Information Technology Reference
In-Depth Information
lead on to other technologies is relatively simple. In contrast, the relationships be-
tween combinations of bits in the L&F model and between complex technologies in
the A&P model form much more complex solution spaces / technology spaces.
Another view of novelty is as reinterpretation. In several of these models ideas and
technologies can acquire new functions or values due to the appearance or disappear-
ance of ideas and technologies. For example, a kene in the SKIN model can form an
input in many different innovation hypotheses, some of which may only appear later
in the simulation run. In the A&P model the discovery of a new node can connect
other technology nodes and shorten paths to the best-practice frontier. In the hypercy-
cles model the disappearance of a rule or firm alters the amount of production work
for other rules and firms.
The hypercycles model simulates the emergence of novel, identifiable self-
maintaining structures - a new type of thing with some significant degree of perma-
nence or stability. Thus, whereas some models represent the application of heuristic
search methods to combinations of pre-existing objects, models of emergent struc-
tures, such as hypercycles and auto-catalytic sets, have the potential to explain how
there come to be objects in the first place.
The models vary in the degree to which they represent open-ended systems. L&F's
model is relatively closed. Agents respond to their neighbours, while they differ in
beliefs, and to their fitness evaluations. The landscape is static and fixed at the begin-
ning, and does not respond to any events either inside or outside the model. Once a
population has converged on a single peak in the fitness landscape, the model's dy-
namics come to an end. Change in A&P's model could also come to an end, since the
list of desired functions is static, but in practice the list used by A&P is sufficiently
long for the phenomenon of technology extinctions to continue throughout the length
of their simulation runs. The models in both S&V and CJZ can permit indefinite ex-
tensions / additions of innovation, but S&V's percolation grid does not gain columns
and CJZ's knowledge vector does not gain dimensions, so in both cases new qualities
do not emerge. The SKIN model contains a limit on the number of products, but in
practice this limit need not restrict the dynamics of the model. Firms and their kenes
come and go throughout a simulation run. The hypercycles model reaches either a
dead state in which the firm-rule network becomes too fragmented and no more learn-
ing by doing can occur, or it forms a self-maintaining system of firm-rule hypercy-
cles. With no processes to introduce new firms or rules to the system, dead systems
cannot be resurrected and live systems are unable to change much except by a particu-
larly unlucky sequence of rule decay events. (Padgett and Powell [34] however con-
tains proposals for future extensions to the basic hypercycles model that will make it
more open-ended.)
Nearly all the models assume pre-defined structures. L&F assume a fitness land-
scape, for which they use Kauffman's NK fitness. S&V assume a network structure
for their percolation grid technology space. A&P assume their list of desired func-
tions. The hypercycles model assumes the initial structure for the network of firms,
the initial allocation of rules to firms, and the various types of rule themselves (the
“chemistry”). For this latter assumption Padgett's studies so far have restricted them-
selves to a simple cycle of rules, each rule having one input and one output product,
