The value and dangers of remembrance in changing worlds: a model of
cognitive and operational memory of organizations
Giovanni Dosi1, Luigi Marengo2, Evita Paraskevopoulou3, and Marco Valente4
1 LEM, Scuola Superiore Sant’ Anna, Pisa and visiting professor, Friedrich-Schiller-Universität, Jena
g.dosi@sssup.it
2 LEM, Scuola Superiore Sant’ Anna, Pisa, l.marengo@sssup.it
3 Departamento de Economia de la Empresa, Universidad Carlos III de Madrid, eparaske@emp.uc3m.es
4 School of Economics, Universit`a de L’Aquila, marco.valente@univaq.it
1. Introduction
The notion of organizational memory stands for an elusive albeit crucial feature of the organizational
reproduction of knowledge as distinct from the memory of individuals, namely the ability of
organizations to elicit stored information from an organization’s history that can be retrieved to bear
on present decisions (Walsh and Ungson, 1991). The property of memory of being “organizational”
means that, first, it may well be distributed within the organization in ways such that no individual or
individual subunit embodies the full representation or the full behavioural repertoires contained in the
memory itself. Second, the organizational character of the memory also implies that it is resilient - to
degrees to be established, which we shall indeed discuss in this work - to the turnover of individuals
within the organization itself.
In many respects “memory”, is a crucial corollary of organizations being path-dependently
reproducing institutions – or whatever name is chosen for collective behavioural entities as distinct
from agency-theoretic constructs (no matter whether of the ‘incomplete contract’ kind). Organizations
“remember” because they entail explicit norms and, together, more tacit practices addressed to
collectively solve practical and cognitive problems, ranging from the production of a car, all the way
to e.g. the identification of a malaria-curing molecule. This is another way of saying that organizations
learn, store, elicit and modify over time routines and other “quasi genetic action patterns “ as Winter
(in Cohen et al., 1996) puts it. Together, organizational routines bear the memory (and possibly the
continuing threat) of conflict of power and income distribution which contributed to generate them and
continues (or not) to sustain them.
1
Granted all that, organizational memory concerns, first, the structure of beliefs, interpretative
frameworks, codes, cultures by which the organization interprets the state of the environment and its
own “internal states” (Levitt and March, 1988): in brief, call all this the cognitive memory of the
organization. Second, organizational memory includes routines - comprising standard operating
procedures, rules and other patterned actions- : call that operational memory of the organization.
Both, cognitive models and operational repertoires are the outcomes of learning processes and thus
evolve over time in response to experimentation and feedbacks from the environment. However, they
might often entail quite high degrees of inertia and path-dependent reproduction, as in fact, by its
nature, organizational memory reproduces over time what an organization has learned throughout its
history. In short, the two types of memory concern the organizational capabilities to “understand” the
characteristics of the environment, on one hand, and to coordinate particular sequences of actions
across different decision units and individuals, on the other. In turn, a major question we shall address
below concerns indeed the role of memory in changing environments. “Competence traps” clearly
belong to this domain of analysis.
Cognitive and operational memories entail an “if….then” structure. Signals from the environments, as
well as from other parts of the organization, elicit particular cognitive responses, conditional upon the
“collective mental models” that the organization holds, which are in turn conditional upon the
structure of its cognitive memory. Cognitive memory maps signals from an otherwise unknown world
into “cognitive states” (“…this year the conditions of the market are such that it will be profitable to
produce X…”). Conversely, the operational memory elicits operating routines in response to cognitive
states (“…this year we should produce X…”), internal states of the organisation (“…prepare the
machine M to start producing piece P…”) and also environmental feedbacks (“…after all X is not
selling too well…”).
A promising candidate to model both types of memory finds its roots into the formalism of classifier
systems (cf. Holland, 1975 and 1986). The model that we shall propose below finds its ascendancy
there, and in their application in Marengo (1992), albeit with significant modifications. On purpose in
the following we do not commit to any specific “if …then” structure, which in an organizational
setting implies also specific commitments to rules of selection and evolution (so, technically, we also
depart from the “bucket brigade” formalization of rule reinforcement).
2
Moreover, in the model we explore below, we allow the possibility of having on the “if” part both
“cognitive organizational states”
associated with explicit organizational rules but also more
idiosyncratic present or past “inner states” and cognitive frames. Moreover, we implicitly take on
board the coupled dynamics between e.g. standard organizing procedures (SOP) belonging to the
organizational domain of explicit rules –whose likely memory entails devices such as operational
manuals and straightforward commands –, on the one hand, and partly uncodified responses to
whatever “if” in the “then” behavioural response – ranging from tacit repertoires all the way to
conflictual responses or sheer mistakes.
