, and world model (e.g., a filter) as a situated
agent. Goals are just one priority, but goals to not really drive
perception, they mostly seek
to manipulate the frame to achieve a desired result. Loosely, I think
the frame problem is much more about constructing a useful and tractable
model that the situated agent can use towards building a plan to achieve
its goals.
|
Well, I think "goal" is the wrong way to understand the operation of the
brain though it's not too far off.
The true goal of a reinforcement learning machine is to maximize expected
future reward. So it's a reward maximizing machine with one prime goal.
What the prime goal translates into is some internal systems of values for
all possible behaviors which in turn translates into some behavior
probability distribution. This in turn must drive whatever mechanism is in
place to select behaviors. The system that decides what behavior to select
for the current context is using the internal system of values to pick
between alternatives.
Our higher level ideas of "goal seeking" is simply the fall out of a the
lower level behavior selection system picking the best behaviors for any
given context.
When you translate the implementation of the system into a reward trained
behavior selection system, the frame problem doesn't even make much sense
to talk about. The frame problem arises nearly as much out of incorrectly
framing the question of what the purpose of the agent is. However, the
issues that surround the frame problem are real. But they are all answered
in the context of a system which has the power to prioritize all possible
responses to stimulus signals. That is, which reaction the system chooses
at any point in time based on its learned values (priorities if you like)
is the answer to how the system deals with the frame problem. That is, the
one problem it must solve (how to select which behavior to use at any
instant in time) is the same answer to the frame problem.
Finding a workable implementation of such a system is the path to solving
AI.
You said above that goals do not drive perception. That's just not true in
my view. II think our perception and our behavior selection are one and the
same problem. Perception is a problem of behavior selection.
On the sensory input side of the network, the major function is perception,
but as the signals flow through the network, the function transforms into
behavior selection. So near the output side of the network, it's mostly
"goal driven" an on the input side it's mostly "perception driven" but I
believe it's a fairly even continuum though the network as raw sensory data
is translated to raw effector output data.
We see how this works when we test color perception of people raised (aka
trained) in different cultures with different words for different ranges of
colors. Our perception of color bends to correctly fit the classification
of light frequency labeled by the words of our language.
| Quote: |
Attention is paid only
to world states which bear on the system's goals (as a background
process).
Of course, goals to play an important role in how to focus attention, and
to some extent this colors the frame problem, but I do not see how it
drives it exclusively as you put it.
|
It drives it exclusively in my view because behavior selection is all the
brain is doing and behavior selection works by picking behaviors that are
estimated to produce maximal expected return for the given context. And
this general process of selecting the "best" behaviors at the lowest level
is both the mechanism which creates what we think of as goal seeking and
the behavior which is think of as attention focus. I see them as one and
the same process at the low level.
| Quote: | If I look at a movie I have no
goals other than to be entertained, however, I clearly create the proper
frames needed to understand and appreciate the movie and its meaning.
|
Yes, but what you call "creating a frame" I would call "picking the
appropriate behavior for the current context.
For example, we see a picture and we recognize various items in the
picture. It might be a street scene for example with a car parked on the
street in front of a pizza restaurant. How does our brain react to this
stimulus? It will react by activating part of our brain which represents a
car, and another part of the brain which represents a city street, and
another part of the brain which represents a pizza restaurant. But the
brain is doing this because it has been trained by experience to classify
that sensory data in that way.
If we expose a person who has never seen a car or never seen a city street
or never seen a pizza restaurant before (say someone who spent their entire
life in a jungle), their brain will react to this stimulus in a very
different way. They might notice the tree in the background because it's a
type of tree they know very well but have no real clue what anything else
in the picture is.
This is because this person has never had experience with this type of
"frame" in the past and has very little experience with how to correctly
react to this combination of stimulus signals. None the less, the brain
will still pick a reaction to the stimulus signal based on the past
experience the person has had. For this guy, the "reaction" to the frame
might be to move the eyes to focus on the tree in the background because
all the city street stuff in the foreground looks mostly like "noise" to
him.
| Quote: | In
that case, I am actually trying to discover the goals of the movie, and
not my own, to understand it. No goals on my part, but the frame problem
seriously exists.
If not enough information is in hand to solve the current problem, then
the
system returns to "the world" to gather additional information. (There
is no need to "store millions of facts." Facts are gathered as they are
needed, i.e., in light of the present goal and problem).
I don't think anyone would say that classic AI would not return to the
world to gather more facts to add to its "millions of facts". The issue
that Dreyfus says is the problem with AI is that it creates rules that
are representations (or symbols) and are compartmentalized, both of which
he says the Philosopher Heidegger espouses, which Dreyfus and his set of
philosophers/researcher say is not the case. I think every Intelligent
system will end up effectively having a constantly evolving set of
millions of "rules", so that is not the question. Do you have any
counter examples?
Cheers!
Ariel B.-
|
I don't fully understand what you are suggesting here because I don't tend
to read or study the work of the type of people you are studying. I'm not
sure for example what the debate is on representations.
