Between
the years 2000-2003, I studied Environmental Biology at the
University of Wales, Bangor. For my final dissertation I decided
to write about evolution as a learning process. This was not
an easy task given that there was a 20 thousand word limit
as to what I could write. I found myself constantly going
over that limit. Consequently, at the time, there was much
I had perforce to omit or condense into 'verbal formulae'.
In this revision, however, I am free to reword certain passages
so as to make them flow more easily.
As I said at the start of the original dissertation, the key
ideas explored throughout I credit to an author by the name
of S.G.Powell. A psychology graduate and keen naturalist,
Powell's latest -and arguably most important- manuscript still
awaits publication. This is despite the fact that it has been
described by the ecologist Edward Goldsmith as "a treasure
trove of interesting material", as well as earning the
expressed adoration of, for example, the author Dorion Sagan.
I cite Powell wherever necessary, despite the fact that he
remains unpublished. However, much of what I have to say represents
my own 'fleshing-out' of the framework of his ideas, ideas
that were mostly derived from lengthy discussions whilst sat
around a camp fire, somewhere in wilderness. Despite these
reservations, it is still my hope that this document will
be acceptable and valid as well as interesting to read. Ideally
it will be inspiring.
"From approximately the same causes follow approximately
the same effects - in nature as well as in any good experiment.
If this were not so, we would not be able to ascertain any
natural laws or build any functioning machines" (Peitgen
et al, 1992).
[Image of female garden spider by Ronald
Patterson, 2001]
Contents
1.
Introduction
Raising the issue of evolution as a learning process.
3.
Principles and Mechanisms of Learning
"You can't learn anything unless you almost know it already."
Learning in artificial neural networks is homologous with
learning in evolution.
4.
Methinks it is Exactly Like a Weasel
A debate on the direction of evolution and the polarity of
certain characters.
Examples of fixed selection criteria at play in the ongoing
learning process.
5.
The Vine of Life
Set atop a sturdy, palpable physicochemical framework, life
evolves both directionally and non-directionally upon or across
that framework; life comes to mirror the framework upon which
it so elegantly sprawls. From where could the framework have
sprung?
1. Introduction If the common garden spider (Aranea diademata L)
can build a successful web without being 'taught' the necessary
skills via paternal or social interaction (Dawkins, 1996;
Purves et al, 1998), then evolution must equate to a form
of learning process (Powell, unpublished). Such a view explains
a great deal about life, simply by extending the scope of
learning beyond what is acquired within one lifetime, and
hence encompassing acquired biological function. For example,
as well as the skills necessary to build an intricate silken
web from special protein, a garden spider must also 'know',
without ever having been shown, how to digest its catch, how
to extract energy from it, and how to excrete waste products.
Our garden spider must also 'know' how to move across a heterogeneous
terrain, how to perceive and to recognize patterns - e.g.
potential enemies and mates, within that environment, and
hence to act in accordance with real-time gain of information.
No mean feat this, as any roboticist will admit. Most importantly,
via sexual reproduction our garden spider must inherently
'know' how to communicate such learning, let us call it 'genetic
learning', passing on what it 'knows' to more or less exact
copies of itself that can thence do the very same, and possibly
to an even greater standard. Regarding genetic learning in
garden spiders, Purves et al (1998) comment on how remarkable
it is that the genome of A. diademata, essentially
a chain of molecules with non-periodic crystallinity, can
come to store such a wealth of useful know-how.
Fig.1: An orb web of a garden
spider
you try making one at birth!
"All
animals learn" (Jablonka et al, 1998) within the span
of one lifetime. It is by learning that we Homo sapiens,
for example, come to talk, to read, to write, to ride a bicycle
or to operate a computer. But there are more fundamental layers
of learning besides those acquired within one lifetime. Those
deeper layers of learning have been, so to speak, deposited
over time and now constitute readily packaged metabolic and
functional know-how; of which, in a blurred and somewhat ill
defined manner, we might typically refer to as 'instinct'.
Because we did not personally have to learn mechanisms such
as peristalsis, blinking or coughing, we do not appreciate
such as learned. Nonetheless, within us mammals lies a stored
know-how which relates to the typical environment in which
that 'know-how' has been historically acquired.
To
take just one example, all mammals (and birds) thermoregulate
to maintain a relatively constant body temperature of around
37oC (or 42oC for birds). This attests to the fact that the
immediate environment fluctuates between hotter and colder
ambient temperatures. Of course, you and I know this fact,
but so does the region of the brain called the hypothalamus
- albeit indirectly and unconsciously.
Thermoregulation also attests to the fact that energy can
be gained and lost via conductance and radiation across a
given surface (depending on factors such as temperature, surface
area to volume ratio, thermal conductivity, and albedo of
the surface). Hence, heat gain or loss can be controlled via
special, dilatable blood capillaries traversing the dermis,
or by sweating or shivering, by insulation, as well as by
other behavioural means. As the Peitgen et al (1992) quote
cited above must convey, a physicist and a physiologist both
know these discernible facts vis a vis heat exchange, but
again so does the homeostat embedded within the mammalian/avian
body: Thermoregulation is a biological mechanism that has
evolved only because the world is so dependable. Evolution
has endowed mammals and birds with the knowledge of cause
and effect, which manifests itself as bio-logical function.
That
we do not regard thermoregulation as biological learning of
approximate cause and effect is partly due to current evolutionary
paradigms (discussed at some length below). That is to say,
our general scientific model or story of life and its worth
currently prevents or even denies such a perspective of physiological
function as acquired learning. Instead, science and the dogma
that it wields appears to be content to label biological function
as mere 'adaptation'. Indeed it might go so far as to say
that adaptations arose in relation to past environments, and
thus the term 'abaptations' 1,
denoting learning as some kind of biological luggage, would
therefore be more appropriate.
1a.
Everything Makes Sense of Something Leaving aside all notions of tense - past, present or
future, science has failed to appreciate how 'sensible' biological
adaptations are within the context in which they appear and
continue to exist. Given that energy and its transfer and
transformation will always observe an approximate cause-effect
relationship, the evolution of life has, so to speak, picked
up on that fact. Adaptations that arise in accordance with
the loss or gain of energy therefore make very good sense
of manifest physical effects - in the context of maintaining
enzyme functionality and cellular optimality. Consequently,
mechanisms could evolve.
The notion of adaptation as learned biological or behavioural
response to an essentially palpable, pattern-rich world will
form the central theme running throughout this dissertation.
As Powell suggests, evolution is akin to a 'classical conditioning
experiment', whereby nature teases life into performing certain
smart and relevant tricks, by virtue of its sensible, orderly
and hence 'meaningful' cause-effect setup (personal correspondence).
Order imparts meaning. For example, consider the effect that
a traffic light system, a subway or a wildlife underpass beneath
a busy road have on life. Their persistent orderliness ultimately
fosters meaningful behavioural responses. In the same way,
a chimp (i.e. Washoe) comes to employ sign language to interact
with its environment - a learned (and intelligent) response
in that it makes sense of the meaningful human contextual
experimental setup around it. Likewise life comes to employ
metabolism, photosynthesis, motility, and language - equally
learned (and intelligent) responses in that they too make
sense of the orderly and meaningful natural contextual milieu
around it. For life the reward is survival, continued autopoiesis,
being.
Organisms
that are 'better adapted' to their environment, as Darwin
argued, tend to leave more offspring. This is Nature's way
of saying that such 'better adapted' individuals somehow make
'better sense' of their meaningful cause-effect environment.
Such a better sense-making capacity is therefore preferentially
transmitted to future generations - either genetically or
epigenetically, as increasingly refined genetic learning or
as time-proven individual learning, respectively. We will
focus more closely on these issues later.
But
before we discuss possible mechanisms of evolution as a learning
process, as peculiar fruits of that process, firstly we must
better understand how evolution is currently portrayed (and
downplayed) throughout the scientific community that studies
it.
2.
Evolution as Delta f genes T-1, and that's it! Remarkably, Charles Darwin (1809-1882) was unfamiliar
with the mechanism of genetics - a mechanism that was nonetheless
discovered by a contemporary of his time, namely Gregor Mendel
(1822-1884). Nonetheless, from his numerous observations of
life, Darwin clearly sensed that the traits displayed by biological
organisms have a mutable and heritable nature. He knew that
all individuals vary from their parents to some degree, and
because of Malthus (1798) he understood that, at any given
moment, for any given population, the environment can only
support a certain number of individuals. These facts led Darwin
(1859) to argue that individuals whose traits were better
adapted to their environment stood a better chance of winning
what he termed "the struggle for survival". These
traits, these adaptations that are heritable and are correlated
with reproductive success pass preferentially from generation
to generation, and eventually they become common throughout
a population (or conversely, less 'fit' individuals leave
fewer offspring, and so their traits disappear from the population).
Darwin referred to the whole process of the transmission of
mutable, heritable traits via differential reproductive success
as 'decent with modification' by 'natural selection'.
Charles
Darwin
Evolution
is an unfinished business.
Our interpretation of evolution is also an unfinished business.
