Emergence of behavior through software
Lynn H. Maxson
Thu, 05 Oct 2000 09:27:32 -0700 (PDT)
"That said, your theories really sound like you believe life comes
from some extraphysical divine "soul" that somehow directs empty
physical receptacles that are bodies. I bet your theory is
isomorphical to this soul thing."
I would remind you that it is not I who gave software a "life of
its own" based strictly on whether we understood it or not. I try
not to inject my faith or belief system as a causal element in a
discussion of this nature. Such are unprovable and have no place
in such a discussion.
"I bet, that, by induction, you can recurse down the bigbang,
at which time there was some fully developed seed for each of
modern-time species, as created by god. This is just ridiculous."
You lose the bet. I've been through this in another response. In
it I also did not inject God. We do not know how life or the
organism started or if this even makes sense in a timeless
universe. What we do know is that currently we have no means of
creating life except through transmission of a living organism.
The question easily becomes is there a difference between
artificial (man-made) life and life as somehow formed within the
universe? The answer I hope you would agree is "no". The
difference lies in the process man uses to create life forms.
Will that difference lie in attempting to replicate the process in
the manner in which it occurs or not.
"Why couldn't they? I imagine AI and A-life programs precisely
as starting as some simple symbolic organism, and evolving
There are two problems here. One lies in the difference between a
physical organism, one that exists physically in the universe, and
a symbolic one that exists entirely within human systems. The
second problem lies in the means of their evolving. You presume
that you can give a symbolic organism life. You regard software
as such a symbolic organism and when initiated within a hardware
body that the combination can become a life form, an organism.
"Maybe not yours. Maybe not most anyone's today. Yet, I've seen
people who did grow software. Genetic systems, reflective expert
systems, data-mining pattern-extracting systems, etc, do evolve
(and I hope to eventually bring them all together)."
How does software grow? How does software evolve? It grows by
someone writing. It evolves by someone writing. In both it
involves direction by an external agent. I don't care if it is
genetic, reflective, data-mining, whatever. It is not me who
injects God-like responsibility into this discussion.
I accept that software will always reflect the writing of its
authors regardless of how well they understand or can predict the
results of what they have writing. They may very well have
written it for just that purpose. However that purpose remains in
them and does not transfer into the software, which as a
non-intelligent mechanism, as a non-life form, does exactly as
Evolution within a sequence of organism generations occurs from
some intrinsic "property" within it with respect to the
environment of which it is a part. To achieve this with software
means having no "external" writing, only "internal". The
challenge lies in programming the seed, that initial piece of
software that acquires a "sense", an "awareness", and a "purpose"
of its own.
Now can that happen? Without a better understanding of how these
occur in living organisms we cannot rule one way or the other.
However we can fairly safely rule out von Neumann architectures
for the hardware and Turing rules for the software. I'm not into
denial here. I do say we do not have the right "materials"
currently. If and when we determine just what those right
materials are, we may find that hard-coding, not soft-, is the
You keep talking about meta^n-programming and ever higher levels
of languages, all of which we may understand, but none of which
have we invested in the physical means, the computer architecture.
No existing computer executes a meta^n-program or a HLL. What it
executes is their translation into a language (its instruction
set) it is constructed to respond to. I have to exercise caution
here and not use such terms as "understand" or "know", because a
non-intelligent mechanism can do neither. Here the metaphorical
use of language deceives.
If the answer lies in what you propose in terms of languages, in
terms of symbolic forms, then the machine must have the ability as
the authors do in terms of "understanding", "intent", and
"purpose". That says that you cannot do it through software. The
software cannot execute without a machine. The machine cannot
"know" more than its instruction set. Therefore that instruction
set must be capable on its own to "understand" and "create" ever
higher level of abstraction of its own. That, my friend, is
"evolution" from internal growing.
That is not a von Neumann machine. Nor are the governing rules
Turing. The secret here lies in developing both in sync as a
single system with no separation between directing and doing, the
same system that occurs in every organism. That means they do not
grow with "external" assistance, but only in response to it as
part of their interaction with their environment.
I would have thought as one respecting cybernetics that you would
have at least picked this up from Ashby's work, an understanding
of homeostasis, and the homeostat. In retrospect I should have
gotten a clue when you proposed that software should acquire all
these attributes except "purpose" which remained that of the
authors. That would imply that you hold volition, awareness,
thinking, understanding, knowing, and purpose as separable
components and not interdependent, integrated, interacting
"You blank out the notions of input and persistent state. Not to
talk about chaotic behavior and evolution."
