Complex Neuro-Cognitive Systems - Foundations on Natural and Artificial Computation - page 2

cases, however, the identification steps (2) and (3)have proved to be very dicult
or unfeasible. The bottom-up approach consists of describing the structural and
functional properties of given brain circuits, and then bringing this function into
congruence with the cognitive function under study.
Previously [1,2], I have shown that both approaches set up the method of
reverse engineering . This method combines analysis with synthesis in the follow-
ing way. Analysis is carried out top-down by specifying first a certain cognitive
function which is assumed to be computed through the cortex or some cortical
subsystem. Then a decompositional analysis is performed, i.e. the cortical sys-
tem is both functionally (computationally) and structurally decomposed, and
the interactions between components are determined. Following the localisation
concept, the functional components (computational units) are assigned to the
anatomical components.
Synthesis first requires modelling, i.e building a structurally adequate, func-
tional model of the computational unit. Based on knowledge of localised com-
ponents and their interactions, a structurally adequate network model of the
cortical system is build composed of the computational unit models. Simula-
tions of the network model should eventually prove that the specific cognitive
function under study is generated this way.
Recent efforts to build artificial brains 1 employ both approaches [6,7]. Large-
scale brain simulations attempt to model in a realistic fashion the details of the
brain organisation, i.e. its structure and function. The Blue Brain Project is one
prominent example of this bottom-up modelling strategy; its explicit goal is to
reverse engineer the brain. On the other hand, biologically inspired cognitive ar-
chitectures (BICAs) rely on the top-down approach. They attempt to achieve the
brain's cognitive functionality by emulating its high-level performance without
capturing the neural details.
In the survey [6,7] it was concluded that the two approaches display very dif-
ferent strengths. While bottom-up brain simulations are confined to syntactic as-
pects like how collections of neurons synchronize their electrical discharges, they
do not tell anything about semantics, i.e. how brain processes enable cognitive
agents to achieve goals, select actions or process information. In contrast, BICAs
propose how brains may realise cognitive functions, but as yet they demonstrate
rather simplistic behaviour compared to real brains. The authors conjecture that
the deficiencies may be due to the fact that ”BICAs lack the chaotic, complex
generativity that comes from neural nonlinear dynamics - i.e. they have the sen-
sible and brain-like higher-level structures, but lack the lower-level complexity
and emergence that one sees in large-scale brain simulations.” [7, p. 48]. Bringing
large-scale brain simulations and BICAs together, they suggest, will accomplish
progress toward the goals of cognitive science - understanding the brain and
creating artificial cognitive systems.
1 ”Artificial brain” is a term used to describe research that aims to develop software
and hardware with cognitive abilities similar to humans or other mammals. Promi-
nent examples are SyNAPSE (Systems of Neuromorphic Adaptive Plastic Scalable
Electronics)[3], the Blue Brain Project[4] and the China Brain Project [5].