However, I do believe that there is a tendency in AI to build into the
machine the wrong type of representations. Such as classic GOFAI tried to
build representations that were roughly equal to the level we would create
in our natural language. Meaning, in the AI implementation, there would be
something like a database which was close in structure to a natural
language dictionary which attempted to define the meaning of the words
based on their relationships to other words and to perhaps sensory inputs.
This didn't work well because it was making the assumption that the brain
worked by manipulating world-level symbols. These attempts made it obvious
that there was a much finer level of detail missing from our definitions -
the common sense problem. Some thought that simply using more words would
solve that problem. In theory I think that's correct, but in practice,
it's the wrong direction. Tje entire GOFAI approach seemed to be based on
the idea that since humans produced sequences of words and that we called
this type of behavior "thinking" that we could make a computer "think" by
also producing sequences of words. I think that is where the entire GOFAI
approach went off track. How the brain works, and what the brain does, are
not the same thing.
However, just because the GOFAI approach ran into a wall after awhile, that
doesn't mean that using "symbols" to represent something was wrong.
The brain uses a common language of symbols to represent everything as
well. Those symbols however are spikes. Digital computers use 1 and 0
symbols to represent things. Manipulating symbols which are
representations is exactly what the brain is doing. Any argument to the
contrary is misguided.
The solution to creating human like behavior in a machine is to build
symbol manipulating machines (aka signal processors), but the symbols must
be a level closer to spikes or bits, than to English words.
I also think that the correct implementation is along the line of a
confectionist network which is processing multiple parallel signal flows.
So from that perspective, I think it's more useful to think of the network
as a signal processing machine than as "representations with symbols". But
it's the same thing no matter which way you talk about it. An AM radio
signal is still a representation of the vibration of the air, and is also a
representation of the thoughts of the DJ which was speaking on the radio.
How you choose to label these systems is matter of viewpoint far more than
a true matter of what the system is doing or how it works.
--
Curt Welch http://CurtWelch.Com/
curt@kcwc.com http://NewsReader.Com/ |
|
| |
|
| Back to top |
Curt Welch
Guest
|
| Posted: Wed Nov 19, 2008 9:29 am Post subject: Re: Critique #1 Re: A Critique of Prof. Hubert Dreyfus' "Why |
|
|
Josip Almasi <joe@vrspace.org> wrote:
| Quote: | Publius wrote:
Disputes about representationalism appear in AI discussions because the
disputants are not distinguishing between intelligence and
consciousness. The latter almost certainly entails representationalism;
the former need not, but natural intelligent systems may employ it.
It's an empirical question.
Actually there's yet another big question - is consciousness emergent
property of (sufficiently high) intelligence.
And there's an empirical part of it - how to make it:)
|
Well, I think people in general are very confused about the entire subject
of consciousness, so when they talk about it, they are often talking about
many different aspects of humans. However, I believe the foundation of
where all this confusion comes from, is in how the brain tends to model
itself. That is, our entire understanding and perception of reality is
created by the processing that happens in our brain, and that includes our
perception of our own brain. Our perception of self. Because the brain
has only limited access to itself, the model it creates is also limited.
Based on the sensory data the brain has access to, it is forced to model
internally generated brain signals as having no association with external
sensory signals. It does this simply because there is no temporal
correlation between auditor sensory data and the brain signals which
represent our private thoughts. That is, we can not hear the brain making
any physical sound as our neurons fire. Because the brain makes no noise
(that our ears can pick up) when it operates, there is no correlation
between auditory sensory signals, and private thought signals. And because
there is no correlation, the brain doesn't create an association between
these different signals. This is true for all the physical sensory data
that flows into the brain. We can't hear our private thoughts, we can not
feel the brain vibrate, we can not smell it. There simply are no
correlations between our physical sensory data and our internal thought
signals.
The result of this fact is that the brain builds a model of this data by
indicating no association between private thoughts, and the physical events
represented in the sensory data. The result of that is very simple - it
leaves us with a model of realty where thoughts are disconnected from all
things physical.
This model the brain builds based on the signals it has access to, is the
source of all the confusion about human consciousness. It's why people
think the mind and the brain are two different things. People think that,
because that's exactly the model of reality the brain builds to describe
itself.
The net result of this is that "consciousness" is the illusion of
separation between our thoughts and our physical body. It is the source of
the illusion that humans have a soul which is separate from the body and
the reason the mind body problem has been debated for hundreds of years.
When we look at the world around us, we believe we see the world as it is.
Rocks are hard, and pillows are soft because this is the way those things
actually are. But in fact, rocks are hard and pillows are soft because
that's the way the brain has modeled them for us. For the most part, the
model of reality the brain creates is an accurate and close match to the
way the universe actually is. It's so close and so accurate that we just
accept that what we "see" is what is really there. This faith in our own
ability to see what is "really there" is what creates the hard problem of
consciousness.
When people look at their own brain, they see a mind which is separate from
their physical body. When I sense words and images popping up in my mind,
I don't in any sense "see" it as the firing of neurons. And the fact that
I don't "see" it as the firing of neurons, or the physical actions of my
brain, is the illusion of consciousness. We believe that what we "see"
when we look into our own mind, is what is "really there" because we trust
that when we look at things, we see what is really there.
