- @#$%&! That wasn't how it appeared in the so-called

"preview". (One guess how soon I'll use the Yahoo!

Groups web posting form again.)

This one should come out properly:

I haven't gotten back to this religion thread because

I've been swamped, not because I didn't have anything

else to add.

If you go back and look, some of you might be

surprised to realize that I did not in fact profess to

be on either side of the "science is just another

religion" debate, because in fact I'm not on either

side. I do appreciate Chris Phoenix's exuberant

confirmation of my up to that point thinly supported

assertions about one of the common stances, and I hope

he won't attribute too malicious an intent to my

deliberately delayed confession of sympathy for both

viewpoints.

The problem, as it is so often, is that the sides are

talking right past eachother. Of course it's not

really true that science is just another belief

system, and it is true that some of the people on the

other side of academia mean to flatly deny this. But

there is another contingent which will concede that

science is in fact a more sophisticated and

theoretically distinguished belief system, while still

insisting that this distinction is not very

significant. And their point is much more than just

that scientific "knowledge" is always by definition

both contingent and incomplete�the much bigger point

is that much of our "reality", including particularly

most of the morally and politically important aspects

of it, are socially constructed, and thus in a much

more profound sense our reality really is _not_

objective.

Ironically, in fact, the more advanced our scientific

and technological knowledge become, the less and less

relevant it becomes to moral and political issues.

While on the one hand technology often seems to take

issues out of the hands of legislators, by

distributing capabilities to such an extent as to make

them beyond governmental control, and on the other

hand it produces issues the political system and

culture are ill-prepared to deal with, both of these

are merely the immediate, incremental effects. The

overarching broader effect is to successively remove

scientific and technological constraints on the range

of feasible political, economic, cultural systems

people can adopt, thereby putting a progressively

greater demand on our collective capacity for

imagination, courage, and discretion in order to

successfully determine and follow wise paths, rather

than go down very dystopian ones.

Stewart Brand made a similar observation in his book,

"How Buildings Learn"�the most successfully adaptable

buildings turn out to be those with constraints, such

as support columns, which greatly reduce the "design

space" which can be considered when contemplating

modifications. (Perhaps professional architects could

do more with less constraints, but most building

dwellers are not architects themselves, so apparently

less quite often turns out to be more.) I think many

video game critics (and some movie critics) have also

similarly suggested that games (or movies) were better

back when designers (or directors) couldn't fall back

on eye-popping graphics (or stunts & f/x, or sex and

violence) to keep players (audiences) entertained. And

Jaron Lanier is one among several who's voiced the

opinion that while the capabilities of software have

in fact gone up as hardware has improved, it has not

maintained the same pace of improvement, largely

because the quality of the _code_ has at the same time

gone very much downhill.

This doesn't bode well for our ability to "cope", as

it were, with the continually expanding possibilities

that accelerating scientific and technological

progress will continue to bring us. JFK observed that

we had the power to eliminate hunger in the world back

in the '60s, and yet it still hasn't happened. Instead

our politicians spend their time, for example,

facilitating ever greater abuse of increasingly

counter-productive IP laws to hinder all kinds of

things from online music sharing to the provision of

patented drugs to third world patients. Both are due

not to technological constraints but rather to

political ones. I don't want to preach to the choir so

I'll stop there, but I'm sure all of you have at least

a couple of other widely-recognized problems which

come to your mind, which society is either failing to

address or is continuing to itself cause because of

"political constraints".

On a related theme, "Mark L."'s musings on the likely

nature of a native or innate philosophy in AIs

actually made something click for me though, in a

moment of tiredness when I let my guard down enough to

truly consider it. One of the memes Jaron Lanier puts

forward in his Half a Manifesto is "cybernetic

totalism", which is basically the digerati version of

George Soros's "market fundamentalism" schtick. It's

also a fair definition of the philosophy that could I

think fairly be considered the obvious

pre-disposition, if there is any, of any A.I. system.

It is essentially a perfection of the reductionist

hypothesis, holding that not only is reductionism

valid, but that perception _is_ reality, and that

recognizing this "fact" is essential to true

understanding and sound moral judgment. The problem,

of course, is it's exactly the same type of

ends-trump-means philosophy which produced the

devastating seduction of much of the world by nazism,

fascism, and despotic communism last century. This

philosophy _is_ dangerous, to an even greater extent

than Lanier tried to explain.

