210 CATALYZING INQUIRY
It is also interesting that biological function often relies on what might be called exploration with
selection—the production of many intermediate products resulting from stochastic subprocesses that
are then refined to unique and appropriate solutions.^11 Taken across the entire population, explora-
tion with selection exploits the difference between creating a solution and testing a solution for
correctness—the first being in general a much more difficult computational task than the second.^12
Random processes are used to explore the space of possible solutions,^13 and other machinery culls
these possible solutions. As Hartwell et al. argue, “Similar messy and probabilistic intermediates
appear in engineering systems based on artificial neural networks—mathematical characterizations
of information processing that are directly inspired by biology. A neural network can usefully de-
scribe complicated deterministic input-output relationships, even though the intermediate calcula-
tions through which it proceeds lack any obvious meaning and their choice depends on random noise
in a training process.”^14
6.2 AN ENGINEERING PERSPECTIVE ON BIOLOGICAL ORGANISMS
6.2.1 Biological Organisms as Engineered Entities
Engineering insights can be useful in understanding biological organisms as engineered entities,
and the rationale for seeking insights from engineering is based on three notions. First, although the
physical scales may differ in some cases, human technology and natural systems operate in the same
world and must obey the same physical rules. Knowledge that engineering fields have accumulated
about what techniques work and the limits of those techniques can serve as a potentially valuable guide
in investigating the physical basis of the operations of natural systems. This is especially true for
biomechanical feats, such as structural support, locomotion, circulation, and so on.
The second rationale is that because evolution and a long history of environmental accidents have
driven processes of natural selection, biological systems are more properly regarded as engineered
artifacts than as objects whose existence might be predicted on the basis of the first principles of physics,
although the evolutionary context means that an artifact is never “finished” and is rather evaluated on
a continuous basis.^15 Both engineered artifacts and biological organisms demonstrate function, embody
(^11) For example, the immune system relies on the random generation of pathogen detectors, which are then eliminated when
they match some definition of “self.” In single molecules, kinetic funnels direct different molecules of the same protein through
multiple, different paths from the denatured state to a unique folded structure (K.A. Dill and H.S. Chan, “From Levinthal to
Pathways to Funnels,” Nature Structural Biology 4:10-19, 1997). Within cells, the shape of the mitotic spindle is due partly to
selective stabilization of randomly generated microtubules whose ends happen to be close to a chromosome (R. Heald, R.
Tournebize, T. Blank, R. Sandaltzopoulos, P. Becker, A. Hyman, and E. Karsenti, “Self-organization of Microtubules into Bipolar
Spindles Around Artificial Chromosomes in Xenopus Egg Extracts,” Nature 382(6590):420-425, 1996). Within the brain, the pat-
terning of the nervous system is refined by the death of nerve cells and the decay of synapses that fail to connect to an appropri-
ate target.
(^12) This point can be formalized in the language of theoretical computer science. See J. Hartmanis, “Computational Complexity
and Mathematical Proofs,” pp. 251-256 in Informatics: 10 Years Back, 10 Years Ahead, 2000, Lecture Notes in Computer Science,
Springer-Verlag, Berlin, Heidelberg, 2001.
(^13) For example, random processes are at the heart of stochastic optimization methods that can be used for protein structure
prediction and receptor ligand docking, including simulated annealing, basin hopping, and parallel tempering. (An interesting
introduction to stochastic optimization methods can be found at W. Wenzel, “Stochastic Optimization Methods,” available at
http://iwrwww1.fzk.de/biostruct/Opti/opti.htm.)) Also, the systematic exploration of ecological models discussed in Section
5.4.8 is also based on the use of random processes.
(^14) The quote is taken from L.H. Hartwell, J.J. Hopfield, S. Leibler, and A.W. Murray, “From Molecular to Modular Cell Biol-
ogy,” Nature 402(6761 Suppl.):C47-C52, 1999. Hartwell et al. credit Sejnowski and Rosenberg with the neural network example
(T.J. Sejnowski and C.R. Rosenberg, “Parallel Networks That Learn to Pronounce English Text,” Complex Systems 1:145-168, 1987).
(^15) A classic paper on this subject is F. Jacob, “Evolution and Tinkering,” Science 196(4295):1161-1166, 1977.