It
can be argued that, because Darwin had an incomplete knowledge
of the mechanism of mutable character transmission (i.e. genes),
rather than focusing on how selection quantitatively affects
a gene pool per se, he was free to focus more on the
concept of fitness.
"You can understand
fitness quite literally, as in a jigsaw piece fitting nicely
into place" Powell, 2007
The
question 'what is fitness' is of great importance to evolutionary
theorists, and yet it has a subtly different answer depending
on how evolution is defined. Later we shall come to see how
fitness can be defined given that evolution is a learning
process. It is to the contemporary definition of evolution
(and hence of fitness) that we shall now turn.
Simply
stated, evolution is currently defined as:
"a
change in the frequencies of genes in a gene pool over time,
and that's it"
(Prof. J.Wright, personal correspondence, echoing a widely
and warmly held view of evolution).
An
important point which must be noted at this early stage, is
this:
The
effect that natural (or neutral) selection has on a given
gene pool over time - namely that certain genes become more
common over time relative to others, rather than to serve
only as proof that evolution does occur, or to measure the
rate at which it does occur, instead has increasingly become
the defining principle of evolution:
just
mere change over time - and that's it!
2a.
Gene Salad
That there has been a change of gene frequencies over time
is not disputed. Arguably, there must be change (of some form)
for evolution (by any definition) to occur (Mameli, 2002).
However, such a quantitative definition of evolution does
not acknowledge all-important qualitative factors. That is
to say, such a 'numbers only' definition of the process that
sculpted extinct and extant biodiversity says nothing of the
direction that 'changes' in gene frequencies have taken or
might take in the immediate, foreseeable, or long-term future.
Likewise, by such a square definition nothing is said
of the relationship between 'genes' and 'groups of genes'
whose frequencies are or are not changing.
Analogously,
this entirely naïf contemporary definition of evolution
is akin to saying that the difference between a 1st edition
publication and a 6th edition publication is, ignoring all
editorial selective forces, simply a change in the frequency
of words over time, and that's it': just a casually tossed
word salad.
2b.
We're on a Road to Nowhere
Life on Earth has been subject to 3.9 billion years of 'changing
gene frequencies'. Accordingly, the incredible array of living
organisms only partially catalogued by science is merely the
result of considerable temporal change, and that's it. Regarding
the direction that such change may be taking, it is currently
held that evolution heads specifically nowhere; as Dawkins
(1991) asserts, mutable genes are pulled this way and then
that way, heading for anywhere but somewhere intentional,
chasing an ever-moving, quasi goal-like criteria or shifting
selective forces. (We shall return to this idea in section
4). The somewhat polarised debate as to whether most non-lethal
mutations are neutral (Kimura, 1975) or adaptive continues
to date. 'Genetic drift' that results from population sampling
error and base mutation rates in unconstrained parts of the
genome, are taken by neutralists to imply that the gene salad
is tossed almost entirely randomly - that is, relatively independent
of natural selection. Adaptionists, on the other hand, argue
that mutations change the sense in which genetic information
is read by the surrounding environment, and therefore most
or all non-lethal mutations must come under selective pressure.
Either way - natural selection or neutral selection, biodiversity
with untold novel and common characters is portrayed and accepted
pretty much as a purely unintentional state of affairs: just
mere, short-term, direction-less change.
Although
the concept of intention in evolution is negated, the idea
that some things may nonetheless be inevitable is considered
to be just unlikely. As the author Stephen J. Gould (2000)
argues, if we were to re-run the process of evolution starting,
for example, 600 million years ago - Precambrian era revisited,
life would probably 'change over time' in a wholly different
manner. Gould's main bone of contention is that 'there is
no guarantee humans would exist a second time around'. It
is perhaps naïve to expect that we humans should evolve
a second or third time around - or ever again, as though the
human biological design were somehow written into the fabric
of nature. But as we shall see later, perhaps some elements
or attributes of living organisms were/are inevitable, not
unlike stations on a train journey. For now it is enough to
realise that the tree of life, as depicted by leaders in their
scientific fields, is merely the result of change that branched
whither it would roam.
2c.
Selfish Genes
With scientific focus squarely on changes in gene frequencies
over time, it follows that genes are regarded as the focal
point of evolution. This focus on the smallest parts to explain
the whole process of evolution 3
has come about despite the fact that the process of natural
selection acts at a much higher level than the genetic material,
namely it acts on the phenotype - the expressed, material
organism. This orthodox view of evolution depicts the individual
organism as though a mere runner in a blind genetic baton
race; hence you and I, as phenotypes, have no consequence
in evolutionary terms save to affect the composition of the
gene pool from which we sprang: that is, to mate. We cannot
[naturally] alter the baton for there is no known mechanism
whereby the germ cell line can intentionally modify its own
genomic instructions. Accordingly, we are but passive gene
machines; a mass of supposedly selfish replicators; and nothing
else matters but survival to an age where sexual union is
attained just to pass on those selfish genes.
Chromosomal
DNA
Are we "nothing
more than the accumulation of its mistakes,
generated randomly and guided non-randomly by a blind watchmaker"?
Thanks
to everythingispointles for posting it on YouTube.
[As
the story goes, if one cannot pass on one's own genes it might
pay, in terms of reproductive fitness, to help one's kin pass
on genes that are statistically, probably closely related
to your own (Hamilton, 1964). Paradoxically, from such a purely
quantitative viewpoint of one's genetic composition, one's
relatedness to kin becomes diluted during subsequent sexual
events such that, after just two generations one is only r=0.25
related to one's grand-offspring. In terms of reproductive
fitness it supposedly would therefore not be worth sacrificing
one's own life just to save one dear grandchild. Beyond a
few generations, one's relatedness -and hence one's proud
genetic composition- are more or less dissipated into the
temporal realm for good. Hence, not only is the baton race
depicted as being blind in both eyes but also the baton falls
to pieces in the hands of statistically probable non-relatives,
over time 2.]
2d.
Mind Over Matters of Little Importance
It is ironic that such a reductionist view of evolution demotes
mind and its fruits (i.e. the capacity to read) to
a mere outcome of change over time, and that's it. Mind has
no intrinsic value in this peculiar scheme of things, save
that it helps to spread -or as implied, dissipate- one's genetic
composition, over time. From such a mechanistic and 'merelyistic'
3 perspective of the individual,
of mind, and of the process that supposedly 'willy nilly'
churned them both out, it then follows that all learning in
evolution (either genetic or epigenetic), the garnish on the
salad is merely a chance occurrence that helps to ensure that
these replicating nucleotide sequences are tossed into the
giant gene salad bowl of nature. Music, art, literature, philosophy,
religion and, indeed, science, these are all but mere epiphenomena;
interesting side-dishes resulting from change over time -
period. Society and culture are just elaborate doodling on
the canvas of history. All that matters is passing on your
genes - whatever those genes encode or regulate -or constitute-
en masse.
So
long as mutant alleles change ratios with wild alleles, this
is apparently sufficient for most to define the entire process
of evolution, rather than to serve as proof that something
inexplicable is afoot in the biosphere.
2e.
Salad or Science?
If what has just been said is an accurate depiction of evolution
and its astonishing fruits, then equally we should be able
to depict scientific advancement and its equally remarkable
fruits likewise. After all, science is a product of evolution
(however it is defined). Hence, let us ask the following question:
Is scientific advancement achieved via a mere change in the
frequencies of thoughts and ideas in a given culture, over
time? Perhaps a crab Louis Pasteur tossed by the chef of appointment
to the blind watchmaker? In short, the answer has to be 'not
at all'. Nobody would doubt that science has progressed over
time as a stepwise integration of often hard-won information.
The collective body of science has built upon discoveries
(i.e. of approximate cause/effect relationships) by modifying
ideas already existing in the pool of learning, as new information
becomes available. This is not to say that there is no change
in information vis a vis science, over time. Although such
a depiction of science is right, as Powell would say, "clearly
it is not right enough by half."
Just
as mutations happen randomly within a genome, most scientific
discoveries happen from chance (Rigden, 2002). However, chance
discoveries can only be understood and assimilated into the
body of science in the light of what is already known. In
a sense this is somewhat akin to a life history of science
imposing constraints on future discoveries. And the door swings
both ways too. Information can also be lost to science by
drift. Consider that Gregor Mendel's discovery of the mechanism
of genetics, established by him during years of plant breeding
experiments, these were almost lost were it not for a chance
discovery of his notebooks after his death. Such 'new sense'
that was made of the world of genetic inheritance led to many
new techniques and outcomes in science and in industry - just
as it leads to many new techniques and outcomes in biological
machines over evolutionary time.
For
now, let us inquire further into the true nature of change
over time by asking the following question: Does society
in any way act as a selective pressure during the process
of education? To this question we can answer a resounding
'yes'. At every stage of a child's education rewards are given
for tentative steps taken along a relatively well-defined
route. That is, through infancy to primary school, to college
and, on and up, to university level and beyond, to work, a
child's learning is forged via meaningful feedback from peers,
and is thus guided to a certain inner state with corresponding
outer manifestations by forces of selection. This is a form
of conditioning. A higher grade is a fitter grade and, ultimately,
a fitter grade is the first thing that any good employer selects
from variant applicants. Hence,
analogously, the modus operandi of the education system
and of evolution seems to be the selection of the fittest
- the term fit herein being meant in the sense
of a piece fitting nicely into place within a larger system.
