| |
ISSN 1320-0682 | |||
| Volume 02 | April 1995 | |||
Paul-Michael Agapow
Computer Science
La Trobe University
Bundoora Vic. 3083, Australia
Email: agapow@latcs1.oz.au
"Memories are just dead men making trouble."
Gabriel Marquez Garcia, The General In His Labyrinth
Culture, the accumulated information of society, is the most striking and perhaps the sole unique feature of humanity. The accumulated libraries, art, databases and architectures of the world are a massive store of extra-somatic information that has been snow-balling since the first crude cave paintings. Yet in terms of biology and evolution it is one of the most mysterious, passed off either as the inevitable product of "intelligence" or the result of socio-biological machinations by genes. This neglect is perhaps understandable. It is an experimentally formidable problem, and too holistic for the reductionist approach that underlies a great deal of the "Modern Synthesis".
There are several good reasons why culture should be considered in a biological context. At the very least, it is one of the most complex and recent products of the human genetic system and deserves investigation for that status alone. In an opposite sense, the impact (or potential impact) of such a system on human evolution should be studied. Thus in the following work, I propose to examine these questions :
Culture is a tricky term to define if we wish to encompass all possible means of long-term recording of information patterns that may convey meaning and be transmitted from organism to organism, including (to paraphrase Langton [10]) "culture as it could be". Fortunately for the purposes of this investigation, a precise meaning can be given. Borrowing from Csikszentmihalyi [3], extra-somatic or cultural information is :
"[the] transmission of information contained in artifacts"
There are some important things to note about this definition. First, while there are other forms of extra-somatic information (bacteria for example can be highly promiscuous in scooping up loose nucleic acids from their environment), we are interested only in those are deliberately created as external information stores [9]. Further, although culture necessitates communication (that is the ability to introduce and receive signals from the environment), it is not just communication. For such signalling to contribute to cultural information, communication must possess two qualities: it must be potentially one-to-many (a way of one organism broadcasting to several others) and must be decoupled in time from its originator (its existence and use does not explicitly require its maker).
Next, when we speak of the cultural, we are not speaking directly of society. Social institutions do not exist in the absence of a population, unlike extra-somatic artifacts.
Finally, note that, in principle, all creatures are capable of some form of communication in that all alter their environment and react to changes in their environment in a purposeful way. The quality and sophistication of communication of course reflects how sophisticated this environmental manipulation is. While even crude communication is of considerable benefit to a creature, it may not be sophisticated enough to support culture.
It is important when considering evolution to distinguish between how a structure arose and what role it currently fulfills. In the case of the mechanisms necessary for cultural information, their origin is unfortunately obscure. For brevity, it must suffice to note that the mechanisms are of very recent evolutionary origin and that there are no anatomical clues to be found in the brain, the human brain appearing to be exactly what one would expect for a large primate .
To envisage how extra-somatic information might be an advantage to somatic evolution, it helps to envisage how the various information stores of a biological system interact. Goonatilake [8] has proposed a scheme whereby information flow is divided into three domains: the genetic, the neural and the extra-somatic. The meaning of the first is obvious, the second refers to that data "stored" in nervous systems and passed on by imitation or teaching and the last to that held in artificial (cultural) information systems. Using this view, the domains may be pictured as a nested hierarchy, with the genetic inside the neural and the neural inside the extra-somatic. As one moves from the centre to the outer domains, information transforms and adapts faster, is more flexible and has correspondingly shorter "lifetime". Each sphere gives rise to the next outer, and co-evolves with it.
Thus it can be seen that each domain is best suited to cope with change on a different time-scale: genetics on a generational basis, the neural over fairly long periods during an organism's life-time and the cultural over shorter periods. The cultural information enables organisms to adapt at speeds their genetic and neural material are incapable of.
Perhaps more significantly, extra-somatic information may, in a sense, substitute for evolution. The accumulated culture of a society can serve as an extended memory, an adjunct to the neural architecture [6]. Yet this belies the advantages this extension has over the original item: it is possessed of a nearly limitless capacity without the metabolic costs or physical limitations of the nervous system and there are no constraints placed upon how the information within is searched or revised.