Fortunately (for my own sanity), I'm still in the John

Holland camp (as he articulated it at the 2000

Stanford "Spiritual Robots" debate, shortly after the

publication of Bill Joy's infamous Wired article), and

don't believe the emergence of A.I. will be nearly as

automatic, inevitable, nor early as Kurzweil an

company expect, so I'm not terribly worried about it.

Barring, of course, the frightening possibility of

Lanier's inversion hypothesis being validated, and

producing a perceived success by moving the goalposts.

If we let this happen, then we will in fact create our

own dystopia, but only by (at least implicit) choice,

not due to any force of technological determinism.

I'll try to elaborate my thoughts on Zen and the

self-other dichotomy soon as well.

--

Kevin D. Keck - For another approach to the problem of science, rationality, and the

real world, I encourage anyone following this discussion to read my

recent Extropy-chat post:

http://www.lucifer.com/pipermail/extropy-chat/2004-April/005790.html

I begin by talking about rationality, building a case that the validity

of thoughts must be considered within their particular context. Usually,

the context is only within our heads, but we have the cognitive error of

believing that it extends much farther. If someone else's thought makes

no sense, it's probably because their context is different. Likewise,

your thoughts, however rational, are generally unlikely to be

trustworthy if applied too widely.

Then I discuss the consistent real world, and how it exists but we have

trouble addressing it even with science. I'll quote myself rather than

trying to restate:

"It's tempting to think that the world is a single context that

everything can be compared to. But this is equivalent to reductionism.

There are lots of things in the world that can be understood far more

completely by approximation than by first principles. For example,

human psychology has some really weird phenomena (phobias, optical

illusions, passive-aggressive behavior, etc) that a study of physics

will not help you understand. To a psychoanalyst or a politician, or

even a medical doctor, a study of shamanism will have more concrete

utility than a study of electromagnetism.

In fact, when dealing with people, not studying at all--not trying to

form postulates and practice formal thought, but just going on instinct,

intuition, and experience--may be more effective. This is because

people are incredibly complex, and we have a strong evolved non-rational

toolset to help us deal with them. In addition to people, things like

ecology may still be too complex for rational thought to improve on

accumulated heuristics, because we simply don't yet know the postulates

and methods. And then there are things like immunology and cosmology

where none of our tools really work yet, so the only way to approach

them is by study and rationality. Eventually, we can expect that study

and rationality will encompass psychology (including religion and

parapsychology) and ecology and everything else as well.

You mentioned the undesirability of chaos. The alternative to chaos is

the belief that a self-consistent real-world context exists. But even

though it exists, we can't access it directly. Science is motivated by

the desire to build conceptual contexts that map to the real-world one.

Its methods include cataloging (an underrated skill these days),

categorization, experiment, creativity, criticism, and more. In some

sub-contexts like electromagnetism, scientists have been very

successful; the mapping is very close. In protein folding, the end is

in sight. Pedagogy, psychology, and oncology are quagmires, though

oncology may be ready for a synthesis.

But back to the practice of science: the trouble is that scientists,

like everyone else, are prone to the illusion that their chosen context

extends everywhere. Let's be clear: I don't mean that scientists should

leave room for the paranormal or magical. They should not. I mean that

chemists should leave room for physics, and physicists should leave room

for psychology, and psychologists should leave room for chemistry.

Otherwise you get absurdities like chemists declaring that Drexler's

physics and mechanics work is worthless, when it's obvious they don't

even understand it.

One thing I never see addressed in discussions of rationality: How does

a rational thinker know when to keep their ears open and their mouth

shut? Obviously, the belief that a rational thinker will be an expert

in everything is irrational. But it's far too common. Scientists are

slowly learning enough to be rational in certain limited contexts. And

in a few glorious areas, those contexts have spread enough to merge.

But anyone who aspires to rationality should learn from the

overconfidence of scientists who, secure in their rationality, talk

nonsense outside their field. That's as big a mistake--I would argue

that it's the same mistake--as religious people talking nonsense while

feeling secure in their irrationality. The mistake is assuming that

their mental context extends farther than it actually does.

And scientists and rationalists have even less excuse than

irrationalists. If as great a scientist as Lord Kelvin could be wrong

about something as mundane and technical as heavier-than-air flight,

surely the rest of us should be extremely cautious when talking outside

our field of study--or even inside it, for many fields. But no, we keep

making the same mistake: our context defines our universe, and

everything we see must be made to conform. Appeals to rational thought,

in the end, are usually just another way to rationalize this process."