As we shall see, in both cases the individual (either a person
in society or an organism in nature) is selected within -and
is thus molded into shape by- the entire meaningful context
surrounding and pervading it.
We
have considered the contemporary view of how genetic learning
arose (ie mere change), and have applied it to cultural learning
to assess whether or not that cultural learning could have
arisen according to the same freakish principles. Hopefully
we can agree that such mere change in information over time
smacks of absurdity vis a vis science or the education system.
Let us now attempt the converse by applying the principles
underlying learning in a cultural context to those underlying
the process of evolution, in order to determine whether genetic
learning could have arisen accordingly.
3.
Principles and Mechanisms of Learning
Martin's law states the following: "You can't learn anything
unless you almost know it already" (sourced from Winston,
1984). The implication of this apodictic statement lends itself
nicely to the idea that evolution is essentially a learning
process. The statement is self-evident in the wise old adage
"one cannot run until one has learned to walk."
Moreover, 'one cannot walk until one has learned to stand'.
Peeling away the layers of learning one by one, we can therefore
note also that 'one cannot stand until one has learned voluntary
muscle control'. If we continue this line of reasoning far
enough, and indeed we should, eventually we determine that
'one cannot learn to have voluntary muscle control until one
has learned to have muscles and a nervous system'. But before
any of those qualities can become, one simply must have some
kind of reason/advantage for such - else they simply would
not make sense. It is the context of survival in any given
environment that calls the shots, and it is bio-logic that
responds appropriately.
Jablonka
et al (1998) maintain that all animals learn.
The genetic recipe that underpins the capacity to learn (to
any degree) is passed on via DNA in all probability because
selection repeatedly favoured it. However, along with the
genes affording a capacity to learn, the interrelationship
between genes -as functional entities- is also passed along.
That is to say, just as a human child learns to associate
words into sentences, natural selection comes to associate
genes into functional groups - called gene complexes.
Within the developing organism, these gene complexes are jointly
expressed like words of a sensible -or not so sensible, context
dependent- sentence. Just as human skills are chosen by an
employer, not as separate abilities to move, to press a button,
to walk to the end of a machine, to fetch a product, and to
return to do the same over and over again, but as a collective
capacity to do a task. So, natural selection favours collective
genetic capacities that are expressed phenotypically, not
individual genes per se. Groups of genes equate to
integrated learning, and biological organisms can be seen
as avatars (Gliddon et al, 1989) for that learning as it manifests
itself via the phenotype, throughout evolutionary time. Genetic
learning is hence not just genetic sequences but is also the
relationship between genes, the manner in which they jointly
unfold a developing organism, and the manner in which they
jointly regenerate a mature avatar that can respond to a world-context
that is essentially written all over that avatar as well as
imprinted genetically within it.
Everything
about an avatar says something about the world in which it
lives, in the same sense that everything about a graduate
student says something about the culture in which they live.
Their degree says everything about the world they live in,
and in turn their world says everything about the natural
laws that underpin and fashion that world.
So,
how does the world find it's way into the genome and it's
avatar?
3a.
Epigenetic Cognition
Learning is essentially a process of 'coming to know' via
experience. Over time, the information gained from experience
is stored, imprinted and integrated by association: this is
a ball; this is an apple; do not eat the ball. Similarly,
this is a worm; this is blade of straw; do not build a nest
from worms. Unlike a computer, which stores information in
discrete locations in memory blocks, the associative models
stored by the brain (and as we shall see later, by artificial
neural networks) are dispersed throughout entire localised
nets (Gurney, 2002). However, before we come to see possible
mechanisms by which a neural network is said to learn, and
hence before we can relate this process to evolution, we firstly
need to recap a few fundamental aspects of how neurons function
in groups.
Without
going too far into detail, recall that a neuron with perhaps
hundreds of individual dendritic inputs sums up electrical
signals [called spikes] as they arrive at what is called the
axon hillock, which is situated in the main body of a neuron.
Due to differences in operational speed evident between various
chemical neurotransmitters at synaptic clefts, some action
potentials arrive late at the hillock (Purves et al, 1998).
But there is a degree of temporal summation, meaning that
relatively late signals can still be included and hence can
influence the chain of events that leads to a physiological
or behavioural response.
3b.
Trains of Thought
To borrow a very good analogy from Powell (unpublished), which
illustrates the workings of adjoining neurons remarkably well,
the summation process itself can be thought of as a train
waiting at a platform (a hillock) for passengers to arrive.
However, the train will only depart from the platform if =n
passengers are onboard. If <n passengers are onboard, the
train does not depart, and the passengers simply disperse.
Hence, the train is cancelled: a big zero somewhere down the
line. On the other hand, if the train has =n passengers onboard,
it departs from the platform, careering off along the course
of the axon body at speeds of up to 100 meters per second.
Eventually it arrives at a synaptic junction (a station),
where the passengers disembark the train, commute via vesicles
(or vehicles, perhaps along escalators) to another platform
(another hillock) and the entire process is repeated. Interestingly,
as is typical in real life, an axon hillock receives a frequency
of action potentials, or passengers, some arriving early and
some rather delayed in their journey, perhaps having to run
for the train. But many arrive en masse - which can be thought
of as rush hour or peak-time traffic. How all of this commuting
to and throw via trains and escalators actually relates to
learning in a real neural network - i.e. in a human being,
for example with many billions of synapses, is still very
much an area of active neurophysiological research. However,
from the groundwork of artificial neural networks, referred
to as ANN's, certain principles can me made all the more clear
for our purpose of demonstrating that evolution is an information-gaining,
storing, utilising, learning process, and not just a mere
change in gene frequencies over time.
Needless
to say, due to their artificial nature, ANN's have been simplified,
as models that emulate what are believed to be the essential
functioning of real neural networks. We can't learn anything
unless we almost know it already, and hence "making computers
learn and artificially intelligent helps us to understand
intelligence" (Winston, 1984) and the learning process
itself. Also, compared to the difficulties of accessing and
observing how living neurons interact, ANN's are very handy
things.
Schematic
of a simple artificial neural network
What I wish to demonstrate throughout
the remainder of this section is that neurons become adjusted
over time to give an output in an homologous way to how gene
frequencies become adjusted in a gene pool over time. Both
systems (neural networks and biological populations) are repeatedly
exposed to patterns and, via a feedback mechanism - by a process
of selection, something becomes adjusted to give an apt output.
So, please bear with the rather heavy technical jargon in
the next few paragraphs because, as I see it, there is something
very important in this idea.
The
neurocomputing term used to describe variation in signal delay
is S.T.D.P., which means signal transduction delay plasticity.
It is this essential plasticity, the fact that signals can
arrive faster or slower at the axon hillock to affect afferent
output, that is believed to play an essential role in learning.
In ANN's, temporally variant input signals can be given a
specific 'weight' (see the 'crossed circles' in Figure.2 below).
This means, for example, an input value of 1 arriving at a
node (the name given to an artificial neuron) might be weighted
as 0.5, such that, even though the input signal carries the
binary value of 1 it will be pre-summed as 1 x 0.5 = 0.5 (see
Fig.2). Similarly, a binary input of 0 from a preceding node
might be weighted at the next node as 1.2 and hence pre-summed
as 0 x 1.2 = 0. Summed together at the virtual hillock, our
two example signal inputs therefore amount to 0.5. If the
firing threshold of the virtual hillock equals -or is greater
than- 0.6, for example, this means that the threshold has
not been reached and therefore the node outputs an afferent
binary 0.
It
is easy to see that, by changing the pre-summing 'weights'
of input signals, different outputs can be obtained using
the same binary input - i.e. from sensors such as cameras
(or eyes or antennae, etc). The output '1' may trigger a different
response than the output '0'. If we imagine this essential
process taking place between, say, one thousand individual-collective
nodes, we can appreciate that there will be considerable learning
power and flexibility at hand, since only two neurons are
responsible for the sense of touch and muscle retraction in
the marine mussel Mytilus edulis (Purves et al, 1998).
3c.
Perceiving Patterns
An ANN cannot recognise a pattern, for example an apple, until
it has been repeatedly shown and told what an apple is, just
as an infant human. This is an apple; it has a more or less
spherical shape; it is green or red (in digital RedGreenBlue
terms) or patches of both. Seen from the top it is more spherical
than when seen from the side. Given sensory input, an ANN
gradually learns to recognise an apple (from different angles)
by adjusting internal input weights between nodes, thus altering
the summation process throughout the entire network (essentially
as a real neural network might do), and consequently altering
the response output. To see how this can possibly relate to
evolution as a learning process, consider:
"The
goal of a neural network's learning procedure is to maximize
the mutual information between its output and input"
(Chechik, 2001).
In
other words, input weights are adjusted until a binary output
signal from an 'output layer' (see Fig.1), as observed by
a programmer, is affirmed as being 'correct' or 'desirable'
or 'mutual with input' or, in a nutshell, as being 'fit'.