On the ground of such formal apparatus we shall first analyze the structure of what is remembered
conditional on the characteristics of the environment. Second, we shall explore the conjecture –well
grounded in several empirical studies- that a memory structure well “fit” for a particular environment
may turn out to be pernicious under different technological or market conditions (cf. among others
Tripsas and Gavetti, 2000 and Bresnahan, Greenstein and Henderson, 2010). And conversely, we shall
try to identify the circumstances under which “learning” implies primarily “intentional unlearning”.
Third, we shall address the possibility of sort of “organizational cognitive dissonance” characterized
by the mismatching between the mental models of the organization –with the ensuing operational
strategies- and its operational repertoires (meaning also the possibility of being “successful” - against
whatever evaluation criterion –with a “wrong” model of the world, or conversely, failing despite
having the “right” one). In fact, in many respects, the proposed work is an exploration of the outcomes
of the mappings between cognitive and operational memory, on one hand, and patterns of
environmental change, on the other.
We shall proceed as follows. In Section 2 we shall attempt a broad even if necessarily concise
assessment of the state of the art of the incumbent knowledge concerning organizational memory.
First, straightforwardly we shall address the meaning of such notion, its structure and the process of its
storing and retrieving. Second, we discuss the inevitable path dependent and inertial nature of
organizational memory. Third, we shall survey the evidence on the role (and dangers) of
organizational memory especially in changing environments. In that, crucial issues regard, as
mentioned, (cognitive and operational) “traps” as well as the importance of organizational forgetting.
Fourth, we explore the (even more scattered) evidence on mismatches between “collective mental
models”, behavioural patterns and payoffs thereof; that is the evidence on organizational cognitive
dissonance.
3
Section 3 will present the structure of the model which, in a simulation environment addresses the
interpretative questions stemming from the foregoing pieces of evidence and explores the dynamics of
collective cognition , behaviour and ensuing (most often opaque) environmental payoff feedbacks.
Section 4 will discuss the major simulation results.
2. Organizational Memory: Characteristics, determinants and dynamics
2.1. Organizational Memory and Organizational Routines
The existence and importance of organizational memory is associated with the very ability of
organizations to interpret their environment, learn how to solve operational problems and, by doing
that, built constructs of knowledge that can be stored and reused. One side of the story is in a broad
sense cognitive.
The view of organizations as fragmented and multidimensional interpretation systems is grounded on
the importance of collective information processing mechanisms that yield shared understandings
(Daft and Weick, 1984), or “cognitive theories” (Argyris and Schon, 1978), of the environment in
which they operate , and assist organizations to bear uncertainty, and, as we shall see, environmental
and problem-solving complexity . If one subscribe to the notion that organizational learning is a
process of refinement of shared cognitive frames involving
action-outcome relationships (Duncan
and Weiss, 1979) and that this knowledge is retained –at least for some time- and can be recalled
upon, this is like saying that organizational learning is in fact the process of building an organizational
memory. This cognitive part of the memory is made of “mental artefacts” embodying shared beliefs,
interpretative frameworks, codes and cultures by which the organization interprets the state of the
environment and its own “internal states” (Levitt and March, 1988).
Together, there is an operational side to the organizational memory involving the coupling between
stimuli (events, signals -both external and internal ones-) with responses (actions), making up a set of
rules that remain available to guide the orientation of the organization and execute its operations. In
this domain the memory largely relates to the ensemble of organizational routines - patterned actions
that are employed as responses to environmental or internal stimuli, possibly filtered and elaborated
via the elements of cognitive memory (much more on routines in Nelson and Winter, 1982; Cohen et
al, 1996; Becker et al., 2005; Becker, 2005 and the literature reviewed here ). As Cohen and Bacdayan
(1994) put it, this procedural side is the “memory of how things are done”, bearing a close
4
resemblance with individual skills and habits, often with relatively automatic and inarticulated features
(p.554)
In fact, the characteristics and evolution of organizational memory mirrors the characteristics and
evolution of organizational routines. In the case of routines, the memory elicits a “relatively complex
pattern of behaviour triggered by a relatively small number of initiating signals or choices” (Winter,
1996). How small or big is the initiating set of signals in itself is an important interpretative question,
which has to do with the ways the organization categorizes environmental and intra-organizational
information. And likewise the behavioural patterns are likely to display different degrees of
conditionality upon particular sets of signals. So, at one extreme the action pattern might be totally
unconditional and “robust”: “perform the vector of actions X* irrespectively of the perceived state of
the world”. At the opposite extreme actions might be very contingent on the fine structure of their “if”
part.