Chris

Ps. Note the very awkward formatting of your post; please correct that.

Pps. I should have cited a source in the Extropy-chat article: the

mundane explanation for the "loaves and fishes miracle" comes from a

book called "The Robe."

Kevin D. Keck wrote:

> I haven't gotten back to this religion thread because I've been

swamped, no=

>

that I d=

> t because I didn't have anything else to add.

>

> If you go back and look, some of you might be surprised to realize

>

another religion" deba=

> id not in fact profess to be on either side of the "science is just

--

Chris Phoenix cphoenix@...

Director of Research

Center for Responsible Nanotechnology http://CRNano.org - On Apr 24, 2004, at 12:41 PM, Chris Phoenix wrote:
> I begin by talking about rationality, building a case that the validity

It is probably worth pointing out that one can prove this

> of thoughts must be considered within their particular context.

> Usually,

> the context is only within our heads, but we have the cognitive error

> of

> believing that it extends much farther. If someone else's thought

> makes

> no sense, it's probably because their context is different. Likewise,

> your thoughts, however rational, are generally unlikely to be

> trustworthy if applied too widely.

mathematically for algorithmically finite systems (which includes a

subset of non-finite state machines in addition to all finite state

machines). In fact, the mathematical expression of this is one of the

more useful theorems of algorithmic information theory. An interesting

theoretical direction of this is that one can compute the limits of

correctness for a particular model in a particular context (the

"predictive limit" of a finite model).

Or to put it in simpler terms: In any finite subcontext, rationality

does not imply correctness, and correctness does not imply rationality.

But it is theoretically possible to compute the maximum probability

that a rational model is also a correct model. For some arbitrary

brain/machine, the actual probability will be of the form:

0 < x < predictive limit < 1

where "x" is the actual probability that some rational model is correct

in some context, and the predictive limit is the maximum theoretical

probability that a model might be correct in that context. Why there

is often a significant difference between "x" and the predictive limit

for intelligent systems is a complex topic that I'll simply avoid.

Humans have an extremely poor grasp of the predictive limits of the

model of the universe that they build in their brains. Not only are

many (most?) people unaware that rationality does not imply

correctness, just about everyone is oblivious to the predictive limits

of their rationality with respect to correctness. There are many

things in the universe that can only be modeled to such low predictive

limits in the human brain that one would have to be skeptical of any

claim as to the correctness of those models.

j. andrew rogers - You mean there's theoretical justification for what I said? Cool! Is

it thought to extend to systems that are not algorithmically finite as

well? What about algorithmic approximations to non-A.F. systems? Can

you give me a reference or two for this?

Chris

J. Andrew Rogers wrote:

> On Apr 24, 2004, at 12:41 PM, Chris Phoenix wrote:

--

>

>>I begin by talking about rationality, building a case that the validity

>>of thoughts must be considered within their particular context.

>>Usually,

>>the context is only within our heads, but we have the cognitive error

>>of

>>believing that it extends much farther. If someone else's thought

>>makes

>>no sense, it's probably because their context is different. Likewise,

>>your thoughts, however rational, are generally unlikely to be

>>trustworthy if applied too widely.

>

>

>

> It is probably worth pointing out that one can prove this

> mathematically for algorithmically finite systems (which includes a

> subset of non-finite state machines in addition to all finite state

> machines). In fact, the mathematical expression of this is one of the

> more useful theorems of algorithmic information theory. An interesting

> theoretical direction of this is that one can compute the limits of

> correctness for a particular model in a particular context (the

> "predictive limit" of a finite model).

>

> Or to put it in simpler terms: In any finite subcontext, rationality

> does not imply correctness, and correctness does not imply rationality.

> But it is theoretically possible to compute the maximum probability

> that a rational model is also a correct model. For some arbitrary

> brain/machine, the actual probability will be of the form:

>

> 0 < x < predictive limit < 1

>

> where "x" is the actual probability that some rational model is correct

> in some context, and the predictive limit is the maximum theoretical

> probability that a model might be correct in that context. Why there

> is often a significant difference between "x" and the predictive limit

> for intelligent systems is a complex topic that I'll simply avoid.

