Information Technology Reference
In-Depth Information
6.3.6 Sieving and Intelligence - Evolving a List of Interesting Genes
As seen from the previous examples, finding sieves to select desired emergent
behaviors among large populations of cellular systems is a matter of educated
guess and experiment. It can be regarded as an intelligent process of interaction
between the human observer and the artificial entities (in our case the cellular
automata), mediated by a synthetic characterization of the CA behavior by a number
of complexity (or emergence) measures. Since algorithms are defined to compute
those measures for the entire family under observation (e.g. it requires no more than
20 h on a modest laptop of nowadays to compute the measures for the entire “2s9”
family) some educated guess in choosing the sieves provides shortlists of candi-
dates supposed to have “interesting” emergent behaviors. Such short lists can be
simulated and analyzed by the human observer in a reasonably short time providing
hints to further tune the sieves. This process is similar to the process of getting
acquainted to a new circumstance (e.g. meeting a new person or facing a novel life
situation) and is essential for what we call “intelligence”. The opposite kind of
action is to simply take a brute force approach, where each particular cell is taken
into consideration, simulated and observed. But this takes an enormous amount of
time, often not available in practice. That is why commonsense says that having
intelligence helps in solving a problem quickly. Is the solution optimal? Quite
often not, but it is better than what you may come up with after a dumb search
trough the whole space of possibilities. In our case, studying how a sieve in our
case acted with respect to a certain behavior, provides hints that gradually tune the
sieve to improve the optimality of the solution.
The above discussion reveals an aspect which expands beyond the scope of this
topic, where we are looking to find cellular automata with a certain behavior. The
conclusion is more general and can be applied to general systems as a process of
discovery : It can be summarized as follows:
Find a set of synthetic measures to characterize the observed system
(e.g. in real life problems our language facilitates descriptions about
humans like bad, good, etc. and defines degrees to which such qualities
are observed).
While enough variations of a certain system are presented, a database of
such measures is formed.
Given a desired behavior, define sieves starting from simple observations
of the relationships between quantifiable measures about the quality of the
object and observed behaviors of the object.
Apply the sieves and observe the match or mismatch of the resulted objects
with the desired behaviors. Whenever mismatches occur, tune the sieves
accordingly.
Coming back to our CA objects, we applied such a discovery process with
several (no more than four iterations in tuning the sieves) in an attempt to get the
most interesting CA cells from the “2s9” family. Such cells belong to the “intelli-
gent life” category and by declaring them interesting we suggest that such cellular
Search WWH ::




Custom Search