That is to say, the feedback provided by the technician to
the net (hence acting as an agent of selection) gradually
teases 'pre-summation weight' adjustments until a collective
state of weights is achieved, stored, and hence learned -
or for our purpose, evolved. Thereafter, the neural net can
discriminate between an apple and modus ponens everything
that is not an apple.
In the above instance, the programmer eventually imparts to
the ANN a basic recognition of a particular palpable pattern
called 'an apple'. But the pattern could be something far
more complex. To give just two short examples of their usefulness
to society, intelligent ANN's are currently in use by the
American National Cancer Research Institute to help diagnose
patients for prostrate and ovarian cancer. Having been fed
information about blood serum proteins obtained by mass spectrometry,
an ANN learned how to distinguish between differences in disease-diagnosed
and disease-free blood serum patterns, and then communicated
the information to a computer that was left to diagnose new
patients - with evident success (sourced from www.AAAI.com).
HNN's, hybrid neural networks, are currently being implemented
for their flexible expertise in helping to design and control
efficient constructed wetland wastewater treatment systems
(i.e. see Pastor et al, 2003). Factors such as inflow and
outflow BOD (biological oxygen demand), pH and temperature
are input into the ANN, which then calculates first order
kinetic constants and adjusts the pH or temperature of the
constructed wetland accordingly to attain maximum reaction
efficiency for a minimum given area of land usage. The HNN
learns the relationship between approximate cause and approximate
effect - how change in one variable, say temperature, affects
other variables, for example outflow BOD. Control and function
of the loop of Henle in the mammalian kidney is principally
identical.
So,
we have seen that ANN's 'learn something' over time by being
repeatedly exposed to a pattern (i.e. a blood serum protein),
being told via selective feedback (a technician) whether the
response (for example displaying the name of the protein on
a screen) is mutual with the input pattern, and hence
gradually adjusting 'internal weights' (pre-summation weights)
until the output is indeed mutual with, or apt or corresponding
to...
Let
us simply change a few words in the above formula, to get
the following:
...a
population 'learns something' over evolutionary time by being
repeatedly exposed to a pattern (i.e. seasonal changes), being
told via selective feedback (nature) whether the response
(i.e. fur, hibernation) is mutual with the input pattern,
and hence gradually adjusting 'internal weights' (genes, gene
complexes) until the output is indeed mutual with, or apt
or corresponding to...
In
other words, learning in ANN's and the process of evolution
are homologous one to another.
3d.
Learning in Natural Intelligence Systems
For living organisms, as members of a population, a variation
in 'almost knowing it already' impacts on survival, and hence
comes under the watchful eye of selection. Those organisms
whom 'almost know it - only more mutually', have a greater
chance of survival to pass on more copies of themselves (with
variation). The population gene pool becomes adjusted over
time with genes that permit a more mutual knowing. The process
repeats itself whilst retaining that which has hitherto made
sense of a physicochemical and biological sea of palpable
patterns we call the environment.
In
this sense, then, we can define fitness as follows:
Fitness
is the degree to which phenotypic output is mutual with all
environmental input
It
stands to reason that if a system such as an artificial neural
network is capable of learning in such a manner and is deemed
to be intelligent (even artificially so), likewise a system
constructed by nature that is capable of learning in principally
the same manner can also be regarded as an intelligent system:
a non-artificial or natural intelligence system (Powell, unpublished).
[Consider for example the immuno-response system.] This is
a moot point only because intelligence is so very difficult
to de-anthropomorphise in order to obtain a working definition.
Given that at a fundamental level intelligence relates to
gaining, storing, and using information and making sense,
however, a natural corollary to artificial intelligence is
plausible in natural systems. Moreover, intelligence is a
process. Clearly evolution is an information gaining, storing
and utilising (and hence is a naturally intelligent) process
that fosters bio-logic.
This
is a good point to take a closer look at what exactly life
and it's evolution is learning all about in such a naturally
intelligent manner.
4.
Methinks it is Exactly Like a Weasel My original intention was to demonstrate natural learning
in botanical ontogenetic processes. I had planned to mention
that Hyasinthoides nonscripta, the common British Bluebell,
extends retractable roots into the ground - roots that are
effectively botanical arms that serve to pull the starch-rich
bulb from which they sprang further into the nurturing soil.
It then occurred to me that such 'learning' has already been
fully documented in every major botanical volume. I would
simply be wasting my time by describing such learning where
so many authors have already done so, albeit unwittingly.
One has only to pick up a book about biology to read about
genetic learning; science almost knows this already but to
see it as such requires a subtle paradigm shift. Therefore,
instead, I will continue by focusing on the issue of how evolution
accrues learning via repeated pattern exposure and feedback
upon output that is more or less mutual to input.
The
author and zoologist, Richard Dawkins was perhaps entirely
misguided to suggest that there are no fixed criteria in nature.
Take Dawkins (1991) classic experiment "METHINKS IT IS
LIKE A WEASEL", in which a sensible phrase was evolved
from an initial 28 random letters via the cumulative selection
of mutations. Note that Dawkins used a fixed target phrase
to achieve such evolution - and moreover a phrase that made
sense, only then to dismiss altogether the idea of fixed targets
in nature. One can only guess at the nonsensical outcome should
the experiment be repeated, this time with a target phrase
that was allowed to wander whither it would roam, and did
not make any sense to start with. How Dawkins got away with
that one we shall perhaps never know. But clearly he is mistaken.
And I shall demonstrate this using several good examples from
the natural world. Take
for example the evolution of plant xylem tracheids and vessel
elements.
4a.
Water
In order to transport water from the soil to the growing canopy,
a tree must take into account the physical and chemical properties
of water, for example
cohesive strength (relating to bonding between like-molecules),
and adhesive strength (relating to bonding between unlike-molecules).
As
water is drawn up through a tube using suction, for example,
the water molecules maintain bonds between themselves that
are strong enough (in terms of tensile strength) to form an
unbroken column stretching from one end of the network of
tubes to the other - depending on the diameter of the tubes.
If the tube is relatively too wide, although the rate of flow
is greater, the column of water more readily breaks (referred
to as cavitation, where air comes out of solution under tension)
(Salisbury and Ross, 1992). If cavitation occurs, vapour pockets
form in the tracheid, causing an embolism by blocking water
transport. For a Douglas fir, for example, which can grow
to a height of 119 meters, natural selection has sculpted
xylem tracheids with lignified secondary cell walls (with
very high compressional strength to avoid collapse under pressure)
as well as spiral thickenings (like a vacuum cleaner hose
designed to avoid collapse of the pipe - e.g. when the nozzle
is blocked). As these cells form, differentiate and die, they
form tiny valve-like pits that allow water to continue to
flow from the cohesive column into adjacent tracheids. Each
of these adaptations is specifically related to the behaviour
of water and cells under pressure.
Sir
David Attenborough highlighting the Natural Intelligence in
Plant Water transpiration.
Thanks to phytoman007 for posting it
on YouTube.
Remarkably,
trees rely on the atmosphere to do all of the work of pulling
the column upwards. Flowing from greater to lesser negative
potential, as though from wetter to drier areas, water is
drawn up through the tracheid drinking straws running the
entire length of a tree by virtue of the dryness of the atmosphere
surrounding its leaves. The evaporative loss of water from
stomates on the under surface of leaves renders their water
potential more negative (or drier) in relation to the (wetter)
mesophyll tissues surrounding them. This in turn draws in
water from the (wetter) xylem vessels serving the leaves to
the (drier) tissues surrounding the stomates. As a result
the entire water column moves steadily upward inside the xylem
tracheids, an unbroken column stretching from the stomates
in the sky to the fine root hairs busily divining its presence
deep within the soil. Roots often extend deep into the soil
profile because, as an input rule, that is where water tends
to be found, and hence that is where a botanic lineage has
been taught by cumulative selection of phenotypic output to
seek for it. The entire water transport mechanism -from roots
to stomates- makes absolute sense of water and how it unfailingly
behaves in the environment. It is incredible natural technology.
Thus,
what Dawkins (1991) asserts by stating that only short-term
selection criteria are at work in nature is, that, plant xylem
tracheid architecture has been pulled willy nilly from an
initial random generation of, for example, 28 genetic units,
to a point where those units eventually spell out the following
phrase:
"METHINKS
LIFE (R)EVOLVES AROUND WATER AND NOT VICE VERSA"
(see for example Margulis, 1995)
Because
the properties of water do not change - will never ever change,
Dawkins' assumption that only short-term goals are chased
in evolution is perhaps non-applicable to xylem architecture
and to plant transpiration streams in general. Evolution by
cumulative selection has steered xylem architecture until
phenotypic output is mutual with contextual input
- simultaneously taking into account perhaps innumerable other
sources of input, such as: climate and probable range of relative
atmospheric humidity, patterns of rainfall and groundwater
availability (field capacity), soil structure (that we can
classify soils implies that they have a pattern), soil nutrient
status, redox potential and pH, cell osmotic pressure, as
well as other patterned environmental factors such as atmospheric
[CO2]. The result is an astonishing example of evolution as
the prime learning process.