As we shall explore below, it might well be that the coarsesness of the “ifs” and the “robustness” of
the “then” parts might well depend on the nature of the environment and its dynamics. A conjecture in
this respect is that the more complex and unpredictably changing is the environment, the less
contingent is the behaviour (Heiner, 1983; Dosi et al., 1999). After all, routines can be seen as an
uncertainty reducing device (Becker and Knudsen, 2005; Dosi and Egidi, 1991): robust and largely
uncontingent routines might be those memorised under highly complex and changing environments.
The importance of memory is not confined to its value as a stock of information referring to past
activities and decisions of organizations; most importantly, its significance resides in its knowledge
contribution for current problem solving and possibly to learning (Huber, 1991), to organizational
legitimacy and consolidation power (i.e. “this is how we have always been doing things”) that can be
used to overcome intra-organizational conflicts (Feldman and March, 1981) and influences the
establishment of future strategies and their implementing actions. The other side of the coin is the
ensuing danger of hindering organizational change. Indeed, depending on environmental dynamics,
organizational memory can become an asset for success or, on the contrary, an obstacle to change (see
Section 2.4)
With regards to its structure, organizational memory represents a stock of knowledge whose elements
are characterized by different degrees of latency and “tacitness” and are unevenly distributed manner
across the organization. As the elements of the memory are distinguishable –although often
5
interrelated- in terms of the problem they address (or addressed), the actors that embody them etc., it
follows that there exists a variety of storage locations within the organization. The organizational
location of different elements of the memory is determined by their content: a first distinction to make
is between “storage bins” that carry elements containing information about the stimulus of a decision the “if” part - and others that contain the elements with primarily the knowledge on the response to
that stimulus –the “then” part- (Walsh and Ungson, 1991)1. These distributed “knowledge reservoirs”
(McGrath and Argote, 2000) can nevertheless be retrieved and possibly recombined under the
“appropriate” environmental and internal states. The frequency of use (or length of inactivity), the
capacity of the storage location, the degree of interconnectedness and the cost of maintenance and
retrieval of the elements of memory are all factors that determine the characteristics and (re)use of the
constituents of organizational memory. As such memory structures tend to differ across organizations
and over time time.
The retrieval of multiple memory elements entails the necessity of mechanisms that bring together,
interpret and categorize environmental stimuli, supported by the inner social network of the
organization, its sequence of routines and the physical technologies used (Argote and Ingram, 2000).
Retrieval is collective in nature, which might also mean that it can be accompanied by tensions and
conflicts whose resolution relies upon “horizontal” coordination and hierarchal authority.
2.2 Path dependence
Inertia and path dependence are an almost inevitable corollary of the very existence of organizational
memory. The organization is able to recall specific cognitive frames and behavioural repertoires
precisely because they are stored and inertially reproduced (possibly with slight modifications) over
time. Organizations are inertial because their cognitive and behavioural performances are a far cry
from this response “plasticity” normally postulated by any view of organizations as “bundles of
optimal contracts” (Rumelt, 1995). Organizations path dependently carry with them their birthmarks
and what they have subsequently learned throughout their history. It is true that firms typically live in
selective environments which tend to “weed out” the most dysfunctional traits and behaviours.
However, typically their overall “fitness” (say, their revealed competitiveness) depends upon multiple
inter-related traits: in such cases, selection happens on a fitness landscape with multiple local maxima
that are determined by (random) initial conditions (Levinthal, 1997; Castaldi and Dosi, 2006). Indeed,
1
See Walsh and Ungson (1991) also for a detailed discussion of the different “storage bins” of organizational
memory.
