Information Technology Reference
In-Depth Information
Several models simulate the emergence of a scale-free frequency distribution rep-
resenting the size of innovations (S&V, A&P and SKIN). In the first two cases inno-
vation size is defined in terms of the number of technologies rendered obsolete or
becoming extinct. In the case of SKIN the distribution applies not to the size of new
kenes but to the size of new production networks [27]. L&F and A&P simulate sys-
tems that show progressive improvements, but with diminishing returns over time, as
a peak or optimum is reached in the L&F model and all of A&P's list of desired func-
tions become satisfied. SKIN, CJZ and the hypercycles model simulate the emergence
of networks of firms. Only CJZ specifically analyse the structure of these in their
paper, though the other models could certainly support analysis. But all three sets of
authors are interested in discovering the sensitivity of this emergence to various
parameter settings.
4
Towards Future Models of Innovation
Future models of innovation are likely to continue the concepts of innovation as heu-
ristic search and recombination of parts, and to continue to describe the emergent
networks and frequency distributions. As more data on industrial eco-systems, inno-
vation networks and other networks of firms become available, the structures of these
can be compared with the output from the models. Likewise, empirical frequency
distributions can be analysed to find out to what extent they tend towards being scale-
free, perhaps log-normal in shape, and with what parameters. Although some of the
above models could generate scale-free distributions of change sizes, familiar from
the theory of self-organised criticality [26], it is not yet known whether real-world
creative destruction follows this distribution of sizes. Stylised facts concerning geo-
metric growth in quality (better, faster, cheaper etc. - i.e. quantitative innovation) and
the immense growth in number of types of goods and services (qualitative innovation)
were not explained by the above models. Future models might address these and also
the scaling laws relating innovation rates (number of patents per capita, number of
entrepreneurs per capita) to city population sizes recently highlighted by Bettencourt,
Lobo, Helbing, Kuhnert and West [35]. The chapters and models in Lane [18] repre-
sent a first step towards meeting this latter end.
Empirical studies should also inform the inputs and design of these models. For
examples, fitness landscapes, the artificial chemistry of rules in the hypercycles
model, and the initial firm networks in L&F and the hypercycles models play impor-
tant roles in the models' behaviour, but abstract or arbitrary choices are currently used
for these inputs. Some of the models made a distinction between incremental and
radical innovation processes, as is common in innovation literature, but the further
identification of modular and architectural innovations [17] was missing. This will
require some thought as to how technology spaces and fitness landscapes should be
structured. In addition, agent learning behaviour and market mechanisms could also
be grounded in real-world cases.