The
physicochemical behaviour of water can either be seen as an
'environmental constraint' on life, as some scientists may
dismissingly regard it, or as an attractor point around which
evolution gravitates. Much more could be said about water
(i.e. that multicellular organisms maintain an inner intercellular
ocean echoing the marine origin of life), but space permitting,
such is outside the scope of this dissertation.
4b.
Metabolism
Another important factor which life simply had/has to learn
about is chemical potential energy. ATP, the universal cellular
energy currency, as it is referred to and as that term suggests,
serves as an agent for energy transactions within all living
cells. The specific molecular structure of ATP and its dephosphorylated
form, ADP, are such that metabolism is everywhere and everywhen
geared entirely around these pair (as well as other Lego brick-like
components - e.g. NADP+) (see Fig.4 below). Consider:
"The
interplay between [several alternate photophosporylation]
pathways [i.e. cyclic, pseudo-cyclic, and non-cyclic] might
explain the flexibility of photosynthesis in meeting different
metabolic demands for ATP (Allen, 2003).
…the
key words in the above quote being that of the flexibility
of photosynthesis toward a specific end. Hence, again, starting
with a randomly chosen flexible-mutable string of nucleotide
sequences, unfixed selection criteria can supposedly, gradually
come to spell out the following phrase:
"METHINKS
THE LAWS OF CHEMISTRY ARE IMMUTABLE WHEREAS GENOMES ARE NOT"
One
has only to ask if short-term selective criteria could maintain
this universal biochemical aspect underpinning all of life
over an estimated 3.9 billion years. Indeed, how short is
short-term? How many genes are involved in metabolic processes
that are fixed - in terms of the components they must unfailingly
make sense of?
Continuing
with our account of evolution as a learning process, and simultaneously
challenging Dawkin's idea that no fixed criteria exist in
nature/evolution, heterotrophs perforce evolved output
metabolic pathways that became mutual with the palpable input
that is chemical properties of autotrophic storage products,
such as glucose and starch. Likewise, chemotrophic bacteria
must evolve output metabolic pathways mutual with, for example,
input pyrite mineral properties or other law-conformable sulphur-based
compounds. Metabolism gravitates around specific electron
acceptors (H2, H2S, NH3, CH4, Fe2+) (Margulis and Sagan, 1995)
whose properties are fixed (fixed with a capital F). Again
such a fact can be written off as an environmental constraint
on metabolism, which is typical of some scientific attitudes.
But herein it is presented positively as life making sense
of chemical potential energy in the context of an ingrained
striving toward survival and autopoiesis, via long-term cumulative
selection.
The
generation of energy has much to do with the next environmental
aspect which life perforce had to learn about: gravity.
4c.
Gravity
It is remarkable that a fundamental force keeps bodies with
specific mass held firmly -but not too firmly- together. The
important point to make about gravity in this context is that
it is a constant. Natural selection, as we have seen regarding
water chemistry and metabolism, fosters mutual output, steering
life toward mutuality, always getting closer over time. Again,
if we take Dawkins gene jumbler example, starting with a given
number of randomly generated nucleotides, Dawkins would assert
that, in relation to gravity, a given random sequence could
come to spell: I can walk on water - if I run fast enough;
or, I can fly; or, deposit feeding as a way of life, depending
on the given niche:
I
Can Walk On Water - If I Run Fast Enough!
Common pond skaters or striders (Gerris lacustris)
manage to stay afloat on water because they distribute their
relatively negligible mass over an area that is large enough
to prevent any one point breaking the surface tension of the
water. Evolution of output characteristics in the Gerris
genera has gravitated toward mutuality regarding aquatic environmental
input. It is not perfect; rather it is mutual with. However,
the Jesus Lizard (Basiliscus vittatus), so called because
it can walk -or rather it can run- across water, achieves
the seemingly miraculous despite its relatively bulky mass
(see video stream below). B. vittatus traps air bubbles
within special folds of skin beneath its feet. Slow-motion
photography has shown how a lizard rears up on its hind legs
and moves them in a 'free-wheeling' motion whilst moving across
water - relative to the pull of gravity. Its legs free wheel
so rapidly that its feet are not in contact with the water
for long enough to break the surface tension (sourced from
www.fusionanomaly.net). Hence, there is something in the information-rich
genome of the Jesus lizard that takes into account such tangible,
palpable, environmental factors, learned via evolutionary
processes.
A great video clip of a Jesus Lizard
running across water.
Thanks to dmac202124 for posting it
on YouTube.
To
say that running on water has just happened over time is certainly
not right enough by half. Natural selection is weighing a
lineage within a given context, and by trial and error that
lineage will acquire the necessary learning to enable apt
behaviour. As Jablonka et al (1998) argue, it was perhaps
new learning that initiated the phenomenon of running on water,
not mutation per se. It is very unlikely that a gene
or a group of genes mutated to allow running on water. All
animals learn, and hence may come to associate 'faster and
faster speed' with 'greater and greater buoyancy' thus 'distance
traversed on water'. Hence, this view of how running on water
originated is altogether more plausible - given the ubiquity
of water, the need to escape predators or to catch prey items,
and that all animals learn. As R.F. Ewer puts it, "behaviour
is always a jump ahead of structure" (sourced from Thorpe,
1978, p33). In other words, new behaviour from new learning
leads the way whilst the genome follows in its wake until
'assimilated' (Jablonka et al, 1998). [In essence this is
Baldwinian
Evolution - which I think deserves far more attention]
I
Can Fly
Just as the JC lizard lineage literally learnt to run on water
over time, likewise "bird lineages literally learnt to
fly over time" (Powell, unpublished). By flapping their
[wings] whilst trying to scale obstacles during escape they
perhaps achieved a very apt uplift and hence a selective advantage
(referred to as the cursorial theory of flight evolution).
The lightest birds with the greatest uplift tended to live
to tell the tale, genetically and epigenetically. Again, new
learning carries around old learning until old learning is
worn nicely into shape by the (selective) friction of inevitable
change. Perhaps others simply glided into selectively advantageous
safety (referred to as the arboreal theory of flight evolution)
(see for example Henderström, 2002). Over time, avian
physical attributes (i.e. weight, shape) as well as behavioural
attributes (i.e. diet) have centered around the gravitational
constant, making sense of its ever-reliable approximate cause-effect
on matter.
Deposit
Feeding as a Way of Life
Perhaps it may sound trivially obvious to say, it is no coincidence
that deposit feeders are found on the ocean floor. Dead and
decaying plankton and the remains of other marine organisms
continually rains down from the productive, light-rich photic
zone to the benthic zone below, therefore constituting a pattern-rich
input derived entirely from this fundamental law affecting
matter. Recall that artificial intelligence systems require
repeated exposure to a pattern whilst being informed via feedback
that their output is mutual with an input pattern. Benthic
deposit feeder lineages have been exposed to the pattern-like
rain of nutrient-rich debris over vast periods of time and,
via natural selection, genetic weighting and its corresponding
output as function-behaviour now mirrors, or is mutual with,
the palpability of gravitational pull and the patterns it
unfailingly induces. Eversible stomachs (e.g. in echinoderms)
and mucous threads (e.g. in vermetid gastropod molluscs),
as just two examples, make good sense given the predictable
deposition of food from above and the protein-rich microorganisms
attracted to it (Barnes et al, 2001). Such is evolution as
a learning process; such is manifest classical biological
conditioning.
It
can be argued that statocysts have evolved to make sense of
the force of gravity in the context of the need for orientation;
and where burrowing makes such good survival sense if you
can tell where is down and where is up. Skeletal and muscular
traits in larger animals mirror the prevailing force of gravity.
Tarlo (1964) proposed that bone (ossified tissue) originally
evolved not for protection or support, but as a reserve for
phosphate and calcium - both of which fluctuate in availability
in the environment (sourced from: Carroll, 1988). Only later
was bone advantageous as the function for which we now more
readily associate it with. In all cases, the evolution of
bone (and cartilage - and indeed chitin and lignin) makes
sense of something. Gravity is a long-term selective criteria
that impinges heavily on physical and behavioural characters
and, via evolution, life comes to make better sense of it
over time.
4d.
Light
I now wish to move on to another example of how Richard Dawkins
demonstrates evolution using fixed criteria, but then denies
fixed criteria in nature. In this instance he demonstrates
the evolution of a simple eye from a flat bilayer of cells,
using virtual evolution software and virtual light.
The
Virtual Evolution of a Simple Eye
Note how the light does not change throughout
the simulation, such that the evolved virtual eye
gradually begins to make more and more sense of the fixed
properties of that light.
Thanks
to everythingispointles for posting it on YouTube.
Without
going into details or indeed repeating the account made by
Dawkins, but instead asking that the reader perform a thought
experiment, I can thence make my point vis a vis the inevitable
evolution of an organ of vision which embodies real learning
of the orderly principles pertaining to light. Let the thought
experiment begin.