The brain works so well at correctly modeling reality, that we make the
mistake of believing that what we see must be real. If we see "thoughts"
that are separate from phsyical objects, we assume the thoughts must not be
physical. But that's the error. That's the source of the hard problem of
consciousness. What we see is what the brain has decided is there, not
what is really there. And the brain has decided that thoughts aren't
physical simply because the brain doesn't make any noise when it operates.
So, having said that. What exactly is "consciousness"? Is the the fact
that we have a signal processing brain that is producing output reactions
in response to stimulus signals? That's nothing special because all our
robots do that already.
Or should we say the robot was "conscious" only if it made the same error
humans make of believing its signal processing hardware existed in a domain
separate from all things physical?
The illusion of separation is not anything special that evolved in us for
odd reasons. It's just a normal and expected side effect of the way the
brain builds associations between signals. It's known as classical
conditioning and it's already fairly well understood. But what few seem to
grasp is that it's the reason all these humans think these "voices in their
head" are not physical events.
The reason the brain builds associations is also simple in basic nature.
It's a data compression technique which is an important feature of any
strong reinforcement learning system. It's done to allow a finite amount
of internal signals to represent a maximal amount of information about the
state of the environment by removing redundancy in the signals. Signals
that are correlated are merged to maximize the amount of unique information
in the signals. It's an information maximizing function that has obvious
and clear advantage to the reinforcement learning problem.
The side effect of this, combined with the fact that our thoughts don't
produce external physical effects, is the illusion of separation between
mind and body and this endless waste of time by the philosophers debating
the mind body problem which they have no hope of solving on their own
because they aren't investigating the behavior of reinforcement learning
algorithms in real time parallel signal processing networks.
| Quote: | But seems noone around doubts it is an emergent property, so no need to
emphasise distinction.
As for empirical part, representationalism is sort of top-down approach,
while say Curt seems to prefer bottom-up approach.
Can't say for Dreyfus though:)
Regards...
|
There is value in looking at the problem from the top down. But until the
bottom up understanding meets the top down work, the problem won't be
solved.
From the bottom up, we are going to first produce intelligent man made
machines, and then, we are going to use that understanding to resolve all
the unanswered questions about the implementation details of the brain.
Then finally, last in line, the philosophers will finally catch on to what
they never could figure out without all the empirical data by our working
AI machines and a full understanding of the brain.
I know exactly why everyone is confused about consciousness and I know (at
a high level) what the brain is doing and how it solves the AI problem
using a distributed signal processing network trained by reinforcement.
But it's going to take a lot of time for the rest of the world to catch up
to this because there's a large set of people who never will never
understand the fact that the mind body problem is an illusion created by
the way the brain builds models from the data it receives.
--
Curt Welch http://CurtWelch.Com/
curt@kcwc.com http://NewsReader.Com/ |
|
| |
|
| Back to top |
Curt Welch
Guest
|
| Posted: Wed Nov 19, 2008 9:29 am Post subject: Re: A Critique of AI research |
|
|
turtoni <turtoni@fastmail.net> wrote:
| Quote: | On Nov 18, 11:48=A0pm, c...@kcwc.com (Curt Welch) wrote:
People working on real world robotics questions are already trying to
solve
this problem and their work is already pushing AI towards higher levels
of intelligence.
Good luck but don't be surprised if other fields of research beat them
to man made artificial intelligence.
|
Which other fields are you thinking about?
--
Curt Welch http://CurtWelch.Com/
curt@kcwc.com http://NewsReader.Com/ |
|
| |
|
| Back to top |
Alpha
Guest
|
| Posted: Wed Nov 19, 2008 2:49 pm Post subject: Re: A Critique of Prof. Hubert Dreyfus' "Why Heideggerian AI |
|
|
On Nov 18, 9:41 pm, c...@kcwc.com (Curt Welch) wrote:
| Quote: | Alpha <omegazero2...@yahoo.com> wrote:
On Nov 17, 12:55=A0pm, c...@kcwc.com (Curt Welch) wrote:
All mental activity is a reflex.
This is silly nonsense. Perhaps *you* do not ever think about what
you are going to think about , (and that would explain a lot), but do
not assume that others do not partake of directed thought processes/
scenarios.
I can simply will myself to think about a blue cube for example! And
then proceed to do so.
The idea of a "blue cube" just popped into you head at some point as you
were witting this response. Right? Did you will yourself to think about a
blue cube before you first thought about the blue cube? Of course not.
|
Of course I did; I was thinking of what might be a good example of
directed thought and that you might understand and I consciously
searched my memory for such an example that I might have used before
and I then located the blue cube directed thought. And thence
operationalized it.