6
organizations typically compete on such complex landscapes and interrelated technological and
behavioural traits are responsible for path dependent reproduction of organizational arrangements
(Marengo, 1996; Levinthal, 2000). The link between what firms do and the way they are selectively
rewarded in the market is utterly opaque for at least three reasons: (i) the complexity of the
environments where they operate; (ii) the already mentioned multiple “epistatic correlations 2” amongst
behavioural and technological traits; and (iii) significant lags between organizational actions and
performance-revealing feedbacks. In such circumstances, path dependence can also be fuelled by
behavioural/procedural and “cognitive” forms of inertia (Tripsas and Gavetti, 2000), as well, not too
paradoxically, by the reinforcement of traits, behaviours technologies which have been highly
successful in the past but might turn out to be detrimental under changing environmental conditions
(see also the next subsection): competence traps (Levitt and March, 1988) belong to this general
heading. Widespread path-dependent properties emerge also from the “organizational ecology”
literature investigating the long-lasting bearing of founding conditions of new enterprises, which
become imprinted in the organization and mold their subsequent development in terms of
organizational practices and broad strategic orientations (among several more see Carroll and Hannan,
2004).
In the model which follows we shall address below isomorphic issues by means of simulation
exercises addressing, among other issues, the relationships between the “depth” and inertiality of
memory and path-dependencies in organizational behaviours.
2.3 Organizational memory in changing environments
Organizational memory carries overtime what the organization has learned, directly through its past
experiences, vicariously by observation of the experiences of other entities, or has been so to speak
“brought in” by members of the organization – especially by the top management, executives and
technicians- with their strategic orientations, cognitive views, heuristics and know-how. A crucial
question regards the usefulness over time of the outcomes of such learning activities as carried by the
organizational memory. The question, in turn, boils down to both the characteristics of learning and
the depth of environmental changes the organization faces.
2
More on the application of this notion to the economic domain in Levinthal (1997) and Marengo and Dosi
(2005); see also below
7
To a considerable extent, we have already mentioned, organizations learn from their experience,
reproducing beliefs and actions that have been associated with good outcomes and avoiding actions
associated with bad ones. If the world makes simple sense and is stable, the repeating the “good”
routines is an effective organizational behaviour. However, the world is rarely simple enough to make
experience an infallible teacher (March, 1981; March and Olsen, 1976).The interpretation of
experience itself is ridden with cognitive frames and subject to cognitive biases, so that what is learned
depends as much upon history as on the frames applied to that history (Levitt and March, 1988;
Fishhoff, 1975; Pettigrew, 1985), while beliefs, stories and routines will be conserved notwithstanding
disconfirmation.
Moreover, quite independently from any possible cognitive bias, the environment may well change in
ways that decrease the “fitness” of cognitive and behavioural patterns which were well suited to the
“old” environment or even make them detrimental. This is indeed what competence traps are
essentially about (Levitt and March, 1988). Note that competence traps may refer primarily to the
cognitive domain or alternatively to the operational one. In the former case the “trap” concerns
primarily the reproduction beyond their times of usefulness of previously successful strategic
orientations and heuristics. Call them cognitive traps; relatedly, the escape is likely to involve
“strategic reorientation” possibly linked with the substitution of the top management. Conversely, the
“trap” might concern the “way of doing things” – that is the ensemble of routines and other recurrent
action patterns. In these circumstances the remedy is likely to involve also procedural and
organizational changes. In actual fact, the cognitive and operational lock-ins are likely to come often
together.
Bresnahan, Greenstein and Henderson (2011) present an excellent illustration of this point3 in two
cases of “Schumpeterian transitions” across different technological trajectories and of the vicissitudes
of the firms which were market leaders under the old ones - in their examples, IBM facing the
emergence of personal computers and Microsoft vis-à-vis the arrival of the browser-. Take the IBM
case. Strong technological capabilities match a commitment to incrementalism in product
architectures, cumulative learning, vertical integration, proprietary standards, coordinated strategic
governance and, on the market side, a reputation for post-sale service. This “IBM model”, Bresnahan
et al. (2011) insightfully show, is well aligned to market requirements under the mainframe/mini
computer trajectories, but becomes misaligned to the requirements of effective production and
3
Even if, admittedly, the authors are inclined to offer a somewhat different interpretation of the evidence in terms
of economies and diseconomies of scope in presence of jointly shared assets
8
marketing of personal computers. It is not that the “raw” capabilities are not there. They are. And in
fact IBM even proceeds to a rather successful exploration of a new combinatorics between elements of
technological capabilities, organizational set-ups and market orientation well suited to the personal
computer world. However, that very success accelerates the clash between the “PC organizational
model” and the incumbent “IBM (mainframe) model”. This latter wins and by doing that IBM
ultimately kills its PC line of business.