Imagine
if you will that the virtual milieu (e.g. a computer running
windows and a virtual evolution package) is all set up, the
bilayer of cells are programmed into place, and the source
of light begins to shine out. However, unlike 'normal' light,
here the given virtual light in our experiment behaves entirely
capriciously - very unbecoming of electromagnetic radiation.
Firstly, occasionally-very-often, the virtual light does not
travel in a straight line and hence cannot readily be focused
to a point. Secondly, the energy imparted via a single photon
is randomly inclined to be not inversely related to its wavelength,
and hence the virtual light may behave like regions of the
EM spectrum either side of the visible (380-750nm) spectrum.
In other words, our wiggly light may often be microwave- or
ultraviolet-, or even xray-like...
The
point of the experiment is, then, to determine what might
happen -in an evolutionary sense- to the virtual bilayer of
cells where, in this instance, the selective criteria is not
fixed (i.e. as Dawkins strongly asserts as being the case
throughout all of nature but contradicts by way of his examples).
For example, what effect would such 'lawless light' have on
the refractive index of the fledgling protoeye? On pigment
evolution and its functional effect on action spectrums? On
the sensitivity of the photoreceptors to dangerously high
or perhaps lethal random frequencies of EM radiation?
Thankfully,
as Peitgen et al (1992) noted at the very start of this dissertation,
from approximately the same physical cause of electromagnetic
radiation follows approximately the same physical effect of
light such that functional [biological] machines can evolve
from its tangible nature - that is, because of its dependable
orderliness. Because light travels in a straight line, and
because light from all possible sources can simultaneously
occupy a single point in space in front of an eye - because
light carries information so faithfully, the refractive index
of an eye (a measure of how well light is focused to a single
point) is therefore a good gauge of just how fundamental long-term
selective criteria are for life as we know it.
The
eye has evolved (or rather, how to use light to advantage,
has been learned) independently some 42 times in separate
lineages (Krebs and Davies, 2002); if one includes photosensitivity
in microorganisms, the obvious advantage of such learning,
of such skill (Blackmore, 2002) is beyond dispute. Eyes are
selectively advantageous information-gaining appendages, as
are all sensory appendages, as is life itself.
4e.
Co-evolution
Although Dawkins was not right enough about the direction
of evolution - of learning, of the inevitable polarity of
certain characters, he was right enough in some respects.
There are also what can be termed 'short-term' goals in evolution,
of course fully reliant upon the necessary long-term learning
underpinning life (regarding which we shall return to in section
5). For example co-evolution between predators and prey can
be regarded as 'short-term'. Certain interrelationships can
change rapidly over time and morphological forms can be seen
to counter each other - which is depicted as an 'arms race'.
Recall once more that learning in artificial intelligence
systems is feedback-derived, such that output (function-behaviour
- i.e. running fast) gradually becomes more and more 'in mutual
agreeance' with input (i.e. the presence of fast-running predators).
(Note that one does not make sense without the other.) For
example, slow running Oryx (Oryx gazella) make easier
prey items for female Lions (Panthera leo) than do
faster Oryx. If and when an Oryx is caught, that is natural
selection's way of feeding back to the Oryx gene pool that
such output (slow running, or perhaps a lack of perceptiveness)
is not mutual with input (fast hungry predators); the gene
pool 'weights' adjust correspondingly as changes in gene frequencies.
Conversely, Oryx as potential prey that repeatedly escape
predation for long enough to reproduce offspring that likewise
survive to reproduce, thus behaving and functioning in a way
that 'makes sense' within that context, that in turn is natural
selection's way of feeding back that such 'output is mutual
to input'; again the weighted gene pool adjusts to the new
patterns around it, like a distillation or concentration of
aptness. Such is evolution as the prime learning process.
Consider:
"The
method of learning by the selection of successes from among
many acts is the most fundamental method of learning"
(Thorndike, 1901, p38).
Function
and behaviour that are successful, mutual with input, apt
under the circumstances, are sensible in the light of context,
spread by virtue of survival advantage; like weights in an
ANN; like beneficial ideas spreading throughout a society;
like skills in a workplace. CO-evolution is one particular
example of biology learning all about, or making sense of,
its tangible, palpable environment, and transmitting that
sense via the mechanism of reproduction or social interaction.
There is no 'war', only manifest sense making from chance,
necessity, and intelligibility, that makes life collectively
such a robust phenomenon. Of course there are limits to the
speed at which an animal can run; to how fast oxygen can be
circulated to tissues. Hence, life learns other strategies,
for example camouflage.
Summary
So far we have considered how truly remarkable it is that
certain skills can be passed on via hereditary chemical molecules,
and hence independently of social learning (i.e. web-building
in Aranea diademata L). By classifying the storage
and transmission of such skills as 'genetic learning', we
then asked how such genetic learning could arise - given the
current 'wanting' definition of evolution. Having discussed
how genetic learning is as much to do with the interrelationship
between groups of genes as it is with gene sequences, we then
discussed some principles of social learning in general and
specifically in artificial intelligence systems (whose output
gradually becomes mutual to input via selection/feedback).
At that point we noted neat parallels or homologies with the
process of evolution itself.
Thus
we were able to see how the world gets into the avatar as
well as plausibly into the genome. Section 4 built upon this
very subtle perspective of 'evolution as a learning process'
by demonstrating that biological functions and behaviours
perforce gravitate toward mutuality in the same way that ANN's
gravitate toward mutuality. We then discussed how some functions
and behaviours by necessity gravitate toward short-term pattern
mutuality. Now we shall discuss some of the main points that
have been covered already - in the light of an orderly, meaningful
world context.
5.
The Vine of Life
"Is
there a [tree] of life? The answer seems to be yes"
(Baldauf, 2002).
Our
garden spider sprawled across the front page of this dissertation
is a female preparing to spin an egg case into which she will
deposit all that she knows, genetically. The symbol of an
orb web constructed between structural supports such as rocks,
branches, garden fences and washing lines, also symbolises
or captures in its essence the very nature of the tree of
life. As we saw in section 4, biological metabolism is reliant
upon an orderly molecular structural support for its very
existence. In turn, molecules -as the product of the periodic
table of elements- are reliant upon known fundamental forces
of nature: the strong and weak nuclear forces, the electromagnetic
force, electrostatic attraction, and gravity. It does appear
that everything rests upon something else, from silken webs
to physiological function all the way down to the fundamental
level of subatomic particles obeying immutable laws. Take
away the support and the edifice crumbles. It is in this sense
that Powell (personal correspondence) refers to life not as
a tree but as a vine, since a vine needs a support
up or over which to develop. One has only to visualise Hedera
helix entwining its way to the very top of a temperate
forest canopy in close association with its lignified structural
support to get a sense of what Powell refers to as 'life coming
to mirror or reflect its framework'; taking on its shape or
character. Alternatively, one can visualise life as a fluidic
mass or substance slowly pouring itself across a molded surface
that becomes increasingly intricate and detailed over time,
and that life is therefore a negative casting, shaped according
to the detail and character of the positive molded surface.
In
the same way that goldfish foraging behaviour becomes classically
conditioned to an orderly hence meaningful experimental framework
that encourages association between coloured spatial cues
and food availability (Hughes), the vine of life becomes classically
conditioned to an orderly hence meaningful framework that
encourages association between environmental patterns and
survival, with being. In order to understand this idea -of
life reflecting its framework, taking on its character- more
clearly, we will now turn to the subject of virtual evolution.
In the same way that virtual learning in ANN's has demonstrated
to us that evolution can be regarded as a learning process,
virtual evolution can likewise serve to highlight the significance
of an orderly framework for evolution.
5a.
Virtual Entities
Just as Dawkins (1996) used virtual evolution software to
evolve a virtual eye, Powell (unpublished) proposed that virtual
entities could be evolved with the capacity to predict the
next number in a given numerical sequence. These virtual entities
"…can be thought of as a domesticated form of computer
virus that lives in, and adapts to, a controlled environment"
(Wilke and Adami, 2002). Powell envisaged the entities comprising
of strings of mutable computer code analogous to genetic information.
Each replication event could impose a degree of variation
in code sequence, thus simulating biological mutation and
DNA replication error.
He
proposed that they could be rewarded differently for predicting
numbers accurately or not so accurately, for example by being
allowed to reproduce differentially. If an entity failed to
predict at all after, say, 5 iterations of the same numerical
rule, it simply died. However, Powell proposed to start the
evolutionary simulation by repeatedly exposing the virtual
entities to a sequence of randomly generated numbers between
1 and 100, for example:
2, 1,
13, 86, 10, 4, 41, 12, 17, 78, 14, ?
Hence,
an entity has just 5 attempts to predict the next number in
the sequence, to a degree of accuracy, to make sense of the
inherent pattern - if any. Clearly there is no underlying
rule that can be deduced in the sequence since each number
is generated purely by random. It can be said that there is
no possible strategy that could evolve in relation to predicting
random numbers, no 'sense' to be made here. If perchance a
virtual entity did manage to guess relatively correctly and
hence to reproduce, no heritable advantage would be transmitted
to its offspring simply because it was purely a lucky guess,
a fluke, carrying a probability of 1/100. As Powell indicates,
the offspring of the entity would be "none the wiser"
vis a vis a capacity for prediction and hence acquiring a
survival advantage. Therefore, evolution would not be possible
in such a randomly generated, nonsensical world. There might
still be 'mere change' in frequencies of virtual gene sequences,
however, but unless there was sufficient order from which
to gain a survival advantage, a framework across which to
sprawl, to pick up on, to learn about, life would soon run
out of luck. This is an important point.