So much for your contentions.
| Quote: | At
some point there, the idea of a blue clue showed up in your thoughts
without any prior will to think about blue cubes on your part. Why did
that happen?
|
It did not happen that way; it was a directive to thought one to find
a thought 2 that had the qualities that I wished to express.
| Quote: |
It happened because you had some sort of thought such as "what is a good
object to give as an example"? And as a _reflex_ to that thought, the
"blue cube" idea showed up in your thoughts.
|
It was not a reflex; it was a conscious find operation. Sorry, but
your inability to think ooutside your cave and without blinders on
dooms your critique.
| Quote: | And as a _reflex_ to the
"blue cube" thought in the context of "create thought example", you
produced the sentence in the post about "I can will myself to think about a
blue cube and then think about a blue cube".
The point here is that you can't will yourself to think about a blue cube
before you first think of what you are going to will yourself to think
about.
|
But If I think about (in thought @ t=0)what I want to think about in
the immediate future (a thought @ T=1) - that future thought is not a
relfex; on the contrary - it is the result of a conscious process to
find the best fit for the preconditions in the first thought.
| Quote: |
Whether we call this sort of thought sequence your "will" is irrelevant.
|
Nope; it is part an parcel of the cause-effect chain.
| Quote: | It's still happening as a reflex to what was just happening in your brain..
|
Nope - perhaps all your thoughts are simply reflexive; and that would
explain your inability, as Casey puts it, to think more abstractly,
(or even beyond some limited non-abstract areas).
| Quote: | Each thought we have follows from the current context set up by recent past
events in the brain. The path of thoughts that get produced based on
context is a function of how our brain has been conditioned by a life time
of experience.
My brain doesn't produce this constant stream of English words because it
was built by the DNA to produce sequences of English words. My environment
conditioned me to string these sounds/words together in this sequence.
|
But brain is beyond mere conditioning in that it can create new
thoughts that could not have been the direct result of prior thoughts
but come into being on their own. Otherwise NO new ideas would come
forth.
| Quote: | It
makes no difference if I'm talking about the stream of words being produced
as private thoughts in my brain or the stream of words that get typed into
the Usenet message. It's all just conditioned behavior coming out of me as
a constant stream of brain behaviors based on recent past context. And
because the brain has feedback loops in it, that context is based both on
recent past sensory inputs as well as recent past brain behaviors.
What you talk about as will is just the fact that one brain behavior is
likely to regulate the next brain behavior. But when behavior A controls
behavior B, we say that B is a reflex to A.
|
No, we do not say that; your antiquated classical viewpoint has been
outmoded for decades now; seems you refuse to come out of your cave.
| Quote: | It's the same thing no matter
which way you choose to talk about it.
|
In your opinion, which as we have seen, it astoundingly naive. It is
not the same thing and that is why those with more knowledge about
these matters and who do not live in intellectuial caves choose (*AHA!
- CHOOSE*) to talk about non-reflexive thought. |
|
| |
|
| Back to top |
Alpha
Guest
|
| Posted: Wed Nov 19, 2008 3:03 pm Post subject: Re: A Critique of Prof. Hubert Dreyfus' "Why Heideggerian AI |
|
|
On Nov 18, 10:22 pm, c...@kcwc.com (Curt Welch) wrote:
| Quote: | "Isaac" <gro...@sonic.net> wrote:
"Publius" <m.publ...@nospam.comcast.net> wrote in message
news:Xns9B57F131CB7B0mpubliusnospamcomcas@69.16.185.250...
"Isaac" <gro...@sonic.net> wrote in
news:491f9f87$0$33506$742ec2ed@news.sonic.net:
Minsky, unaware of Heidegger's critique, was convinced that
representing a few million facts about objects including their
functions, would solve what had come to be called the commonsense
knowledge problem. It seemed to me, however, that the deep problem
wasn't storing millions of facts; it was knowing which facts were
relevant in any given situation. One version of this relevance
problem was called "the frame problem." If the computer is running a
representation of the current state of the world and something in the
world changes, how does the program determine which of its represented
facts can be assumed to have stayed the same, and which would have to
be updated?
Dreyfus is pointing out one consequence of the lack of a useful
definition of "intelligence."
Actually, he is doing much more than that in his paper. I just posted a
portion of its background section, but his paper sets forth what he
believes is why AI fails and how he (and certain philosophers/researchers
he relies on) thinks intelligence works. I will email you the paper for
your reference. Let me know if you are interested in me posting my
critiques of his paper for your response. Maybe you will defend his
positions?
It is problem which plagues most programs for producing
AI (which is not to deny that much progress has been made in that
endeavor).
We may define "intelligence" as, "The capacity of a system to generate
solutions to novel problems," and "problems" as, "Obstacles or
impediments preventing the system from attaining a goal."
I defy you to contrive a definition of Intelligence that works. For
example, using your current definition above, the Earth would be
intelligent because it is a system with the capacity to generate
solutions (e.g., extremely complex, yet stable atmospheric weather, ocean
currents, etc.) to solve novel problems of, for example, maintaining a
stable global temperature in the face of many (thousands) changing
(novel) variables that are constant obstacles preventing the Earth (Gia?)
from attaining her goal of minimizing temperature differences globally.
The earth is intelligent. So is the universe as a whole.
