Emergence of behavior through software
Lynn H. Maxson
Sat, 30 Sep 2000 09:44:20 -0700 (PDT)
I have to thank Jecel Assumpcao for clarifying some issues about
emergent and non-deterministic behavior in software. I did not
set up the two-question pair to Fare as trick questions. I use
"consistent" as the qualifying term to determine whether the
results conformed to the embedded logic of the software or not.
If they (the results) did, then regardless of anyone's ability to
fathom or predict them, the software executed as instructed and
thus brought nothing "extra" or "special" to the process. In
short it did nothing on its own "volition".
If they did not conform, then we have to account for (1) the means
through which it achieved "take off", the ability to fly on its
own, and (2) why given this "independence" it would choose (as
clearly it has the choice) to continue to pursue the "original"
purpose and not determine to pursue a different one as part of its
Fare despite his cybernetic leanings will not grant the software
any choice other than pursuit of the original purpose. He will
allow it to improve upon its internal logic, but this raises now a
larger issue. How does software on its own organize itself
internally to "develop", "recognize", and "construct" higher level
of abstractions from the only thing it can execute: the machine
You see, how do we bridge this gap from externally set,
instruction-conforming behavior to internally set,
instruction-conforming behavior? What initiates the internally
set behavior from the externally set? If we disallow passing it
to it somehow as a form of inheritance, which we must disallow in
order for its "new" behavior to be its "own", then we are left
with the issue of "spontaneous generation", something occurring on
its own independent of the current consistent execution.
You see it cannot occur through meta^n-programming regardless of
levels. That presupposes that we have some ability to encode a
"triggering" event in the execution which will spawn the necessary
spontaneous generation. We do not. What we have is the machine
instruction set, the only thing that the software can direct the
machine to execute.
The only direction it can offer is that externally set by its
authors. The authors may or may not know (predict) the results.
They may or may not take or have the time necessary to retrace the
executed logic. But what they do know (something that the
non-intelligent software cannot) is that whatever results is
consistent with the encoded logic.
Truth is that Fare knows this as well. He knows that there are no
triggering events not present in nor otherwise determined
inconsistent with the embedded logic of the software. It can't
happen, because in von Neumann architecture any such occurrence is
an "error", something to be fixed. Contrary to his statement that
no one can know completely the internal logic of the machine, I
began my career being trained to know just that. If it failed, if
the results differed from the expected result of a machine
instruction execution, the embedded IPO (input-process-output)
logic of a machine instruction, my job was to diagnose and repair.
Spontaneous generation then in a von Neumann machine is an error,
again something to be fixed. The hardware does not support it
except as an error. There is no means in software translated into
machine instructions to make this possible.
Fare in acceding that software retains its "purpose" throughout
knows this. He also knows that regardless of our ability to
predict or fathom software-produced results it has nothing to do
with the results consistency with the encoded logic. We may be
surprised. The non-intelligent software lacks this altogether.
This is true for AI systems whether rule-based or neural-net.
Neural-net implemented in software is rule-based. Rule-based
software systems created by humans may reflect the result of human
"intelligence" (determining the set of rules), but neither the
input, the process, nor the output (result) do more than "reflect"
but not "absorb", that intelligence. In an of themselves absent
of an "observer" they cannot produce information. For information
implies meaning. Meaning does not exist in data or software.
Meaning, if it exists at all, does so only in the observer.
Are the results consistent with the embedded logic of the software
regardless of our ability to fully predict or fathom them? If the
answer is yes, then we do not differ in terms of causes and
effects. If the answer is no, then what spontaneous generation
does not occur in error? If it does occur, how does it still stay
within the "purpose" encoded in the software?
If we can all agree on results consistent with the embedded
software logic, then we can ignore considering inconsistency.
That leaves us then free to differ in our degree of individual
wonderment relative to what "we" have "created" through software.
Therein we can expect that in terms of wonder some are less or
more so than others. If that's the only thing which separates us,
then we can return to the main Tunes list.<g>