>

> Humans have an extremely poor grasp of the predictive limits of the

> model of the universe that they build in their brains. Not only are

> many (most?) people unaware that rationality does not imply

> correctness, just about everyone is oblivious to the predictive limits

> of their rationality with respect to correctness. There are many

> things in the universe that can only be modeled to such low predictive

> limits in the human brain that one would have to be skeptical of any

> claim as to the correctness of those models.

>

> j. andrew rogers

>

>

>

>

> Yahoo! Groups Links

>

>

>

>

>

Chris Phoenix cphoenix@...

Director of Research

Center for Responsible Nanotechnology http://CRNano.org - On Apr 24, 2004, at 2:44 PM, Chris Phoenix wrote:
> You mean there's theoretical justification for what I said? Cool! Is

It is only true for algorithmically finite cases, but since this seems

> it thought to extend to systems that are not algorithmically finite as

> well? What about algorithmic approximations to non-A.F. systems? Can

> you give me a reference or two for this?

to cover all likely "real" spaces, you get a lot of bang for that buck

as a pragmatic matter. In terms of references, they are sparse but

what you are looking for is probably "non-axiomatic reasoning systems",

and Pei Wang's work in this area is probably the best and most

accessible on the Internet. There has been an interesting bit of

activity over the last year or two toward the unification of the fields

of probability theory, information theory, computational theory,

reasoning/logics, and a couple other bits and pieces as different

facets of a single elegant universal conceptual model for

algorithmically finite systems. My theoretical point comes from some

of the bridgework that is unifying reasoning logics and algorithmic

information theory. There isn't a lot out there; the first mentions of

this general result is implied in some papers from the early '90s on

universal predictors and Pei Wang's stuff, but we've really only worked

it all out in the last couple years (and is still a work in progress).

Finite versus Infinite mathematics:

Algorithmically infinite systems are actually the standard assumption

for classic theory in these areas, and it is of limited utility. That

is how you end up with things like standard first-order logics. The

problem is that we missed a lot because of this. Some very interesting

things emerge when you restrict the mathematics purely to the finite

case, often in areas that were considered mathematically "undefined" in

the general case (mostly because the inclusion of infinite parameters

force an undefined value for theorems and functions that have rich,

interesting, and definable properties when restricted to purely finite

parameters).

As for what "algorithmically finite" means:

The classic "finite state" is an inadequate system descriptor for the

above area of mathematics, and the term "algorithmically finite"

denotes something distinct from "finite state", though there are

conceptual similarities. I actually coined the distinction a couple

years ago. I used to regularly argue with a math-savvy retired

Christian lady about the nature of religion and God in a mathematical

context -- I've developed a lot of good pure theory angles in the

course of trying to prove mathematical points to her, best exercise of

theory I ever got. She made the poignant observation that the apparent

algorithmic finiteness of the universe did not seem to have any obvious

dependency on the universe actually being a finite state machine in the

classical sense. And she seemed to have a point after I thought about

it for a bit, which I later formalized.

"Algorithmically finite" means (very roughly) a system that can only

express finite intrinsic Kolmogorov complexity in finite time. A

properly rigorous definition is fairly difficult to express well, and

tonight is not that night. Interesting things that fall out of this

are:

1.) This is inclusive of all finite state systems.

2.) The effective Kolmogorov complexity of these systems can vary in

time.

3.) This is inclusive of some infinite state systems.

The second property looks mundane, but is actually relatively

interesting. This essentially replaces an important given constant in

classic computational theory with a function. Since expressible

intelligence also varies with Kolmogorov complexity, this has

interesting implications. It is worth noting that this can also break

the assumptions of some theorems from classic theory.

The third property is interesting in that you can have infinite state

systems that are mathematically bound to express the computational

properties of finite systems over any finite span of time. An example

of such a system would be a system with a countably infinite state

fabric (say, at the resolution of the Planck length) and a finite bound

on information propagation (say, the speed of light), resulting in a

system which would be mathematically required to do things like express

an analog of the Laws of Thermodynamics that falls out of algorithmic

information theory. While such a system is nominally infinite state,

it is theoretically limited to the expression of finite algorithms with

a Kolmogorov complexity limit that varies in finite time.

From a functional standpoint, I would say that the AF model is more

general than the classic finite state machine model.

Okay, its past my bedtime,

j. andrew rogers