Life looks like it was designed by intelligence because
it was. The process of evolution is just one more example of the many
intelligent processes at work in the universe. Evolution is an example of
a reinforcement learning process
|
No, it is not.
| Quote: | and I basically consider all reinforcement
learning processes to be examples of intelligence.
There are many similar examples that use your language but are not
considered to be intelligent to anyone reasonable in science. Care to
update your definition or defend it?
Many people in science have no clue what they are talking about when they
use the word "intelligence". As such, they define what is, and what isn't
intelligent based on total nonsense and ungrounded speculation - as I've
said before - without using any empirical evidence to argue from.
Of course that doesn't stop them, because they like to claim things such as
"subjective experience is outside the scope of empirical evidence".
And then they tell us what _their_ subjective experience is like and use
their beliefs about their own subjective experience to "prove" an endless
list of nonsense ideas about the universe.
The typical argument and thought path starts with the belief that human
consciousness is something that exists only in humans.
|
No, that is not the typical argument as most in the field of cognitive
science and biology for example believe that creatures other than
humans have consciousness.
You spout nonsense once again; your premeses are incorrect/false and
therefore so is your entire argument.
>Then from there,
| Quote: | they make the argument that since humans have this magical attribute called
consciousness and other things like the Earth doesn't, that intelligence
requires consciousness.
|
No, they do not suppose that I requires C. Deep Blue has chess-domain
intelligence but is not conscious in the least.
| Quote: | But since they don't have any clue what creates
human consciousness,
|
Yes - there are clues - many of them. Brain syncrony (local or
partial) for example. Solcves the binding problem for example.
| Quote: | they also don't have any clue what creates
intelligence and don't really have any way to determine if the earth is
intelligent or not.
And when asked to explain what evidence they have to suggest this attribute
exists only in humans, they use the self serving argument that since they
|
Ther is no "they" there; that is one of your false premeses.
You argue here just like you argue elsewhere- strawmen and red
herrings and falsehoods and misconceptions galore.
| Quote: |
"known" it exists in them, and that other humans are physically similar to
them, that this stuff they known exists in them must also exist in others..
But all that argument and the arguments that grow from it are based on a
belief that has no support. The belief that "consciousness" is something
other than simple brain function. That consciousness is not an identity
with physical brain function.
|
There are few that believe that C is not generated by brain function.
False premise again.
| Quote: |
However, all the empirical evidence we have tells us that assumption is
wrong. And if we choose to believe what the empirical evidence shows us
(materialism) - then we know that there is nothing here to explain, other
than the physical signal processing that happens in the brain which
produces human behavior.
|
That's it folks - that is *all* theri is to underestanding everything
- it is just reinforcement learning - or wait - maybe it is just *all
*signal processing! BWHAHAHAHAHA! Yo cannot even get your mis-shapen
characterizations straight.
| Quote: |
Once you grasp the significance of what the empirical evidence is telling
us, all the need of defining intelligence as some sort of link with "being
conscious" goes away.
|
Strawman.
| Quote: | We are left with defining intelligence is some class
of signal processing algorithm that describes how the brain works. And
though there are multiple options there, none of them make intelligence
hard to understand. It's no harder to understand than any typical machine
learning algorithm for example.
I choose to use the fairly broad and generic definition of intelligence
being a reinforcement learning system which allows the concept to include
many processes other than just what the brain does - such as the process of
evolution.
You could easily restrict the definition to something closer to what he
brain does, which would be something more like a real time distributed
parallel signal processing network trained by reinforcement instead of the
far broader "all reinforcement learning processes" I like to use.
Introducing goals into the definition gives us a handle on the "frame
problem": the problem is framed by the current goal.
Goals are really related to the frame problem in that the "frame" that
matters is the one that reflect "reality" in the context of your
priorities, experience, and world model (e.g., a filter) as a situated
agent. Goals are just one priority, but goals to not really drive
perception, they mostly seek
to manipulate the frame to achieve a desired result. Loosely, I think
the frame problem is much more about constructing a useful and tractable
model that the situated agent can use towards building a plan to achieve
its goals.
Well, I think "goal" is the wrong way to understand the operation of the
brain though it's not too far off.
The true goal of a reinforcement learning machine is to maximize expected
future reward. So it's a reward maximizing machine with one prime goal..
|
Wrong of course. Thee are multiple interacting goals in any human.
ANd of course, humans oft do things that have little to do with one
goal or even several important goals that you would think would
produce consistently intelligent behavior - but it does not!
<remainder of nonsense snipped> |
|
| |
|
| Back to top |
Alpha
Guest
|
| Posted: Wed Nov 19, 2008 3:14 pm Post subject: Re: A Critique of Prof. Hubert Dreyfus' "Why Heideggerian AI |
|
|
On Nov 17, 5:01 pm, c...@kcwc.com (Curt Welch) wrote:
| Quote: | "Isaac" <gro...@sonic.net> wrote:
I completely disagree, Beyond learning and building knowledge, AI also
includes transcendental aspects of consciousness and self (soul?), which
are in metaphysics. Do you really think there is an E=mC^2 equation for
that?
