Evolution and art
Lance
paper on intermediate levels of explanation and variation and selection
mechanisms. Still, I think it might be of some interest to you folk. The
theory of art discussed is mainly that of Colin Martindale.
----------------
It is customary to start introducing some system into the many phenomena
encountered by man by classifying them either as "natural" or as
"artificial". But this dichotomy is misleading, for there is a class of
phenomena which lie between the natural and the man made -- the category of
"invisible-hand" explanations, where phenomena are produced by the actions
of people but are not the result of intention or planning. I wish to broaden
this intermediary class of explanations in two ways: firstly, so that it
contains not only invisible-hand processes but also many variation and
selection processes; and secondly, so that it contains not only the
unintentional products of the actions of people but also the products of the
actions of other creatures and even the effects of the activity of what can
best be described as "automata."
This broadening will result in a new hierarchy of explanations for the
phenomena which we encounter, ranging from the natural, through the
intermediate, to the intentional products of human action. Natural
explanations encompass both simple cause and effect explanations and
explanations which depend upon cybernetic principles -- feedback,
feedforward, and the general principles of control. Intermediate
explanations will encompass phenomena which arise from the collective
actions of self-regulating systems, which may include humans, but which
would also include creatures such as earth worms and corals, and even
relatively autonomous parts of more complex systems, such as assemblies of
neurones. The class of intermediate explanations would also include both
"invisible-hand" explanations and variation and selection mechanisms.
Intentional explanations, the last category of explanations, will apply to
products which are the result of human planning, and which change either
through a relatively slow process of incremental improvement of existing
products or through new conceptions of how to accomplish certain goals. Slow
incremental changes to existing products can easily be conceived of in terms
of variation and selection mechanisms, but the variation and selection
mechanism is weaker and has different properties from those found in the
intermediate class of explanations.
Now, it is easy to see that, at the intermediate level of explanations,
variation and selection mechanisms are specialized versions of
invisible-hand explanations. Both begin by postulating automata of various
sorts which are interacting with each other in various ways. However, in
invisible-hand explanations, the automata usually interact with each other
and the environment according to the same rules, whereas in variation and
selection mechanisms, the automata will differ in at least some of the rules
they will follow, or at least some of the properties they will display.
Consider the most famous example, Darwin's natural selection. Here the
automata are living members of a species which will vary naturally amongst
each other in their characteristics, and which will compete for scarce
resources, because of the pressure of population (Malthus's law), amongst
each other and amongst other species which inhabit the same environmental
niche. The variation in their characteristics means that some of them will
be better able to compete and hence will be more likely to survive and
reproduce. Here the selection mechanism resides outside of the behaving
automata, so that natural selection occurs as a side-effect of their varying
characteristics and competition. But Darwin also proposed a selection
mechanism which resides within the competing species -- sexual selection.
Here the variation between individuals is in the sexual attractiveness of
each individual creature, so that selection resides in the behaviour of the
opposite sexed members of the species.
A model somewhat closer to Darwin's sexual-selection is Colin Martindale's
(1986) account of the processes which shape the evolution of art. This is a
complicated model and we can at best only summarize a small part of its
subtleties. As with all other intermediate level explanations, the model
turns on the behaviour of automata -- in this case artists -- rather than
something abstract like the features of works of art. Martindale accepts the
Freudian theory of creativity, claiming that all existing theories of
creativity are variants of it. A work of art is created in two stages:
initially, an artist must access something like primary process thinking in
which relatively novel combinations of material can occur, and then these
novel combinations of material are elaborated upon by secondary process
thinking at a later stage to produce a finished work of art. However,
artists may differ in their access to primary process thinking, and in their
ability to employ secondary process thinking for later elaboration. Here
then is the analogue to the variation among animals within a species in
natural selection. Further Martindale points out that artists are subject
to two further constraints: an artist must be original and an artist must
maintain some contact with his or her audience. Nobody can become a poet by
copying out Shakespeare's sonnets -- a poet is someone who writes original
verse. A requirement that an artist produce original works produces a
requirement for novelty. It is clear that artists with better access to
primary process thinking will be able to produce more original works than
will artists with a more restricted access to primary process thinking.
However, here we bump into the second constraint: the artist must be able to
keep contact with at least some of his audience. It is unlikely that John
Cage's 4'33" of silence would have been accepted as a work of art in Johann
Sebastian Bach's time, but at least some modern musicians will accept it in
the twentieth century. The need to be original will select among artists
those more able to access primary process thinking; the need to maintain
contact with an audience will select among artist those able to access
secondary process thinking. Further, the constraint that an artist keeps
contact with an audience will also act as an invisible hand to create and
maintain artistic styles.
Consider an artist working at the beginning of a new artistic style. Very
few of the possibilities of the new style have been exploited, so it does
not require very deep access to primary process thinking to be an artist
making original contributions to the new style. However, nearly all the
possibilities of the old style have been exhausted, so that it requires very
considerable access to primary process thinking to be original in the old
style. Further, works in the old style now display such creative novelty
that the audience has difficulty following them, and is inclined to reject
them. Any particular artist might try to begin still another new style or
art form, but it takes considerable resources and effort to get the audience
to accept a new style, so that most artists will find it convenient to
exploit any existing style, for this simplifies their communication problem.