It is a story vividly illustrating the path-dependent reproduction of capabilities, shared strategic
models, specific organizational arrangements and the ensuing traps. To repeat, it is not that IBM
lacked any of the elements underlying successful “PC-fit” combinations. It is just that capabilities,
“visions” and organizational set-ups and their specific combinations are better described at least in the
short term as state variables rather than control variables, in Winter (1987) characterization. Of course,
also state variables can and are indeed influenced by purposeful discretionary strategies, that is, by the
explicit manipulation of control variables. However this takes time and is tainted by initial birthmarks
and subsequent historical paths the organization has taken with respect to both operational repertories
and higher level collective visions concerning the very identity of the organization.
In fact, technological and market discontinuities, –quite a few analyses suggest-, demand forgetting
and unlearning (Hedberg, 1981; Huber, 1991; Nystrom and Starbuck, 1984; Walsh, 1995; Klein, 1989)
involving also changes in the organizational structures and the purposeful erasing of at least parts of
the cognitive and procedural memory of the organization. A revealing example regards the
“unlearning” activities involved in the Merger and Acquisition processes. Kunisch, Wolf and Quodt
(2010) distinguish three domains of possible “misfit” between the two merging organizations - at the
level of artefacts, behaviours and corporate cultures. On the ground of a large database on M&A in
Germany, they find that cultural misfits are particularly conducive to a lower subsequent performance,
while - irrespectively of the sources of misfit – more unlearning is associated with easier absorption of
new knowledge and better post-merger performances.
Clearly, unlearning comes at the cost of the loss of a good deal of the experiential wisdom of the
organization itself (Gavetti and Levinthal, 2000): however, whether this is actually a cost, in terms of
organizational “fitness” is likely to depend upon the depth of the changes in the appropriate
technological capabilities and in the market environments. The general intuition stemming from the
empirical literature is in fact that the value, or the cost, of cognitive changes and procedural forgetting
is a function of the changes in the fitness landscape which the organization faces. Indeed, in the
9
following, we shall explore more formally this conjecture by explicitly modelling shocks on such
landscapes and studying the ensuing impacts upon organizational performances under different
degrees of cognitive and procedural inertia, or conversely “forgetfulness” of organizations.
A related but distinct question concerns the type of cognitive structures and routines characterizing
organizational memory, depending on different environmental features and dynamics. In particular ,
could it be that, as already mentioned, in highly turbulent , highly uncertain, environments it efficient
to develop “robust”, relatively invariant, routines which explicitly neglect pieces of available
information.
2.4 Organizational cognitive dissonance
The patterns of change of organizational memory are determined by changes in both its cognitive and
operational part and may or may not be parallel and synchronized. Changes in the cognitive and
procedural memory are driven by different processes, with cognitive search often broad and loosely
specified, while
more narrow (local) experiential research often guide the latter (Gavetti and
Levinthal, 2000).
In changing environments, the presence of different hierarchical forms and the possibility of
competence traps (cognitive or procedural or both) increase the risk of mismatching between the
mental models of the organization –with the ensuing strategic orientations - and its operational
repertoires. Such an organizational cognitive dissonance may translate into being “successful” against whatever evaluation criterion –with a “wrong” model of the world, or conversely, failing
despite having the “right” one.
Changes in the environment trigger processes of adjustment and adaptation of organizations and ignite
processes of interpretation of the new states of the environment. That very process tends to decouple
the “if..then” rules. In such circumstances organizational memory is activated in two ways. First, it is
called for in understanding the new state of the environment as the receivers of new signals will
commence their interpretations relying on existing cognitive structures. Second, new interpretations
are followed by a retrieval process of elements of procedural memory involving the evaluation and
revision of existing operational routines, deciding whether the existing operational repertoire is useful
under the new conditions or whether new rules need to be created. Re-coupling the cognitive and
10
operational elements of the memory under new environmental conditions is a task that might not be
symmetric either in terms of depth or speed: hence the mismatching between cognition and action.
Inertia and path dependence are ubiquitous attributes of both the interpretation process as well as the
evaluation heuristics addressing the degree of fitness of incumbent routines, – with evaluation “metaroutines” too being themselves part of an organization´s memory . It is indeed the case that inertial
evaluation heuristics can slow down organizational adaptation and change in so far as they are crucial
in altering or reinforcing the cognitive states of organizations (Garund and Rappa, 1994).