Most definitely. As I said, I'm a strict physicalist. To me, the belief
that consciousness is something other than physical brain function is just
a widely held illusion or myth. People believe it, and debate it, just
like they waste their time believing, and debating, the nature of God.
It's just silly crap man made up for reasons that have nothing to do with
the nature of reality.
AI also covers the creation and appreciation of beautiful things, which
is in the 3rd pillar of philosophy: esthetics.
Beauty is created by the value the brain assigns to sensations and those
values are there because humans are reinforcement learning machines.
There's nothing more to it than that. Beauty isn't a mystery. It's simple
and obvious once you understand what we are - reinforcement learning
machines.
However, this is exactly the type of thing which is nearly impossible to
understand by using philosophy alone to try and uncover the nature of
beauty.
So, I believe AI touches
on nearly all aspects of philosophy.
Yes, I agree completely with that. Philosophy is one of many human
behaviors and if you don't understand where human behavior comes from and
what controls it, you will have no hope of answering any of the big
questions of philosophy such as the mind body problem and the nature of
consciousness or the nature of aesthetics. Those questions can't be
answered from within the field of philosophy alone. All you can do from
within philosophy is identify which concepts are compatible with each other
and which are not - you can't identity which set of beliefs are a valid
description of reality without checking the beliefs against empirical data
- which is something philosophy chooses to treat as being outside their
domain.
All you can do from within philosophy is create multiple possible answers..
|
Incorrect; philosophy, in relation to science, is the *question-
generating* or *question-vetting* operation.
| Quote: | You can't tell which is correct or how correct or incorrect a given
approach might be.
Moreover, (reverse) engineering
will not solve the problem and may actually lead to many dead ends by
just finding ways to go nowhere quicker and better. It will take a new
theory and philosophy to do it.
Reverse engineering has already solved it. Many philosophers however don't
understand this because they have created such a huge cloud of confusion by
spending so much time debating all the impossible answers they can't get a
grip on what the truth is.
Think of it like trying to empirically come up with QED or Relativity w/o
any new theory or philosophy of physics.
You have started with the assumption that there is something there
(consciousness) which is fundamentally hard to understand and explain.
Your assumption is invalid. Your assumption is created by a simple to
explain brain function which created in all of us a natural illusion.
|
Please explain how brain genrates consciousness. Since it is simple
in your mind, it should be simple to tell us how that happens, get it
published in a referred journal (perhaps Journal Of Consciousness
Studies or Brain and Behavior etc.) and thence cvlaim your Nobel.
| Quote: | If
you assume the illusion is real, you are left with the hard problem of
consciousness. If you assume the illusion is only an illusion, then there
is no problem at all - all hard questions are answered and explained
leaving a fairly simple material world to understand. By Occam's razor, I
choose the answer that makes everything simple and answers all the
questions instead of picking the answer which creates contractions that
have no answer.
|
Please answer - oh sage - how brain generates the thought and the
concommitant mind-visual accompanyment: blue cube? I want an exact
specific answer now in terms of how brain's APs or molecules represent
the visualization and the semantic content inherent in : blue cube
thought. Should be easy for you right!
| Quote: |
But in philosophy, Occam's razor has no place. All alternatives must be
explored - as such, you are forced by your very charter to wander endlessly
into utter silliness. The hardness of the problem attracts you to explore
the depths of the silliness endlessly hoping to find some "new theory" or
sudden enlightened understanding which clears it all up.
As an engineer, I have no need to explore such an improbable dead end. If
I missed something, the philosophers of the world will find it and explain
it.
|
You have missed everything if import.
| Quote: | But after hundreds of years not finding an answer, I'm not holding my
breath on the expectation that there is something there when all evidence
suggests there isn't anything there to be found.
|
But being in the cave, you would have no knowledge of what others
outside the cave have come up with. Or when you do see the shadows of
such on your cave's walls, you dismiss them and skulk back to
reinforcement learning and signals (or wait...is it really all just
"particles" (haha) interacting) as the only thing that exists.. |
|
| |
|
| Back to top |
Alpha
Guest
|
| Posted: Wed Nov 19, 2008 3:26 pm Post subject: Re: A Critique of Prof. Hubert Dreyfus' "Why Heideggerian AI |
|
|
On Nov 17, 10:59 am, Neil W Rickert <rickert...@cs.niu.edu> wrote:
| Quote: | "Isaac" <gro...@sonic.net> writes:
2nd critique, on his page 12, line 4:
"Heidegger's important insight is not that, when we solve problems, we
sometimes make use of representational equipment outside our bodies, but
that being-in-the-world is more basic than thinking and solving
problems;that it is not representational at all. That is, when we are
There's that "being in the world" mystification.
