Re: The problem of intelligence.

Lester Zick wrote:
Unfortunately, nature equipped us with neither source code nor debug hooks so we are working backwards from tangible effects to intangible (through introspection) recipe.

By this line of reasoning, trying to answer the question "by what method do I do this?" with regard to mental effects by introspection alone will fail. However, this does not mean *all* efforts are doomed. What we may not observe we may still invent. 

Well the problem here with this observation is that it assumes there
is any other way to proceed. Are we interested in intelligent effects
or are we interested in intelligence? If the latter we really have no
alternative to the arm chair introspection method because we have
nothing to study except the process itself and that is only evident in
ourselves. If we proceed empirically by studying the process in others
we have to proceed via tangibles to study an intangible. So if we know
what we're interested in is intangible we have no choice but to study
the intangible process itself and not tangible effects. Our study may
not succeed but the success of studying tangible effects will not shed
light on the mechanics of an intangible process.

Ultimately we are interested in intelligent effects. Not only this but we would be satisfied if we had enough for our machine to be useful for processing information. Im not referring to intelligent results however, I dont mean we should find an algorithm for every useful thing we could do (such as searching, pathfinding, sorting etc). I mean we should build a system capable of doing that for us. *Those* intelligent effects are what we need.

In the name of this pursuit, human intelligence, which you consider intangible, and its effects are merely guides. We don't *need* to discover their inner workings.

Explaining the transition within the biological brain would require far more knowledge about its inner workings that we have today. This path is closed to us. 

Well the only reason it's closed to us is we don't recognize what it
is we're looking at or looking for. Neurology has dual implications,
one for impaired brain function in medical terms and one for ai in
mechanical terms. I agree that simply prospecting for intelligence
within the brain is hopeless without some mechanical recognition of
what it is we're looking for and I've argued against the neurological
approach in ai for many years for exactly this reason.

If you know what, say, a watch is and you've used one all your life but never bothered to peek inside and find out how it works, then given adequate observing technology you can open it up and observe it in action and deduce how it works. This is possible. There is no reason why we cannot do the same to the brain in theory, only our MRI scans and hammers don't qualify for adequate technology.

The "pull something our of thin air and run with it" approach is the robotics way. Its failure is what I was referring to. 

I understand. But then do you see the skill of skills tangible or not?
I see it as a measurable tangible which doesn't shed light on the
mechanics of intangibles.

I would draw an analogy with physics. The formulas of physics happen to be a tangible concepts rooted in something "pulled out of thin air". They were derived as approximations to reality. We do not know if reality contains infinitesimal quantities (i.e. intangible) but using them in our calculus proved practically very useful.

Why couldn't we draw up similar concepts for the search for intelligence? (I wouldn't propose "skill of skills" as such but the attempt was in this vein of thinking.)

Then the problem would be one of interpretation. If the brain's "inner code" is cracked, the so-called "intangibles" become tangible. For this reason we can't say they are truly intangible, although inaccessible to us now. 

Well you raise a fascinating issue here.It's true they're inaccessible
to us now and intangible in this respect. But the problem then becomes
whether even if interpreted correctly in mechanical terms the process
becomes tangible or remains intangible. I suspect most cognitive code
breakers tacitly assume the former. I think we may eventually get to
the point where we can decipher, say, a numeral "2" somewhere in the
brain. But making use of that information in specific neural terms
will certainly be difficult and may prove mechanically impossible.

Difficult no doubt, but impossible? Why? I see no reason why it would not be possible in theory (as in the watch example above).

It's a problem in what I call subjective mechanics.Normally we're used
to considering mechanics in terms of objective tangibles. What happens
in the brain that produces the mind and mental effects and yields
objective tangibles however is subjective in mechanical terms and not
objective. So even if we manage to decipher some numeral "2" in the
brain it's not evident we can say exactly how or in what form a brain
will make use of the information. We know the brain will make some use
of the information in mechanical terms. But if the mechanics is really
subjective we may not be able to identify the process in the course of
events in any particular mind.

I don't follow your argument. Either you need to better define the terms you're using or you have made some unwarranted assumptions. What exactly is "subjective mechanics"? How could mechanics depend on the observer? What we are talking about is strictly *objective*.

In the watch example I could decipher the numeral "2" within the inner workings of my watch just as well. I see no reason why I couldn't also observe the system to see what it did to and with my numeral to decipher how it worked. How is the brain different?

Of course this argument makes no practical difference. 

Au contraire, mon ami. Let me give you a very practical example of the
difference it makes. Let's suppose the brain actually functions as a
computer analogous to what we use today and the only epistemological
problem we face is deciphering its instruction codeset. But instead of
conventional bits of uniform size our cerebral computer has bits of
variable non uniform size whose size in any particular instance cannot
be accurately inferred. How do you approach the problem of intangibles
then? Because this is exactly what happens in the brain.