None of these occur within software. None of them change the
execution sequences in software. All execution sequences are
programmer determined, i.e. consistent with the embedded logic. I
haven't blanked out anything. They don't change anything relative
to the embedded logic of the software.
"Rules offer partial information. Internal state provides another
body of information. Still same blanking out."
Rules offer no information whatsoever. Rules apply to inputs
producing outputs as well as changing internal states. Internal
states are data, inputs are data. Software processes data. It
has no ability to "view" it as information. It is a mistake to
imply that software sees meaning in anything that it does.
Software executes. It doesn't even know it is doing that.
Truthfully it will never know that. Only that within which it
executes, the physical form, can acquire that capability. That
form is not von Neumann-based.
"This is seemingly an essential point of disagreement between us:
you're obsessed with the low-level aspect of things, and do not
accept that high-level structure may be independent from
underlying implementational details."
I guess there is a difference between my view that the whole is
equal to the sum of its parts and yours that it is greater than
that sum. Apparently also in your view not all "wholes" are
created equal. That apparently in the evolving development of a
whole than an inequality appears spontaneously. Now just where
and when remains a question.
Thus far in software we have not created higher level abstractions
not composed directly of lower level ones. And these eventually
into the instruction set of the host machine. In fact in
executable form no higher levels exist, only the instruction set.
So where and how in this architectural model, the von Neumann
machine, do you program in an equality?
You will on the one hand berate me for injecting "life" as an
inequality into a physio-chemical equation. To you this means
that I see the hand of God in the process as well as a soul. On
the other hand you berate me for not allowing it in software which
you do. The truth is that I don't inject an inequality in
material composition to account for life but something in material
organization, some difference that exists when it exhibits "life"
than it exhibits "death". That says I am more for "composing" as
a life process than "decomposing" as a death process. In either
case a continuum in process (sequence of sub-processes) occurs.
This means in no instance does a material inequality occur. I
leave it to you to resolve your own contradiction.
"The details vary a _lot_, but the abstract structure stays the
same. Similarly, in as much as some high-level structure can
implement a system capable of having "intelligent" conversation,
it doesn't matter whether the underlying implementational hardware
be human brain cells, interconnected electronics, silicon, or
software running on a von Neuman machine."
Here I think you and Alik Widge make the same mistake (IMHO). You
posit that some (existing) high-level system is capable of
"intelligent" conversation even though you know if it is a human
conversing with it, the intelligence is actually one-sided. I do
not know if two of these machines converse "intelligently" with
each other or, if like us, tend to argue more.<g> The fact is
that it does matter greatly the underlying hardware
implementation. Furthermore it lies well-beyond existing software
techniques to equip a machine with human-equivalent conversation
capabilities. Neither a von Neumann machine nor Turing rules will
ever approach the conversational levels of humans...or even
You keep acting like I am denying something or have some fear of
future developments. What I deny is the ability of current
developments to do the job. Instead of beating our heads against
an impossible task let's get to the developments in which all this
is possible, if not more likely. Until you have embedded
high-level contructs within a machine, something higher than the
current instruction set, and embed the ability for it expand and
extend them on its own, what you achieve with your
meta^n-progamming and ever higher HLLs will never transfer to the
machine, the only place in which software comes to "life".
You have yet to provide an example not burdened by the
restrictions of a von Neumann machine. No amount of elaboration
or sophistication will overcome those restrictions. This makes it
increasingly difficult to transfer (communicate) the author's
thought processes to corresponding machine behavior. The
transmitter (the author) has the ability. The receiver (the
machine) does not. Therefore the author has to translate his
communication (the software) into the language of the machine. It
doesn't take much perusing of executable code to determine that
considerable is lost in translation. Obviously in terms of
communication at a human level (which is what you desire at the
machine level) considerable is lost in translation.
Improve the machine, make basic changes in its abstract form, and
the need for software (external direction) becomes minimal. Of
course, what you get may be as blind, blanked out, and bull-headed
I think the current conversation between Billy and Alik offer more
in substance relative to current hardware and software than will
the pursuit of this. Maybe we can return to it later after
resolving some more practical issues.