state ‘answer’. Examples might include the manufacture, in a
large sense, of a dynamic product, e.g., a chemical substance,
anti-cancer or artificial immune search-and-destroy strategy,
biological signal detection/transduction process, and so on.
Tunable epigenetic catalysis lowers an ‘effective energy’ as-
sociated with the convergence of a highly coevolutionary cog-
nitive system to a final dynamic behavioral strategy. Given a
particular ‘farming’ information source acting as the program,
the behavior of the final state of interest will become associ-
ated with the lowest value of the free energy-analog, possibly
calculable by optimization methods. If the retina-like rate
distortion manifold has been properly implemented, a kind of
converse to the no free lunch theorem, then this optimization
procedure should converge to an appropriate solution, fixed or
dynamic. Thus we invoke a synergism between the focusing
theorem and a ‘tunable epigenetic catalysis theorem’ to raise
the probability of an acceptable solution, particularly for a
real-time system whose dynamics will be dominated by rate
distortion theorem constraints.
The degree of catalysis needed for convergence in a real time
system would seem critically dependent on the rate distortion
function R(D) or on its product with an acceptable reaction
time, τ , that is, on there being sufficient bandwidth in the
communication between a cognitive biological ‘machine’ and
its embedding environment. If that bandwidth is too limited,
or the available reaction time too short, then the system will
inevitably freeze out into what amounts to a highly dysfunc-
tional ‘ground state’. For cognitive systems, unfortunately,
absolute zero is all too attainable.
The essential point would seem to be a convergence between
emerging needs in biotechnology and general strategies for
programming coevolutionary computing devices.
17 Discussion and conclusions
We have hidden the kind of massive calculations made explicit
in the work of Ciliberti et al. and the Reinitz group, burying
them as ‘fitting regression-model analogs to data’, possibly
at a second order epigenetic hierarchical level. In the real
world such calculations would be quite difficult, particularly
given the introduction of punctuated transitions that must be
fitted using elaborate renormalization calculations, typically
requiring such exotic objects as Lambert W-functions (e.g.,
Wallace, 2005).
Analogies with neural network studies suggest, however, in-
tractable conceptual difficulties for spinglass-type models of
gene expression and development dynamics, much as claimed
by O’Nuallain (2008). In spite of nearly a century of sophisti-
cated neural network model studies - including elegant treat-
ments like Toulouse et al. (1986) - Atmanspacher (2006)
claims that to formulate a serious, clear-cut and transpar-
ent formal framework for cognitive neuroscience is a challenge
comparable to the early stage of physics four centuries ago.
Only a very few approaches, including that of Wallace (2005),
that are visible in contemporary literature are worth mention-
ing, in his view.
Furthermore, Krebs (2005) has identified what might well
be described as the sufficiency failing of neural network mod-
els, that is, neural networks can be constructed as Turing ma-
chines that can replicate any known dynamic behavior in the
same sense that the Ptolemaic Theory of planetary motion,
as a Fourier expansion in epicycles, can, to sufficient order,
mimic any observed orbit. Keplerian central motion provides
an essential reduction. Krebs’ particular characterization is
that ‘neural possibility is not neural plausibility’.
Likewise, Bennett and Hacker (2003) conclude that neural-
centered explanations of high order mental function commit
the mereological fallacy, that is, the fundamental logical error
of attributing what is in fact a property of an entirety to a
limited part of the whole system. ‘The brain’ does not exist in
isolation, but as part of a complete biological individual who
is most often deeply embedded in social and cultural contexts.
Neural network-like models of gene expression and devel-
opment applied to complex living things inherently commit
both errors, particularly in a social, cultural, or environmen-
tal milieu. This suggests a particular necessity for the formal
inclusion of the effects of embedding contexts - the epige-
netic Z and the environmental U - in the sense of Baars
(1988, 2005). That is, gene expression and development are
conditioned by internal and external signals from embedding
physiological, social, and for humans, cultural, environments.
As described above, our formulation can include such influ-
ences in a highly natural manner, as they influence epigenetic
catalysis. In addition, multiple, and quite different, cognitive
gene expression mechanisms may operate simultaneously, or
in appropriate sequence, given sufficient development time.
Although epigenetic catalysis, as we have explored it here,
might seem worthy of special focus, this would be a kind
of intellectual optical illusion akin to inattentional blindness.
Epigenetic catalysis is only one aspect of a general cognitive
paradigm for gene expression, and this larger, and very com-
plicated ‘perceptual field’ should remain the center of intel-
lectual attention, rather than any single element of that field.
This is to take, perhaps, an ‘East Asian’ rather than ‘Western’
perspective on the matter (Wallace, 2007).
Developmental disorders, in a broad sense that must in-
clude comorbid mental and physical characteristics, emerge
as pathological ‘resilience’ modes, in the sense of Wallace
(2008b), a viewpoint from ecosystem theory quite similar to
that of epigenetic epidemiology (e.g., Waterland and Michels,
2007; Foley et al., 2009). Environmental farming through
an embedding information source affecting internal epigenetic
regulation of gene expression, can, as a kind of programming
of a highly parallel cognitive system, place the organism into
a quasi-stable pathological developmental pattern converging
on a dysfunctional phenotype.
The probability models of cognitive process presented here
will lead, most fundamentally, to statistical models based on
the asymptotic limit theorems of information theory, in the
same sense that the usual parametric statistics are based on
the Central Limit Theorem. We have not, then, given ‘a’
model of development and its disorders in cognitive gene ex-
pression, but, rather, outlined a possible general strategy for
18