It couldn't be possible that the size of a bit in any particular instance could not be accurately inferred. The brain (our computer) must do this very function, thus it is possible.

There are many analogous algorithms that use variable length data and they have methods for determining data boundaries. Cracking these algorithms may be annoying but its certainly possible.

Is there another perspective, other than mechanical, which would work here? We are looking for a mechanical answer after all. 

Another excellent question. There are two ways to do mechanics. The
first is the conventional approach we're all familiar with in terms of
cumulative mechanics. But there is a second way to do mechanics as
well in terms of differences between or among parts. The first is what
happens outside the brain and the second is what happens inside the
brain that produces the mind and mental effects. Cumultive mechanics
correctly describes ordinary material interactions where differential
mechanics correctly describes operation of the brain in logical terms.

I have never heard of either term and don't understand how you mean to use them. If these are not your terms please post links to some references on them, otherwise explain what you're talking about and state your assumptions and reasoning.

I wrote a couple of interesting posts on the subject a while back on a
thread titled "The Sum of Its Parts" (7/25/5, 8/5/5) which describes
the mechanical implications involved. I'll repost these if you like.


You're gonna have to, google search is coming up empty.

I never meant to say intelligence is intangible. 

Yes but I do. I don't see any apparent resolution to the problem of
intelligence in terms of tangible --> intangible strictly in terms of
tangibles alone. So unless you're prepared to "see" intelligence in
some tangible I doubt you can approach the problem successfully
in terms of tangibles without resort to some intangible mechanics.

My approach is to define what it is we *want* in terms of the tangible. That is, what would satisfy our idea of intelligence. Then we ought to work top-down to satisfy those conditions. 

Sure but then anywhere you start you don't know will be the top of
anything except the problem you set. Besides which I don't see that
definition of any "top" tangible gets us any closer to any intangible.

We can't aim to comprehend the intangible as thats impossible by definition. Defining a "top" tangible will result in a tangible product which is the only way the whole idea makes any sense.

The mind-body problem (as my hasty Wikipedia search reveals) relates to the relationship between mind and body. The big questions pertain to whether or not there is some kind of "mind-stuff" or if the entire thing is an effect of matter within the brain. Questions regarding consciousness fit in here as well. 

Yes but the real significance of the mind-body problem is the same
tangible-intangible issue faced in cumulative versus differential
mechanics. Only the body is considered tangible and mind intangible.


Define any terms you make up before you use them please.

I think for purposes of AI, for us to even say that it is possible for us to build a (truly) intelligent machine we must assume that regardless of the mind-body relationship, it must be possible to replicate intelligent function mechanically. Note that we don't mean to say we are replicating the human mind, we are intent on creating an /intelligent/ mind, a problem solving device which may little resemble humans. 

Agreed. The problem I have with tangible definitions is intangible
approaches may not appear to produce tangible results initially if
tangible measures are used.

Intangible approaches? Approaches incomprehensible by definition? What are you talking about..?

* First and foremost, we are looking for an algorithm. We want this algorithm to run on a computer and exhibit certain properties. 

Okay.

* There may not be a single answer. There are likely many ways in which our constraints can be satisfied. There is no one "intelligence", what we are seeking may one of many different ways of doing the same thing. It is common in Computer Science to see completely different algorithms do the same thing optimally. 

I disagree. If we're looking for intelligence we need to find it first
and then contemplate what intelligence can do artificially.

Keep in mind the first point (to which you agreed). We are looking for an algorithm which produces certain effects.

I agree except that what you appear to mean by "certain effects" are
tangibles whereas what I mean by "certain effects" are intangible and
mechanically intangible. Huge difference. 

How can we look for effects which we cant comprehend by definition..?

That doesn't mean there is only one such algorithm. You say we need to "find intelligence", but what makes you think there is only one such "thing"? 

Because there are only two ways to do mechanics, in cumulative or
differential terms. Material interactions work in cumulative terms and
sentient operations work in differential, logical terms. What you get
though is a compounding of differences between differences and that
principle compounded in terms of itself which produces a vast array of
superficially different sentient effects whereas material interactions
in cumulative terms only produce one kind of material interactions.


Terms again...

the same way and function by the same underlying principles? Do we really know?

I think the case is readily demonstrable. Sentient organisms
function according to the level of compounding of differences
between differences they're organized around. 

You're reasoning from assumed positions neither disclosed nor proven.

* Our answer is not the silver bullet. Humans are capable of a great variety of skills and tasks, but we should not be overly optimistic with our goals. What we create may only be capable of intelligence within a limited domain. We are looking for generality, but we should not expect to find it right away. 

Well here you're making conflicting assumptions: one, that we
approach the problem top-down and, two, there is no silver bullet.
If there is no silver bullet there is no top-down. As far as I can see
top-down requires a starting place in mechanical terms and that
implies some kind of silver bullet which I would characterize as a
tangible-intangible transition mechanics.