The existence of distinct bundles of evaluation heuristics concerning cognitive and procedural memory
can become the source of cognitive dissonance. In that, the distinction between automatic and
controlled retrieval of memory elements (Walsh and Ungston, 1991) – especially with regards to the
procedural part of the memory - assists in the explanation of failures to accord the cognitive state with
the operational routines of an organization. Automatic retrieval entails the danger of using existing
operational routines, when the “new reality” would actually call for the development of new ones.
Conversely, controlled retrieval (comforted by hierarchical power) may again induce mismatches
between cognition and action as, for instance, top layer management may impose the development of
new actions although the existing ones could well fit the new environmental conditions.
Section 3: A model of cognitive and operational memory of organizations
3.1 Scoping of existing model (Evita)
3.2 Formalizing the notion of organizational memory
•
The organizational problem is to develop a vector of interdependent actions in a
complex environment characterized by a (large) set of interdependent features
•
The (large) of environmental configurations can be partitioned in equivalence classes,
where each class requires a different action profile.
•
A payoff or fitness function which, for every environmental profile, gives the payoff of
every action profile
Three notions of complexity of the problem: Three notions of complexity of the problem
11
•
Categorizability: how large are these equivalence classes? The larger, the more
invariant the action. In some of the simulations only a few environmental features
(“core features”) influence the relative fitness of actions, all the others are irrelevant.
•
Neutrality: are such classes made of similar environmental profiles? If we modify one
bit of the environmental configuration, does the fittest action tend to be same or not?
•
Ruggedness: if we modify one bit of the environmental configuration, do the fitness
value of the action profiles tend to change smoothly or abruptly?
More formally
 Set of environmental features: E={e1, e2,…., en}, with ei={0,1}, thus 2n environmental
profiles
 An action profile is the choice of values for m interdependent actions: A={a1, a2,….,
am}, with ai={0,1}, thus 2m action profiles
 The fitness landscape: F: E  A  R attributes a real valued payoff to each of the 2n+m
environment-action states. The shape of this landscape is defined by the three
dimensions of complexity defined above
 Action is chosen by means of a system of condition-action rules that prescribe a
specific course of action when some environmental condition is met. Each rule takes
the form:
c1, c2,…., ck  a1, a2,…., am with ci={0,1,#} where ci sets a condition on the i-th
environmental feature, which is met if ci = ei or ci = #
 Action is chosen by means of a system of condition-action rules that prescribe a
specific course of action when some environmental condition is met. Each rule takes
the form:
 c1, c2,…., ck  a1, a2,…., am with ci={0,1,#} where ci sets a condition on the i-th
environmental feature, which is met if ci = ei or ci = #
12
Π
e1
a1
e2
a2
c1
e3
a3
e4
a4
Rules and Memory
e5
cs
a5
•
The number of rules an agent can store is the size of his memory
•
If a rule’s condition matches
the current environmental .profile, the rule is called active
.
•
Rules that remain inactive
for δ periods are discarded; δ. is a memory decay parameter
.
.
.
.
Learning
•
.
en
ck
am
Learning takes place through rule selection and rule modification
Rule selection
•
Only active rules can act. Among the active rules the one with highest fitness is chosen
Rule Modification
•
On the action part local search (one-bit mutations) is performed
•
On the condition part two algorithms for the generation of new rules:
specification: whenever a rule ci = # acts, it is compared with other rules which,
under the current environmental state, trigger a different action mapping into a higher
payoff.
generalization: if no rule is active, the one which better matches the current
environmental profile generates a new one with enough #’s to be active.
Note: at the outset an agent is endowed with a rule whose condition is made entirely of # and a
random action. Then rules are generated and modified with the above mentioned mechanisms
13
Section 4: Simulations and Results
4.1 Simple landscape and online learning
•
A first baseline bunch of simulations evaluate the learning properties of an agent in a
“simple” landscape with three core bits.
•
Our learning agent develops rules that correctly match environmental and action
profiles.
•
Initially we have an exploration phase in which a large number of new rules are
generated with very low degree of specificity.
•
At a later exploitation stage, actions become increasingly tailored to the correct
environmental conditions. This process is generated by the cumulation of evidence that
a given rule is systematically better when the set of core bits are in a given
configuration. Notice that average specificity does not reach its maximum because a
default hierarchy persists
Exploration in a simple environment
14
If we keep the landscape simple, but increase the number of core bits memory requirements
increase sharply.