My critique #2:
is not the Heideggerian view requiring this unity between the mind and the
world result in a "contrived, trivial, and irrelevant" world representation
scheme in people when the events in the world are so far beyond a person's
ability to cope (relative to there internal representation/value system)
that they just end up contriving a trivial and irrelevant internal world
that is just projected onto a "best fit/nearest neighbor" of a
representation that they can cope with. In this way, there is no absorbed
coping because it requires a perfect and accurate absorption scheme between
our mind (inner) and the world (outer) that does not exist and cannot be
magically created, even biologically. If you ignore this aspect of the
Heideggerian view then what you end up with is nothing much more than an
"ignorance is bliss" cognitive model that is not too different from what you
say is wrong with Brook's approach. That is, your portrayal of the
Heideggerian view of absorbed coping would exactly model the thinking and
representation behavior of insects, which certainly is not the conscious,
cognitive model of humans. Thus, this Heideggerian view of absorbed coping
is either insufficient to describe the human condition or it renders
indistinguishable insects from humans; either way it does not seem to
uniquely capture the behavior at the level of human consciousness and is,
thus, flawed at best. That is, if this Heideggerian view of absorbed
coping equally applies to any animals or insects then it is not really
helpful to modeling or shedding light on higher human intellectual
behavior, which, of course, is the sole subject/goal of AI. Moreover, this
"perfect absorption" is a complete illusion and in practice will only exist
in the most predictable and simple situations. From another angle, how is
this Heideggerian view of absorbed coping much different from the standard
psychological model of projection where our internal model/representation is
simply projected onto the world (or a subset frame of it) and we just trick
ourselves into believing that we are completely and accurately absorbed with
the true essence of the frame problem. this Heideggerian view of absorbed
coping seems to much more fit the unconscious aspects of the human
condition, which is more insect/animal like. This all seems to be logically
flawed and/or a very weak foundation for grandiose conclusions about what
philosophical approach/model is needed to solve the frame problem and human
consciousness. Maybe I am missing something critical here that can make
sense of it. Please clarify the logic.
I agree with the overall sense of your criticism, that Dreyfus
is giving accounts in vague terms which don't really say a lot,
and isn't giving a persuasive argument for this retreat to vagueness.
Any thoughts on this issue?
Here are some of my own views.
AI and epistemology take the problem to be "how do we use
representations". Dreyfus seems to want to do away with
representations. To an extent, I sympathize with Dreyfus, in that
I see an over-reliance on representations in epistemology and in AI.
But we cannot do away with them. Clearly, people use representations
in their use of natural language.
To me, the question is not "how do we use representations?" Rather,
the question should be "how do we form representations in the first
place?" This seems to be a difficult problem.
|
Yes - and one that Curt glosses over (or maybe he has the Nobel-prize-
winning answer to the question I put to him in another post - we shall
see.)
>As best I can tell,
| Quote: | our digital technology has not solved this problem. We have solved
some problems, in that we can digitize music for recording on CD,
and we can digitize pictures with our digital cameras. But the
representations formed with these digitization methods are not at
all similar to the representations that we see in our ordinary use
of descriptions in natural language.
|
Nor are they similar to the way brain may be doing so; ther is no
evidence of digitization functions in brain, even though we summarily
sequester APs (fire/no-fire) as analogs (sorry for the pun) of the
digitization process.
| Quote: |
On page 2, Dreyfus writes "I was particularly struck by the fact
that, among other troubles, researchers were running up against
the problem of representing significance and relevance". And that
seems to be the crux of the problem. AI people like to claim that
AI systems can solve this problem. But I don't see the evidence
that they have solved it. It seems to me that the computational
hardware is wrong for this. That is to say, I do not see how
significance and relevance can be reduced to computation. It seems
more plausible to suggest that it can be reduced to homeostasis.
|
Comes down to meaning (AKA significance); how does meaning arise in a
cognizer? Once we crack that nut we can relate it to goals and
intentionality.
I think meaning arises in relationships that arise between a cognizer
and its external milieu and also with its own internal memory
(thoughts related to other thoughts). And that meaning is in flux as
new "experiments" are being done by the cognizer (that is one aspect
of how we learn - by experiment), that alter the perception, by the
cognizer, of the salience of the experiment and thence forms a
contribution to the goals of the cognizer. I.e., baby does an
experiment - touches a hot stove - withdraws hand. Salience high
(OWW!) Meaning is thence high and thus a goal is structured or
altered - do not come near the stove again.
| Quote: |
Perhaps I should add that I'm an empiricist, at least in the
broad sense. Presumably rationalists can claim that significance
and relevance are innate, and computation need only apply that.
I don't see such an easy out for the empiricist. And even for the
rationalist, that doesn't really solve the problem.- Hide quoted text -
- Show quoted text - |
|
|
| |
|
| Back to top |
Alpha
Guest
|
| Posted: Wed Nov 19, 2008 3:27 pm Post subject: Re: Critique #1 Re: A Critique of Prof. Hubert Dreyfus' "Why |
|
|
On Nov 18, 11:01 pm, c...@kcwc.com (Curt Welch) wrote:
<snip>
| Quote: | I know exactly why everyone is confused about consciousness and I know (at
a high level) what the brain is doing and how it solves the AI problem
using a distributed signal processing network trained by reinforcement.
|
Please claim your Nobel already! |
|
| |
|
| Back to top |
Alpha
Guest
|
| Posted: Wed Nov 19, 2008 3:36 pm Post subject: Re: A Critique of Prof. Hubert Dreyfus' "Why Heideggerian AI |
|
|
On Nov 17, 7:56 am, ©uæMuæPaProlij <1234...@654321.00> wrote:
| Quote: | I'm an engineer, not a philosopher. As such, nearly everything you write
strikes me as silly and odd and misguided. I hardly know where to begin to
comment.