When I say "there is no silver bullet" I am talking about how we decide with what we will be satisfied. For example, here I state a number of functions that I expect from an intelligence. If we were to create an algorithm which displayed such functions in a general enough way to be useful but not general across *all* problems, we should be satisfied. The practical importance of the point is that we shouldn't discard seemingly sub-ideal ways of meeting our constraints because they are not perfectly general (if there is such a thing!) 

Well I agree there is no silver bullet to the problem of intelligence
in tangible terms but only because the problem of intelligence isn't
one of tangibles but intangibles and intangible mechanics. But the
problem of intangibles and intangible mechanics can only be approached
with a silver bullet. So it's really a question of which problem you
want to tackle. An armchair is the only instrument needed for the
analysis of  intangible mechanics.


Intangible mechanics..?

* To define what properties we want to have, we need to pick the smallest set of traits with which we will be satisfied. There are many superfluous requirements that get placed on the problem. Ones such as "it must be reproducible in neurons", "it must be scalable", or "it must be built out of elements x, y, and z". If we heap on too many demands we will never find a solution, but then again, if we require too few we will not find an acceptable answer.

We can make all of these points without even touching on a definition for intelligence. But if we want to go further, we need to look at the defining traits of intelligence. What would we require? What qualities are we looking for? 

Well as far as I can tell the only defining trait for intelligence is
a tangible-intangible transition mechanics. Without that all you're
ever going to see are intelligent effects and not intelligence.

Actually that is all we need. An algorithm displaying intelligent effects to a sufficient degree would suffice. 

I disagree unless you have some way to limit intelligent effects. Just
approximating some effect to any degree doesn't display intelligence
of anything except the programmer designing it.

That argument is against describing, say, Microsoft Word as intelligent simply because it is functional. What I am talking about is a machine that displays intelligent effects not merely intelligent results. I list a few of what I mean below under cognition.

* Agency. Intelligence is a quality that pertains to an agent. Agency requires that there exist input signals from and output signals to an environment. These signals need to be sufficient for whatever task we want to accomplish.

*  Cognition. The agent should be capable of...
 ...accepting an arbitrary goal within the environment.
 ...accepting relevant information about its environment.
 ...planning actions that would lead to a more favorable position.
 ...creating sub-goals and plans for investigating them.
 ...abstracting, generalizing, and classifying information.
 ...generating relevant information from basic environmental inputs.
 ...reasoning about the environment from known information.

Most of these requirements are about mental effects. If we look at each item individually it is not impossible to think of algorithms that could do this. 

Definitely. Historically though the issue has not been the automation
of intelligent effects but artificial mechanization of intelligence
per se. We have all kinds of machines to automate effects but not to
automate the basic mechanism mainly because we don't know what it is.

We have machines that automated specific processes. We have no machine that automates *all* effects of intelligence. That is the goal, the truly intelligent machine that produces all expected intelligent effects. 

Well sure. Except the problem is with the "all". You have no way to
demonstrate what those are. And the only way to demonstrate those
would be with intangible mechanics because that and not tangibles
is what defines and limits the effects of intelligence.

Again I don't know what your "intangible mechanics" are but you are right, we can't say with certainty today what "all" effects are generated by the brain. This does not stop us from creating a wish list of the primary, desired effects and replicating them.

-- Risujin
.

Relevant Pages

  • Re: The problem of intelligence.
    ... what most people refer to as intangibles because we can't find the ... images in the brain. ... tangibles must be intangible. ... It's a problem in mechanics not philosophy. ...
    (comp.ai.philosophy)
  • Re: The problem of intelligence.
    ... >>>we have to proceed via tangibles to study an intangible. ... >> That's only a valid position if intangibles were not tangible. ... >> universe (and the mind) led to the common false belief that intangibles ... The mind and the brain are one and the same. ...
    (comp.ai.philosophy)
  • Re: The problem of intelligence.
    ... Those mental effects are what most people refer to as intangibles because we can't find the images in the brain. ... So if the brain is tangible then the imaging of tangibles must be intangible. ... But I'm definitely categorically opposed to defining intelligence anyway we want just so we can get on with mechanizing the definition without actually defining the mechanics of intelligence just so we can say we've made progress on some arbitrary definition. ...
    (comp.ai.philosophy)
  • Re: The problem of intelligence.
    ... And I consider the intangible mechanics of transition between what's ... >>>Explaining the transition within the biological brain would require far ... >> to considering mechanics in terms of objective tangibles. ... How do you approach the problem of intangibles ...
    (comp.ai.philosophy)
  • Re: The problem of intelligence.
    ... >> we have to proceed via tangibles to study an intangible. ... >> the intangible process itself and not tangible effects. ... >That's only a valid position if intangibles were not tangible. ...
    (comp.ai.philosophy)