Genes, Brains, and Human Potential The Science and Ideology of Intelligence

A CREATIVE COGNITION 205

executive is, exactly, and how it works, is another prob lem usually side

stepped. It is like another little brain put within the brain, rather like the

one that is implicitly put in a gene. As mentioned in chapter 6, there is no

evidence for such a creature in real brains.

However, these liberties over the nature of inputs, predetermined con-

nections, and central executive have been confronted since the 1980s by a

rival version of associationism. Th is approach sets up a layer of pro cessing

units in a computer. Th e units are programmed to be sensitive to experi-

enced inputs. Th ey then send outputs to a further layer of units, usually

in a one- to- all or at least one- to- many correspondence, to integrate the

signals (fi gure 7.1). Th e trick is that these connections can then be

strengthened or weighted by the programmer according to the frequency

with which signals co- occur.

Th e result is that patterns of co- occurring inputs can then be regis-

tered in the interconnections among the units. Th is is considered to

refl ect what happens in real brains (so they have been called “artifi cial

neural networks”). Moreover, the pro cess captures not only the simple

pairwise associations. Th e deeper statistical structure of the inputs to

input

units

ac

e

d¾

c¾

a 2

b 2

a 1

bdb^1

hidden

units

output
unit
FIGURE 7.1
Fragment of an artifi cial neural network. Stimuli (usually some well- defi ned features,
as conceived by the experimenter) are input at the keyboard and received by the input
units. Th ese are passed to “hidden” (association) units. Aft er integration of a series of
signals, the outputs of those should refl ect statistical associations among them. +/− sig-
nifi es features pres ent/absent and c~/d~^ represents their integration (or probability of
co- occurrence).

This content downloaded from 139.184.14.159 on Tue, 17 Oct 2017 13:54:24 UTC