I find this sort of philosophical debate to be a pointless and endless game
at trying to define, and redefine words to make them fit together in a more
pleasing way. You can't solve AI by playing with words. You have to do it
using empirical evidence. It's not a problem which can be solved by pure
philosophy.
I agree with you. Creating AI has nothing to do with philosophy.
|
Except that historically, the important questions have come from
either philosophers, or by other types of scientists that posed
philosophical questions about Universe, per Bohm, especially when
instrumentality was limited (i.e., when our ability to *be*
empiricists (read: perform instrumented experiments) was limited.)
| Quote: | It is just a
technical problem that needs better mathematical tools in order to solve it.
|
What sorts of math tools are you talking about that would solve basic
questions of how brain for example, represents a blue cube, or how APs
represent (if they do) a thought or a memory?
See Koch's The Biophysics of Computation for an example of
sophisticated math applied to biological function in brain , but which
provides no clue as to representation etc.
| Quote: |
Creating AI will reflect on philosophy in only one way - it will prove that some
philosophers were wrong.
|
But then some will have been right! |
|
| |
|
| Back to top |
Alpha
Guest
|
| Posted: Wed Nov 19, 2008 3:38 pm Post subject: Re: A Critique of Prof. Hubert Dreyfus' "Why Heideggerian AI |
|
|
On Nov 17, 3:52 pm, ©uæMuæPaProlij <1234...@654321.00> wrote:
| Quote: | I completely disagree, Beyond learning and building knowledge, AI also includes
transcendental aspects of consciousness and self (soul?), which are in
metaphysics. Do you really think there is an E=mC^2 equation for that?
I don't know what "transcendental aspects of consciousness and self (soul?)" are
and I don't care. I leave this to philosophers.
AI also covers the creation and appreciation of beautiful things, which is in
the 3rd pillar of philosophy: esthetics. So, I believe AI touches on nearly
all aspects of philosophy. Moreover, (reverse) engineering will not solve the
problem and may actually lead to many dead ends by just finding ways to go
nowhere quicker and better. It will take a new theory and philosophy to do
it.
Think of it like trying to empirically come up with QED or Relativity w/o any
new theory or philosophy of physics.
I can tell you one thing - if we must wait philosophers to tell us how to make
AI then philosophers who think it is impossible to make AI are right.
|
OTOH, if we plod along using math tools, without knowing what
questions we want to ask about mind/brain/AI, then the plodding will
continue as it has! |
|
| |
|
| Back to top |
Alpha
Guest
|
| Posted: Wed Nov 19, 2008 4:33 pm Post subject: Re: Critique #1 Re: A Critique of Prof. Hubert Dreyfus' "Why |
|
|
On Nov 18, 9:03 pm, forbisga...@msn.com wrote:
| Quote: | On Nov 18, 6:24 pm, Publius <m.publ...@nospam.comcast.net> wrote:
Josip Almasi <j...@vrspace.org> wrote innews:gfujgq$4el$1@gregory.bnet.hr:
Disputes about representationalism appear in AI discussions because
the disputants are not distinguishing between intelligence and
consciousness. The latter almost certainly entails
representationalism; the former need not, but natural intelligent
systems may employ it. It's an empirical question.
Actually there's yet another big question - is consciousness emergent
property of (sufficiently high) intelligence.
It is more likely an emergent property of a particular implementation
strategy for intelligence.
There used to be a notion that an eye couldn't develop because of
its complexity. The assertion was that no where along the way
would the components have any survival value. I believe that dead
horse has been whipped enoungh. It seems to me that the same
applies to the development of consciousness in our set of evolved
systems. I'm not completely sure what the precursors were.
As for empirical part, representationalism is sort of top-down
approach, while say Curt seems to prefer bottom-up approach.
Can't say for Dreyfus though:)
How natural intelligent systems evolved is a different question from their
structure and function. Artificial systems might be designed which have
different structures, but perform the same functions and are just as
intelligent (or more so).
As with the eye, certain functional systems can be designed
without passing through all the evolutionary precursor steps.
|
Indeed; many such saltations have occurred as part of nature's
evolution via natural experiment.
| Quote: | This still leaves the open question as to whether or not
the implementation of the function(s) from which human
consciousness emerges necessarily entails that consciousness
will emerge from all implementations of the function(s).
|
It sure does! Some claim that merely embuing some entity or process
with general intelligenece wll somehow coerce or result in the
establishment of C. I fail to see how the two are related in that
way. C does serve intelligent functionality by providing a executive
feedback function of course as well as an I-ness that relates to
intentionality. Perhaps it is the other way around; C - even trivial C
as in bacteria or mice etc., - embues intelligence functions with
another aspect of mind/brain from which to draw conclusions etc. |
|
| |
|
| Back to top |
Alpha
Guest
|
|