So we see the working of the invisible hand, pushing most artists via their
need for communication, to accept a common form or style. At the beginning
of the new style most people can make contributions because little access to
primary process thinking is required, so selection will probably be on the
basis of secondary elaboration. However, the audience will become bored with
increasingly elaborated but essentially familiar works, so that as the style
progresses artists will have to increase the primary process content of
their works, but not too rapidly, or they will loose contact with their
audience. As the style progresses, the personalties of the artists will
change: from men of affairs with well developed secondary process thinking
artists will gradually become increasingly difficult people with much
greater access to primary process thinking. The works of art themselves will
change to accommodate the increasing novelty and surprise of their content.
The possibilities inherent in the style will gradually be exhausted and, at
some point, the capacity of the audience to understand the works of art will
be exceeded. At this point, the artistic tradition will break down, and
artists will struggle to establish a new set of conventions by which to
communicate with their audience. Thus the history of art will show
characteristic divisions into periods or styles, each period will follow a
particular course of evolution from being predominantly rational to being
predominantly primary process in content, and artist themselves will vary in
their personal characteristics. Martindale has accumulated a great deal of
evidence for each of these claims.
It is most important to spell out the requirements for a substantive
application of variation and selection mechanisms, both in order to avoid
empty analogy with evolutionary theory and in order to detect gaps in
existing theories. There seem to be four requirements for a fully specified
variation and selection mechanism. Every variation and selection mechanism
should specify:
1. Criteria for determining selection (change criteria);
2. A mechanism for approximating these change criteria;
3. Some mechanism for retaining the selection;
4. A mechanism for creating "variation" from which a selection may be made;
Requirements 2 and 3 can be accomplished by the same mechanism, but are
conceptually separate, and sometimes are separate processes in fact. Also,
requirement 3 is relevant only to variation and selection mechanisms which
are concerned with explaining a continuous sequence of change.
(a) Change criteria
In Martindale's account of aesthetic evolution change criteria correspond to
the novelty of the work of art. In other words change criteria represent the
ideal towards which change is being guided. However, change criteria really
have conceptual rather than practical value. They allow us to understand the
nature of the mechanism -- what it does. Change criteria usually represent
the ideal and cannot usually be wholly instantiated in real variation and
selection mechanisms.
As can be seen, although we usually have a clear idea of the change criteria
themselves, it can be difficult to specify the characteristic 'C' which is
being selected for its contribution to an animals fitness. Colin
Martindale's theory faces a similar difficulty: most works of art are
complex and may be original in many different ways, so that it is not easy
to specify exactly what contributes to the novelty of a work of art.
(b) Mechanisms for approximating change criteria
Mechanisms for approximating the change criteria and selecting the best
alternative are very various, ranging from natural and sexual selection
through backward propagation algorithms in neural networks to the purchasing
and viewing decisions of art patrons and critics. The mechanism may provide
very exact decisions about the best available alternative (e.g., Decision
Theory), or they may only select the best alternative with a certain degree
of probability so that the mechanism works only because the selection
process is repeated very often (e.g., in evolution).
Given the great variety of the selection mechanisms, it is not surprising
that the question of optimal performance comes up so frequently. It seems
best to consider this question in relation to time. Diachronic optimisation,
that is, optimisation for future conditions, is clearly impossible, except
by accident. Variation and selection mechanisms are "induction" mechanisms,
and it is a well known philosophical finding that induction must presume
that the future will be the same as the past. But this presumption is often
false. A species which is well adapted to a tropical rain forest will not be
well adapted after mankind has destroyed the rain forest. An artist with
limited access to primary process thinking producing original works at the
beginning of an artistic tradition will not be able to contribute original
works at the end of the artistic tradition.
All variation and selection mechanisms can, at best, only optimise their
selection according to the conditions which prevail at the time in which
they operate: none of them have foresight about future condition. Therefore
it is possible for any of these mechanisms to lead to what might be called
"evolutionary dead-ends" in which over-adaptation to specialized
circumstances occurs. This is such an obvious and crucial limitation of
these mechanisms that it is surprising that a lively debate about
optimisation in variation and selection mechanisms has nevertheless arisen.
But, of course, this debate is about synchronic rather than diachronic
optimisation.
Synchronic optimisation -- that is, whether the selection mechanism can
pick the best realisation of the change criteria at a particular time and
place -- must be possible for at least some selection mechanisms, although
it may always be possible to construct particular extreme cases where the
mechanism will fail (see Staddon, 1987). The important point to bear in
mind, if optimisation is to be possible, is (i) that the set of
characteristics which are being selected must remain constant, and (ii) that
they must not constitute unstable bundles of features so that either no best
compromise point is available, or more than one best compromise point is
available (Coombs, 1987). It would appear that the more abstract the
selection criteria the more difficult it will be for any selection mechanism
to attain optimal performance. "Fitness", "sexual attractiveness",
"novelty", and the like are complex and abstract concepts, and can be
satisfied in many different and perhaps contradictory ways. Under these
conditions it seems inappropriate to talk about optimal performance.