Finally, culture can act as a collective accumulation of a society's knowledge, giving individuals a wealth of knowledge that may be physically impossible for them to accumulate by themselves [9]. In a sense, every user of cultural information "stands on the shoulders" of those who have added to the culture before. Furthermore, extra-somatic information is a way for organisms to extend their influence temporally, beyond their physical lifetime.
Since the publication of "Origin of the Species" there have been numerous attempts to apply the evolutionary motif to other fields, often over-enthusiastically. While the difference between simple change over time or trends and evolution can be hard to draw, the point is not so much as to whether the analogy is correct but as to whether it is useful. Given as above that cultural information does interact with biological evolution, it seems natural to ask whether evolution can be used to cast light on the dynamics of cultural information.
Although not the only theory in cultural evolution, the most popular and intriguing is that raised by Dawkins nearly twenty years ago - "memes" [4]. In his own words :
"Examples of memes are tunes, ideas, catch-phrases, clothes, fashions, ways of making pots or of building arches, [propagating] themselves in the meme pool by leaping from body to body via a process which, in the broad sense, can be called imitation. If a scientist hears or reads about a good idea, he passes it on to his colleagues and students. He mentions it in his articles and his lectures. If the idea catches on, it can be said to propagate itself, spreading from brain to brain ... "
The meme concept is simple, yet powerful and self-obviously true. Memes are just like genes, in that they are simply replicators and thus obey the dictates of evolution. Memes with qualities that enhance their propagation (for example, ideas that are easily remembered, encourage repeating or reward those who use them) will proliferate. Memes with disadvantageous or unpopular qualities will perish unless coupled with a favoured meme in a "meme complex". Although the classic example of a meme is evangelical religion, a more subtle case might be urban folklore [2].
Little has been done with the meme idea in the way of verification or practical use. (The few exceptions cover ground all the way from the deliberate contamination of humans with infectious memes [7,13], through analysing market systems [11] to new strategies for Artificial Intelligence [16].) This is understandable in view of the considerable ethical and philosophical problems involved. Despite this, memes illustrate some important distinctions between biological and possible cultural evolution. Firstly, while biological evolution is carried out largely via impersonal "hand of God" selection, cultural evolution is mediated by the mind. Secondly, the existence of a neural domain depends on the existence of a genetic domain. A cultural system does not - although it certainly relies upon the neural (and hence genetic domain) for propagation. It can persist in the absence of the genetic.
As mentioned above, the investigation of cultural information presents some difficulties. The development in recent years of computational ecosystems (in which the dynamics of evolution can observed at an accelerated rate [17]), presents a possible tool for investigation. The following is a description of research aimed at studying:
For brevity, and given the ecosystems' similarity to several others, only an overview of the construction is given. Readers are referred to [1] for further details.
The computational ecosystem, GALILEO II, is in many regards, identical to conventional multiprocessing von Neumann computation. A population of virtual processors ("executors" with the usual stacks, registers and flags) inhabit a shared memory space. Executing in a quasi-parallel manner akin to an MIMD computer, executors perform a conventional fetch-decode-execute cycle. Executors may have privileged memory space and spawn new executors. However, executors may "die" either randomly or after their "health" (a variable initialised to a set value) reaches zero. Health may be decreased (or increased) at intervals (for example, per cycle) or by set conditions (for example, attempting to execute outside of assigned memory space).
Otherwise, GALILEO differs from conventional von Neumann computational largely in administrative details: interaction between ecosystem components is largely bottle-necked through a central "physics" module, so data collection, execution time and memory allocation and error rates can be centrally set. Additionally the decode step of the fetch-decode-execute cycle is done via a separate "instructor" module so different instruction sets may be implemented and used, even so far as changing sets mid-execution. Initial sets utilised a simple pseudo-assembly language.
There are several reasons to adopt such a model :
In order to allow extra-somatic information to arise, a few other modifications are necessary. A secondary shared memory space is allocated and an instruction set used that contains instructions for reading and writing to this space. The secondary space differs from the main one in two important details: there is no protected or allocated memory and its contents may only be read or written, not executed. Thus, we attempt to reflect the nested hierarchy as described above: the genetic domain is represented by the primary memory space, the neural by the stacks and registers of each executor and the cultural domain by the secondary space. Previous computational ecosystems, while providing a division between the genetic and neural domains, have not permitted the development of a cultural space.