As
a next step, Powell proposed that the sequence of numbers
between 1 and 100 could contain at least a measure of orderliness
and hence discernible meaning, for example:
18, 92,
36, 20, 2, 8, 74, 88, 86, 44, 12, ?
By
generating a numerical sequence between 1 and 100 featuring
only even numbers, interesting things can begin to happen.
Already there is an underlying rule to pick up on - at first
by chance, such that entities whom willy nilly preferentially
guess even numbers more often than odd numbers stand a better
chance of predicting correctly, and hence have a slight selective
advantage. Consequently, having reproduced, they would pass
to their offspring the same preference for guessing even numbers
more frequently than odd numbers. As Powell notes, if we reiterate
this experiment enough times, gradually the composition of
the virtual code pool would begin to shift toward entities
that preferentially guessed only even numbers. But as Powell
also notes, such a preference toward guessing only even numbers
is the finite limit to evolution in this particular experimental
world. That is, even though the sequence is no longer generated
purely randomly, and simple evolution can occur, nothing else
at all can happen but a preference for guessing in one direction
rather than another. As with the above scenario, like a game
of roulette, luck would soon run out and virtual life would
probably come to an end.
We
can say something about genetic drift at this point. It is
possible that two relatively different virtual genetic sequences
'do the same job' of guessing, and therefore selection does
not care, as Dawkins would say. Also, entities that were fully
capable of guessing could nonetheless be subject to accidents
or catastrophes, and hence be removed from the virtual world,
perhaps before a reproduction event. This would remove 'fit'
genes from the pool, altering the ratio of fit:unfit genes.
There are no guarantees in life vis a vis survival. This is
what makes a biosphere so remarkable - so astonishing.
As
a further step, Powell proposed that the numerical sequence
could contain still more given order, for example:
2, 4,
8, 10, 14, 16, 20, 22, 26, 28, ?
5b. 30 or 32?
"Pattern
recognition is an activity which enters into almost every
walk of life. One could say that our entire ability to survive
depends on recognition of patterns" (Aleksander,
1978).
As
more and more order -and hence inherent, discernible meaning-
is introduced into the all-important survival-dependent sequence
of numbers, character polarity (the observable direction that
a character takes over evolutionary time) becomes more and
more marked - given the power of cumulative selection acting
from a fixed selection criteria. What we can note in this
experiment is, that, entities whom initially preferentially
guess even numbers more often than odd numbers will again
have a survival advantage. They will again tend to reproduce
more often and hence will transmit the same preference to
their offspring - with variation. By reiterating this experiment
many times over, unlike in previous virtual worlds, there
will be a definite trend toward selection of entities that
guess only even numbers running in an ascending numerical
order (as per above). Eventually an entity may come to 'crack'
the code-like sequence, so to speak, by accurately predicting
the next number - rather than by simply guessing it. (Note
that this entire process is essentially identical to how an
ANN -an intelligent neural network- comes to exhibit output
that is mutual to input - that is, to learn) When this stage
in the virtual evolution experiment is reached, the extant
entities can be said to have made 'maximum sense' of the numerical
sequence of numbers - they have learned the underlying rules
at play.
The
experiment could continue further, becoming more and more
complex, such that two or three or more sequences were generated
simultaneously. The entities would therefore have to crack
several sequences in order to survive and reproduce. For example,
cracking a second sequence could engender an 'added energy'
aspect, such that they had extra attempts at guessing. Again,
evolution could only occur if one or all sequences were 'crackable',
were able to be made sense of, carried about them an approximate
orderliness.
Hence,
we have seen how a purely random, nonsensical world does not
permit evolution - it only permits guessing that carries no
adaptive advantage. (In fact it permits nothing at all because
the ability 'to guess' requires an organ capable of real-time
gain of information and subsequent processing that itself
must evolve in -and be a biological response to- a non-random
environment.) We have seen how a relatively nonsensical world
with a measure of order permits only preferential guessing
before very probable extinction. But most importantly, we
have seen how an increasingly sensible, orderly world permits
correct prediction and hence is supportive of virtual life
and its further evolution. This is precisely because there
is sense to be made in such a world. If we again bear in mind
Peitgen et al (1992), and their assertion that machines can
only be evolved from approximate effects that follow approximate
causes, clearly a number-predicting machine (i.e. a brain)
is only evolvable in a world comprised of orderly numerical
stimuli. Now it is time to exit the virtual world and step
back into the real world, into the sea of palpable patterns.
5c.
On the Origin of The Underlying Rules at Play
According to Einstein the most incomprehensible thing about
the Universe is its mysterious comprehensibility:
The
world is awash with orderly numerical sequences, so to speak,
such that there are libraries filled with volumes describing
order and pattern in great detail - not just verbally but
mathematically. Indeed, science can be regarded as 'the study
of pattern and its ultimate and proximate causation'. When
for example we consider avian migration and mammalian hibernation
in the context of an orderly, seasonal cycle, we find that
real world entities have very much 'cracked the code-like
sequences' evident approximately everywhere in nature. Such
is manifest classical conditioning - a genetic and epigenetic
awareness of pendulum-like cycles. This is not to say that
such awareness is conscious; nonetheless it is a form of awareness
- a natural awareness, if you will. For example, Palolo worms
synchronise their mass spawning events with the cycle of the
full moon (Stearns and Hoekstra, 2000), the orderly celestial
motion of the moon resulting from gravitational attraction
with the earth thus acts as a dependable cue to present (and
future) reproductive events (we can visualise virtual Palolo
entities correctly predicting celestial sequences in an orderly
virtual world; or left to guess in a roulette world; as non-starters
in a nonsensical world). Wet seasons and dry seasons also
exhibit cyclic patterns. Although they are not as predictable
as the motion of celestial bodies, we can nonetheless come
to know that as a rule one always follows the other, for example
as the strategies of desert annual plants attest.
If
we consider again the virtual evolution experiments described
above, bearing in mind that survival and evolution were only
possible given that order and meaning had to be introduced
into the virtual system, then, we must ask where such order
came from? For the virtual world inside the computer, that
order came from human input. Here we are treading on very
heretical ground. However, considering the 'hot big bang'
circa 10-20 bya, Thorpe (1978) contends "it now appears
that the essential nature and development of the universe
must have been determined (perhaps 'programmed' is not too
strong a word) during the first micro-seconds of this cataclysmic
event" (his inclusion, not mine). Hence, we cannot continue
to study the world resulting from the realisation of inherent
potential without asking from ultimately where sprang that
potential.
We
have discussed how biological function is reliant upon molecular
order. In turn we have discussed how molecular order is reliant
upon atomic and subatomic order. Likewise (sub)atomic order
is reliant upon the fundamental natural laws. Equally, therefore,
we should be able to state -one way or the other- whether
or not fundamental laws derive from an orderly source (Powell,
personal correspondence). We could approach the issue by saying
that there is no reason to believe that 'the underlying rules
at play' derive from anything other than an orderly source
- given that all other forms and degrees of order emerge from
previous order. "Order begets order" (Powell, unpublished,
echoing prevalent systemic thinking). Until scientists come
to ask the right questions, however, an answer as to the origin
of order must remain 'intractably metaphysical'.
6.
Discussion
This view of evolution as a learning process, although brought
very much into focus here, nonetheless is not altogether new.
John Corlis, author of 'Life is a Strange Attractor', for
example, comes very close in principle to this new paradigm
by explicitly stating that 'life is a learning process in
the context of an evolving Universe.' For Corlis, life emerged
as a subsystem nestling within -and simulating, in terms of
chemical processes- a larger system - an Achaean submarine
hot spring, circa at least 3.9 billions years ago. Life quite
possibly began in such thermophilic style and diversified
out into mesophilic and psychrophilic forms. However and wherever
life did start, it eventually encountered earth's heterogeneity.
Hence, wherever live ventured from its place(s) of origin
it encountered variations in the patterns prevailing around
it, of which it was exposed to; places where the framework
differed either subtly or profoundly, the intricately molded
surface rising or falling. Plenty of scope for learning and
mirroring the framework, and inevitable opportunities for
changes in genetic composition to accumulate in relative isolation.
As I have been at pains to point out, cause/effect relationships
(e.g. the compression and rarefaction of molecules causing
vibrational waves to spread out from an epicenter, or the
fact that atoms are compelled to join and interact to form
functional groups), such enables science to assemble functional
machines (i.e. seismometers and mass spectrometers, respectively).
