|
|
1.Model Architecture
Sim-I-Space is a multi-agent simulation characterized by mixture of competition and
collaboration. Although built in part on a Swarm platform, in the limited number of time
periods it runs, and given that agents make their decisions at random – i.e., they do not learn -
it only exhibits limited elements of evolutionary behavior. Survival is the aim of individual
agents in the simulation. The rents that agents earn provide them with the means to survive.
If agents run out of money they are ‘cropped’. They can quit while they are still ahead. The
overall value of a given simulation run is the sum of the rents earned by all agents during the
run. The social welfare generated by the simulation is sum total of all knowledge created in
the course of the simulation and then diffused out to ‘society’. Note that ‘society’ is located
outside the simulation. The price paid by ‘society’ for this social welfare is the cumulative
rent earned by the agents inside the simulation – ie, the price paid by ‘society’ turns out to be
the value of the simulation.
How does Sim- I-Space implement and embody the concepts of the I-Space? The I-Space is a
conceptua l framework for analyzing the nature of information flows between agents as a
function of how far such flows have been structured through processes of codification and
3
abstraction. Such flows, over time, give rise to the creation and exchange of knowledge
assets. Where given types of exchange are recurrent, they will form transactional patterns
that can be institutionalized. In Sim- I-Space, we focus on the creation and exchange of
knowledge assets alone without concerning ourselves with the phenomenon of recurrence
and institutionalization. In later versions of the model, recurrence will become our central
concern.
Sim-I-Space is populated with agents that carry knowledge assets in their heads. Each of
these knowledge assets has a location in the I-Space that changes over time as a function of
diffusion and obsolescence processes as well as of what agents decide to do with them. These
have the possibility of exchanging their knowledge assets in whole or in part with other
agents through different types of dealing arrangements.
Natural selection is at work in Sim-I-Space at two levels. At one level, agents survive by
learning to make good use of their knowledge assets. They can make use of these assets
directly to earn rents, or they can make indirect use of these assets by entering into trades
with other agents who will then use them directly. Agents that fail to make good direct or
indirect use of such assets in a timely fashion fail to earn the minimum rent required to
survive and are selected out of the simulation – i.e., they are “cropped”. At another level,
knowledge assets, in turn, and somewhat like Dawkins’ “memes’ (Dawkins, 1982), survive
by inhabiting the heads of many agents. If they fail to occupy at least one agent’s head, they
die out and the knowledge associated with the asset disappears from the simulation as a
resource.
Existing agents have the option of quitting the simulation while they are ahead and before
they are cropped. Conversely, new agents can be drawn into the simulation if the
environment becomes sufficiently rich in opportunities for earning rents.
The rate of entry and exit of new agents into the simulation are based on the difference in
mean rents between between two periods. The rate of entry and exit is a parameter that is set
at the beginning of the simulation for every % change in mean rents. Change in the entry and
exit rates is a function of % change in mean rents. In this way one can control the level of
4
market turbulence – of creative destruction, if you will – that is generated by the performance
of existing players.
We start by discussing the agents and then turn to a discussion of the nature of their
knowledge assets. This is followed by a brief discussion of agent interactions.
1.1 Agents:
Sim-I-Space operates through a number of agents that, taken together, make up the diffusion
dimension of the I-Space. In the model as developed, agents are intended to represent
organizations – firms or other types of information-driven organizations – within an
industrial sector. It would be quite feasible, however, with suitable parameter settings, to
have the agents represent individual employees within a single firm and hence to simulate the
behavior of individual organizations. It would also be possible to have an individua l agent
representing the behavior of a strategic business unit within a single firm. Conversely, one
could run Sim-I-Space above the firm level and simulate knowledge flows within a
population of industries.
As we have already seen, agents can enter or exit Sim-I-Space according to circumstances
and can also be cropped from the simulation if their performance falls below a certain
threshold. Agent entry and exit is an important source of variation within the simulation.
Clearly, the population that is located along the diffusion dimension of the I-Space will vary
in size at different moments in the simulation.
Agents aim to survive within the simulation and to maximize their wealth over the periods of
the simulation. Wealth here is taken to be the sum of rent streams and of rent-generating
knowledge assets. The first accumulate in a financial fund that is used to cover the expenses
incurred in meeting and transacting with other agents. The second accumulates in an
experience fund that is used to finance the creation of new knowledge assets by moving
existing ones in the I-Space. In sum, agents modify their wealth either by trading in
knowledge assets they possess with other agents thereby enlarging or shrinking their asset
base, or by creating new knowledge assets. They do this by moving around the I-Space in a
learning process and by adding and then linking new knowledge assets to their existing stock
5
(Boisot, 1998). In this way they enhance their rent-generating potential. The details of how
this is done are given under the heading of ‘agent interaction’.
From the financial and experience funds, agents draw meetings and knowledge-investment
budgets. Money that is not spent in a given period gets put back into the relevant fund where
it accumulates. An age nt’s financial funds correspond to its tangible assets whereas its
experience funds correspond to its non-fungible intangible assets. Each fund, or both, can be
switched off with a toggle. The program can thus be made to behave in a modular fashion
with agents surviving either through trading and collaboration with other agents alone, or
through knowledge creation alone. An agent’s preference for using one type of fund or for
another – i.e., for trading in existing knowledge or for investing in knowledge creation - is set
at the beginning of the simulation for all agents.
1.2 Knowledge Assets:
In Sim-I-Space, knowledge assets are represented in network form. A knowledge network
consists of a collection of elements and of relations between elements. We sha ll refer to the
elements of the network as nodes and to the relations between elements as links. Nodes and
links can be combined with certain probabilities1 called linkage probabilities. A knowledge
asset, then, can either be a node, a link between two nodes, or a set of interlinked nodes that
can vary in size and complexity. Each node and each link varies in how far it has been
codified, made abstract, or has been diffused to other agents. Thus each node and link has a
unique location in the I-Space that determines its value to the agent and hence it’s rentsgenerating
potential.
|
|