(c) Retaining the selection
The need for some apparatus which can retain any selections arising from
the decision mechanism principally concerns variation and selection
mechanisms which explain a continuous sequence of changes, much as Darwin's
natural selection, Martindale's aesthetic evolution, or even backward
propagation in a neural network, do. Notice, however, that a retention
mechanism would actually be harmful to a mechanism concerned with short term
changes, such as Selfridge's demon model for recognising characters.
However, when a method for retaining change is required, this requirement is
important. After all, it is mainly because such a retention mechanism could
not be found that we have abandoned Lamarkian evolution (the inheritance of
acquired characteristics). Exactly the same objection has been raised
against Martindale's theory of aesthetic evolution: Martindale's theory
presumes that taste will be transferred from one generation to the next,
but, argue the critics, there is no reason to suppose that a new generation,
lacking the experience of the earlier generation, will be bored with the
same works which bore their parents. Martindale's answer is to argue that
generations don't really exist in a social as opposed to a biological sense.
Consider an audience of French poetry on January 1 of, say, 1650. It
consists of people varying in age, not of a cadre that could in any sense be
considered a generation. Consider the same audience on January 2. No doubt
several members have died, but they have probably been replaced by several
new members. However, the vast majority of the audience remains the same.
Whatever habituation occurred for them on January 1 continued on January 2.
New members had to either catch up with or be drug along by this process.
They were in such a minority that they could not influence taste in the
slightest. The same situation has existed on every day since January 2 1650
to today. The audience may have grown or shrunk, but at any moment in time,
those already in the audience must have constituted the overwhelming
majority. It was this majority which transferred habituation across the
'generations.' (Martindale, 1986, pp. 451-452)
In addition, one cold consider various "invisible-hand" mechanisms, of the
kind Gombrich (1979) talks about, such as ridicule for someone with old
fashioned taste, which would ensure that newcomers would rapidly catch up
the greater body of opinion. Gombrich asks us to imagine an individual
returning from the Spanish Colonies shortly after the publication of
Cervantes' Don Quixote. This individual, professing to love medieval
romances, will immediately bring scorn upon his head, for Cervantes had made
that art form appear ridiculous.
(d) Generating variations
Finally, we turn to the variation creating mechanisms in variation and
selection mechanisms. Here intermediate level explanations, which depend
upon automata, have an advantage over intentional level explanations.
Variation, in intermediate level explanations, tends to be embodied in the
different behaviour and characteristics of the automata. In contrast, we
really have no idea of how variation is created in intentional explanations,
beyond such vague formulations as that found in Martindale's theory -- that
it somehow involves access to something like primary process thinking.
Still, saying that variation results from a set of automata having differing
properties does not completely solve the problem, for we need an account of
the genesis of these automata, and that may prove to be very difficult to
accomplish. However, even should we find it difficult to explain how
automata come to differ among themselves, it is still comparatively easy to
demonstrate that the automata do differ, an advantage which is denied to
intentional level explanations.
The complexities which arise when trying to explain why automata differ
among themselves are amply demonstrated in evolutionary theory. These
complexities include: the phenomenon of sexual reproduction along with the
rules of inheritance, as well as such speculative possibilities as the
production of novel genetic material through mutation. Similar complexities
arise in other intermediate level variation and selection mechanisms.
If the processes producing variation are not always clear, two general
problems should nevertheless be considered when thinking about the
production of variation in variation and selection mechanisms. For the sake
of alliteration we may call the first the "cheetah" problem and the second
the "Chomsky" problem. The cheetah problem occurs when there is insufficient
variation to offer a reasonable chance of success in adapting to changing
conditions. I am told that there is now probably insufficient variation
among surviving cheetah's for them to constitute a viable species.
Martindale's account of artistic evolution offers other examples, for
artists with insufficient access to primary process thinking will simply be
unable to create a sufficient number of novel ideas for works of art.
Insufficient variation may lead to the additional difficulty that incorrect
solutions to the change criteria may be magnified. Consider the problem of
interbreeding in small closed populations. Any defective genes are likely to
be activated simply because the smallness of the population ensures that
sexual mixing of genes will no longer provide protection.
The Chomsky problem is the reverse of the Cheetah problem. Here it is
important that a correct solution be found in a short space of time. This is
similar to the problem of a child learning a language in a remarkably short
space of time from limited linguistic input. Chomsky claims that a child
could not learn the language through conventional learning mechanisms
precisely because the range of variation is too great -- conventional
learning or even scientific hypothesis formation and refutation would simply
take too long.
Reflection upon the Cheetah problem leads one to consider a general effect
which may occur, especially in intermediate level variation and selection
mechanisms. Because these mechanisms employ relatively autonomous automata
which interact with one another, we must always be alert for the possibility
of unexpected side-effects of that interaction. For example, it was
suggested, earlier, that styles in art may be the result of such
side-effects occurring in Martindale's account of aesthetic evolution.