Universal selection
The second Darwinian revolution
A second Darwinian revolution is underway - according to Gary Cziko.
In his book
Without Miracles [1]
the case is made that Darwinian evolution underpins a wide range of phenomena.
These include individual development, learning, cultural growth, science and
technology - and indeed ultimately all knowledge.
An iterated process of trial and error is invoked as underlying these
processes.
Variation and selection are seen as key phenomena by which these diverse
processes act.
Dissent
While I sympathise with some of the points expressed, I think the program takes
things too far - and is often more confusing than enlightening.
For one thing, I think that there are a wide range of different search
techniques, which can be used to expand knowledge bases.
Evolutionary search is one technique among many. The others include
simulated annealing, exhaustive search, gradient descent, Newton-Rhapson
- and so on.
To characterise all these approaches as ultimately Darwinian in nature seems
misleading and pointless - when they need not include any random elements that
might be analogous to mutation - do not involve a population of entities - and
need not have any sort of heritability or reproduction.
Also, there is no need to confine our attention to search techniques which are
so simple that they can be easily automated and performed by simple machines.
Intelligent agents can perform complex logical operations as part of their
searches - and are likely to search more effectivey as a result.
I don't think the process behind the development of knowledge would be
best characterised as a Darwinian one, underpinned by trial and error - rather
I see it as a process of iterated experimentation, and reasoning.
While it is true that the only way to expand an existing knowledge base
into areas where deductive reasoning cannot take it is to perform some sort of
trials or experiments a Darwinian explanation seems to fail to take into
account the rather significant role that reasoning can play in the process.
In general, the experimental results can be examined, hypothesis developed to
account for the results, and new hypotheses generated in order to
resolve remaining issues. The process requires intelligent agents to operate
effectively - and are only peripherally related to the process of natural
selection.
Records of old experiments can be kept - there need be no continual discarding
of previous results. The new experiments do not have to be based on old
experiments with variations - they can take any appropriate form.
There need be no equivalent to fitness - experiments are not evaluated and given
a success value which determines how much they influence future work.
Failed experiments can contribute as much to knowledge as successful ones. If
you have a failed experiment there is no need to discard it. You can examine it
in order to determine why it failed and build on that.
Evolution and selection are simply not appropriate metaphors when explaining the
growth of knowledge in fields like science or technology. Trial and error is
one element that is used - not the sole underlying principle.
A significant element of the process is logically reasoning out the consequences
of the new knowledge in the light of the existing knowledge base. This process
requires an intelligent agent to perform at all competently - and has no real
equivalent within a framework involving blind variation, selection and
reproduction.
Inversion
Rather than attempt to shoehorn science and technology into the straitjacket of
Darwinian selection, I would prefer that the reverse approach be taken -
namely that Darwinian selection should be seen as a backwards and degenerate
case of the search for knowledge - which exists because it is prebiotically
plausible, and can "get off the ground" easily.
Expanding an existing knowledge base is best done by intelligent agents - who
can iteratively design experiments, examine the results, deduce conclusions and
then formulate new experiments.
Classical Darwinian selection had to act in the absence of intelligent agents.
If you have to work from this impoverished state, blind variation, selection and
reproduction are some of the best tools available.
However, once intelligent agents come on the scene, traditional Darwinian
evolution and selection seem likely to be displaced by more effective
techniques.
Rather than random mutation being employed, directed variation can be employed.
An intelligent director has a much better chance of producing
interesting variation for selection to act on than a blindfolded agent does.
Experiments need not be regarded as successes or failures. Rather the results
of the experiments can be intelligently evaluated in the light of recorded past
results, hypotheses generated, and new experiments designed.
Past experiments influence future ones - but there is not really any direct
inheritance from one experiment to the next. The knowledge base is
preserved and transmitted to future generations - but a single knowledge base
can easily result in a vast range of completely-different looking experiments -
rather than a string of variations on the same theme.
Intelligent design in evolution
Over time, it seems likely that evolution will take on all the characteristics
of a huge engineering project. Intelligent design will be employed ubiquitously
- extending into every aspect of the process.
No doubt blind variation, selection and reproduction will still be employed in
some areas - but I expect that it will be regarded as a backwards approach -
what you're reduced to if you don't know enough about your problem to come up
with a more effective search technique.
If a revolution is currently underway I would have to classify it
as the revolution of intelligent design.
Not since the evolution of sex has such a significant means of recombing
existing knowledge into new forms come on the scene. It's effect on the
ecosphere seems likely to prove at least as dramatic as the Cambrian
explosion was.
References
- Cziko, Gary "Without Miracles" - Universal Selection Theory and the Second Darwinian Revolution - MIT press 1995.
|