More importantly, in this context, such also enables evolution
to assemble functional biological machines (e.g. mechano-sensors/ears
and the Calvin cycle, respectively). Scientific machines are
essentially the product of the discovery of certain principles;
science picks up on how phenomenon interrelate and, using
the sense made from such elucidation, furthers its understanding
of the world; machines and regimes evolve by trial and error,
becoming more proficient over time. Dawkins' virtual fish
eye picked up on how light behaves; it evolved, becoming more
visually proficient over time. Consequently, so did our understanding
of evolution by selection. In the context of survival, evolution
equally 'picks up' on cause/effect relationships, how phenomenon
interrelate, as life is exposed to persistent environmental
patterns over time. Hence evolution came to instill learning
about those persistent environmental patterns by selecting
from among a pool of individual subsystems only those simulations
of the larger system that made appropriate sense, e.g. those
that have survived, reproduced, staved of entropy against
all odds, for example, simply because they could detect differences
in concentrations of ambient chemicals and move toward or
away from the source better than others could. Consider:
"Evolution
is a learning process or, as Margulis and Sagan put it, evolution
is good at solving the problems of survival, of sense-making,
the solutions being written in robust and heritable DNA. That
which lives is that which makes sense. If an organism cannot
make sense then it is out of luck and out of life"
(Powell, unpublished manuscript).
To
view evolution as mere change in gene frequencies over time
is to view ANN presummation adjustment as a mere change in
weights over time - in ignorance of the processes affecting
those weights; in ignorance of the sense they collectively
make of the patterns around them whilst generating an output
display; and hence in ignorance of the fact that by definition
an ANN is learning from experience. No mere change in gene
frequencies over time in the process of evolution itself;
there is a direction, a form of intelligence underlying it,
albeit one that we do not yet fully understand or even fully
recognise. Sadly, we are happy to deny it outright whilst
slowly dismantling the fruits of that process.
How we see life affects how we live and treat life. Currently
we treat it as a pretty pointless fluke for our perusal, only
now making scant economically viable efforts to preserve that
much deemed necessary for our own biophilic happiness (see
for example E. O. Wilson, the Biophilia Hypothesis). Our ecological
footprint is currently far greater than 1 Earth. If we were
to perceive the biosphere as a pool of learning, potentially
useful learning at our disposal, we may tread more carefully
and respectfully through time. If we were to realise that
evolution was an intelligent process cradled and nourished
by nature, who knows where that may lead?
Footnotes
1:
Abaptation: "…organisms appear to be adapted (fitted)
to their present environment only because present environments
tend to be similar to past environments (which is my point
exactly, evolution is learning of persistent patterns). The
argument is, the word adaptation gives an erroneous
impression of prediction (see 5b-5c above), of forethought,
or at the very least, of design. Adherents claim that organisms
are not designed for, or adapted to, the present or the future
- they are consequences of, and therefore abapted by, their
past" (Begon et al, 1990, p8). My argument is that organisms
are neither adapted to nor abapted to, rather they become
mutual with.
2:
Quantitative relatedness: one must ask how related is one
sperm cell to another if they both derive from the same 95%
homozygous 2n organism? Hamilton (1964) would say 0.5 but
I say there is error in such a quantitative approach. They
could only be 0.5 related to each other if the 2n organism
was heterozygous across each and every loci, and the two sperm
cells happened to get completely different alleles.
3:
Merelyism: Powell (unpublished) refers to 'merelyism' as "the
currently fashionable paradigm which seeks to explain life
and evolution by appealing to life's smallest parts - i.e.
genes", adding that "details can blind one to higher
level meanings and higher level patterns". Such a fashionable
paradigm is not wrong per se, rather "it is not
right enough" because "the dogma fails to capture
the full brilliance of evolution, which is more apparent when
evolution is viewed under more holistic magnification."
A 'merelyist' "is in danger of explaining life away by
reducing everything of apparent significance to merely this
or merely that". For example, mind is merely a by-product
of brain function. Consider that Purves et al (2000) warn
the reader not to be too astonished by the apparent design
of living things because "astonishment is a subjective
human emotion", as though nature were sometimes "a
bit sneaky", and tended to sucker unsuspecting intelligent
individuals into an unbecoming state of reverence toward nature.
An apt example of 'merelyism' is the reference to the design
of living organisms as 'designoid', as though natural design
and technology (i.e. Buggi, on page 2) were somehow not sculpted
to functional mutuality by any form of feedback, or somehow
fell short of an (equally subjective) engineering standard
(e.g. the Beagle2 Mars rover).
7. References
Peitgen
H, Jürgens H & Saupe D
(1992) Chaos and Fractals: New Frontiers of Science. Springer-Verlag,
ISBN 0-387-97903-4.
Dawkins
R
(1996) Climbing Mount Improbable. Penguin Books, ISBN 0-670-85018-7.
Purves
W, Orians G, Heller H & Sadava D
(1998) Life: The Science of Biology, 5th Ed. Freeman &
Co, ISBN 0-7167-3325-0.
Powell
SG
Natural Intelligence: New Perspectives on Evolution. Unpublished
Manuscript.
[Now called: Darwin's Unfinished
Business]
Jablonka
E, Lamb M & Avital E
(1998) Lamarckian Mechanisms in Darwinian Evolution. Trends
in Ecology & Evolution, Vol 13, 5, May.
Begon
M, Harper J & Townsend C
(1990) Ecology: Individuals, Populations and Communities.
Blackwell Science, ISBN 0-632-02344-9.
Malthus
TR
(1798) An Essay on the Principle of Population. Oxford World's
Classics.
Darwin
C
(1859) The Origin of Species.
Mameli
M
(2002) Learning, Evolution, and the Icing on the Cake. Biology
and Philosophy 17: 141-153.
Dawkins
R
(1991) The Blind Watchmaker. Penguin Books, ISBN 0-14-01441-1.
Kimura
M
(1994) Population Genetics, Molecular Evolution, and the Neutral
Theory. Edited by Takahata N. ISBN 0-226-43563-6
Gould
SJ
(2000) Wonderful Life: The Burgess Shale and the Nature of
History. Vintage, ISBN 0-099-2735-6.
Hamilton
WD
(1964) The Genetic Theory of Social Behaviour. Journal of
Theoretical Biology, 7, 17-32.
Rigden
J
(2002) Hydrogen: The Essential Element. Harvard University
Press, ISBN 0-674-00738-7.
Winston
P
(1984) Artificial Intelligence: 2nd Ed. Addison-Wesley Publication
Co, ISBN 0-201-08259-4,.
Gliddon
CJ & Gouyon PH
(1989) The Genetics of Information and the Evolution of Avatars.
Edited by de Jong G. Population Genetics and Evolution.
Gurney
K
(2002) An Introduction to Neural Networks. Routledge, ISBN
1-85728-503-4.
Chechik
G
(2001) Spike Timing-dependent Plasticity and Mutual Information
in Spiking Neurons. Neurocomputing, 38: 147-152, June.
Pastor
R, Benqlilou C, Paz D, Cardenas G, Espuna A, & Puigjaner
L
(2003) Design Optimisation of Constructed Wetlands for Wastewater
Treatment. Resource, Conservation & Recycling, 37, 193-204.
Salisbury
F & Ross C
(1992) Plant Physiology: 4th Ed. Wadsworth Publishing, ISBN
0-534-15162-0.
Margulis
L & Sagan D
(1995) What is Life. University of California Press, ISBN
0-520-22021-8.
Thorpe
WH
(1978) Purpose in a World of Chance: A Biologist's Aiew. Oxford
University Press, ISBN 0-19-217654-4.
Allen
JF
(2003) Cyclic, Pseudo-cyclic and Non-cyclic Photophosphorylation:
New Links in the Chain. Trends in Plant Sciences, Vol8, No1,
January.
Henderström
A
(2002) Aerodynamics, Evolution and Ecology of Avian Flight.
Trends in Evolution and Ecology, Vol 17, No 9, September.
Barnes
R, Calow P, Olive P, Golding D & Spicer J
(2001) The Invertebrates: A Synthesis, 3rd Ed. Blackwell Science,
ISBN 0-632-04761-5.
Carroll
R
(1988) Vertebrate Paleontology and Evolution. Freeman &
Co, ISBN 0-7167-1822-7.
Blackmore
S
(2002) Crossing the Chasm of Consciousness. Trends in Cognitive
Sciences, Vol 6, No7, July.
Catania
AC
(1999) Thorndike's Legacy: Learning, Selection, and the Law
of Effect. Journal of the Experimental Analysis of Behaviour,
Vol 72, No 3, 425-428, November.
Krebs
J & Davies N
(2001) Behavioural Ecology: 4th Ed. Blackwell Science, ISBN
0-86542-731-3.
Baldauf
SL
(2002) The Tree of Life is a Tree (More or Less). Trends in
Ecology and Evolution, Vol 17, No 10, October.
Wilke
C & Adami C
(2002) The Biology of Digital Organisms. Trends in Ecology
and Evolution, Vol 17, No 11, November.
Aleksander
I
(1978) The Human Machine: A View of Intelligent Mechanisms.
Georgi Publishing Co, ISBN 2-604-00023-7.
Stearns
S & Hoekstra R
(2000) Evolution: An Introduction. Oxford University Press,
ISBN 0-19-854968-7.
Corlis
JB
Life is a Strange Attractor: The Emergence of Life in Archaean
Submarine Hot Springs.