Typically, computational "worlds" were set up by allocating primary and secondary memory spaces of up to a million locations. A small number of executors would be loaded at initialisation with allocated memory in the primary space containing a sequence of instructions which, on execution, would replicate elsewhere in the primary (genetic) space and spawn an executor there. Executor deaths were set to occur at random.
As work in progress, the analysis of results from GALILEO II is as yet necessarily incomplete and impressionistic in parts. Nonetheless, some conclusions can be drawn.
At first, evolution proved to be slow and if anything, tending towards simpler forms. On examination, the reason for this was obvious: execution time was being allocated evenly, with all executors receiving roughly equal shares of CPU time. Thus, there was an extreme selection pressure for executors replicating short sequences of instructions (that is, small genetic domains), as these could be copied faster and thus "out-breed" more complex forms.
As a solution to this, execution time was allocated proportional to the memory allocated to an executor (and hence the size of their genetic domain). Thus, small and large genetic domains would replicate at roughly equal rates. Even this proves to be insufficient to allow the evolution of cultural mechanisms, and execution time is eventually allocated so as to favour large, complex genetic domains.
A number of interesting utilisations of the secondary space started to occur. Normally, executors "decide" on which location they will attempt to replicate by simply generating a random location. (This is the scheme used by the primordial executor instruction sequence.) Intriguingly, three new schemes for picking this location evolved that used the cultural space. As the random generation of a location involves a significant instruction sequence (and thus, execution time), one of the first variants seen reads values from the cultural space as seeds for the procedure, allowing it to shorten the instruction sequence. A second used a variant of this where the value of a single location (interestingly, a prime number) was used as a constant in the random number generation.
The most sophisticated mechanism seen was where one "species" of executor evolved the practice of checking the value at a certain location in the secondary space. It would increment this value and use it to generate a location in the genetic space where it would try and reproduce. Thus, the species was able to spread out evenly around memory and avoid crowding.
Although no replicating memes were seen in the cultural domain, it is probably premature to expect this to happen before more sophisticated communications and uses of the cultural space are evolved. Happily, the evolution of the simple cultural mechanisms described above indicates that for some organisms at least, cultural mechanisms turned out to be an advantage. At no time however, did use of the cultural space become predominant throughout an entire population.
Even at this early stage, a number of future directions suggest themselves. On reflection, it may be possible to permit execution in the cultural space. While this was initially forbidden to keep replication activity in the genetic space, the lack of private memory in the cultural space may be sufficient to constrain replication. Conversely, it would be interesting to see if any executors evolve that have shifted replication activity from a private (genetic) to a public (cultural) space.
The execution and evolution time necessary to allow cultural mechanisms is somewhat severe. It is hoped that with longer evolution times (and possibly larger ecosystems) more complexity will arise.
It could be said that the reading of values from a cultural space is not really analogous to the real world, that absorbing information from cultural artifacts is not just a blind recording of data but an incorporation of that data and its meaning into data already resident in the neural domain. This allegation is essentially unanswerable given practical limits to the complexity of the ecosystems we can construct and analyze. If further investigations fail to turn up any use of the cultural space more sophisticated than those described above, this may be a possibility to consider.
Another more basic criticism is possible, that the phenomena seen above are far too simple and basic to be cultural information; that they are just methods of communication. In answer to this, it should be pointed out that these effects meet our basic definition of culture: artifactual information with an existence independent of its creator. Also, given the complexity of the organisms involved, less to a bacteria than a bacteria is to a human, it is to be expected that uses of cultural information is correspondingly simple.
Finally, we have previously reflected that culture may be the only trait unique to humans. The fact that fairly elementary computational organisms can make simple cultural devices is disquieting in a way. Perhaps this indicates that the absence of culture in animals is a claim that should be re-examined.
Bootstrapping Evolution with Extra-Somatic Information
This document was generated using the LaTeX2HTML translator Version 95.1 (Fri Jan 20 1995) Copyright © 1993, 1994, Nikos Drakos, Computer Based Learning Unit, University of Leeds.
The command line arguments were:
l2h -dir test pmaevol.tex.
The translation was initiated by Pam Milliken on Fri Aug 16 14:01:42 EST 1996