Complexity of the environment and number of rules
15
4.2 Complex environment and online learning
A second baseline bunch of simulations evaluate the learning properties of an agent in a
“complex” landscape.
If the number of core bits is low (i.e. the landscape is locally rugged but with a lot of
neutrality), learning is much slower that in a simple landscape, gets stuck in many local
optima.
If we increase the number of core bits instead the learning processes settles into a much
lower number of more general rules (routines emerging)
4.3 Changing Environments
Environment
Slow Changing
Fast Changing
Organization Structure
 Low number of rules
 More specific rules
 Generation of many
general rules
(routines)
constrained by
memory
 Low number of rules
 Generation of many
both general and
specific rules
(sophisticated
routines)
 More general rules
 Highest fitness case
 Both types of
organizations tend to
specialize in the
portion of the
environment they
occupy
 No forgetting due to
lack of use of rules
Limited Memory
Unlimited Memory
16
“Punctuated Equilibria” with system-wide shocks: Rule specificity
The Marks of path-dependency: Even in unchanging environments, firm-specific cognitive
frames and action repertoires…
Persistent cognitive and operational diversities across firms
When (path-dependent) memory becomes an obstacle to adaptation:
Environmental “punctuation”…
17
“Punctuated Equilibria” with system-wide shocks: Rule age
“Punctuated Equilibria” with system-wide shocks: Relative Fitness
Average
Std. Dev
Limited Memory
0.984992
0.0166744
Unlimited Memory
0.988809
0.012073
Limited Memory Erased
0.990349
0.011099
Unlimited Memory Erased
0.992403
0.0100562
18
References:
Alonso, R., Dessein, W., and Matouschek, N., (2008), “When Does Coordination Require
Centralization?” American Economic Review, 98:1, 145–179
Aoki, M., (1986), “Horizontal vs. Vertical Information Structure of the Firm”, The American
Economic Review, 76: 5, 971-983
Aoki, M., (1990) Toward an Economic Model of the Japanese Firm, Journal of Economic Literature,
28: 1, 1-27
Aoki, M., (2001), Toward a Comparative Institutional Analysis, MIT Press, Cambridge, MA
Aoki, M., and Dosi, G., (2000), 'Corporate Organization, Finance and Innovation', in Finance and the
Enterprise, Vera Zamagni (ed.), Academic Press, 1992, 37-61,
reprinted in G. Dosi,
Innovation, organization and economic dynamics: selected essays, 2000, Edward Elgar
Argote, L. and Ingram, P. (2000), Knowledge Transfer: A Basis for Competitive Advantage in Firms,
Organizational Behavior and Human Decision Processes, Vol. 82:1, p.150-169
Argyris, C., and Schon D., (1978), Organizational Learning: A Theory of Action Perspective, Reading,
MA: Addison-Wesley Publishing Co.,
Becker, M. and Knudsen, T., (2005),” The role of routines in reducing uncertainty”, Journal of
Business Research, 58:6, pp.746-757
Becker, M., (2005), A Framework for applying organization routines in empirical research: Linking
antecedents, characteristics and performance outcomes of recurrent interaction patterns,
Industrial and Corporate Change,14:5 , pp. 817-846
Becker, M., Lazaric, N., Nelson, R. R. and Winter, S. G., (2005), “Applying organizational routines in
understanding organizational change”, Industrial and Corporate Change, 14:5, pp. 775-791.
Bresnahan, T., S. Greenstein and R. Henderson, (2011), “Schumpeterian competition and
diseconomies of scope: illustrations from the histories of Microsoft and IBM”, NBER Working
Paper
Carroll, G.R. and M.T Hannan (2004), The demography of corporations and industries, Princeton,
Princeton University Press
Castaldi, C., and Dosi, D., (2006), “The grip of history and the scope for novelty:some results and
open questions on path dependency in economic processes”, in A.Wimmer and R. Koessler
(eds), Understanding Change Models, Methodologies and Metaphors, London , Palgrave
Cohen, M. and Bacdayan, P., (1994), Organizational Routines Are Stored As Procedural Memory:
Evidence from a Laboratory Study, Organization Science, 5: 4, pp. 554-568
19
Cohen, M., Burkhart, R., Dosi, G., Egidi, M., Marengo, L., Warglien, M., and Winter, S., (1996),
Routines and Other Recurring Action Patterns of Organizations: Contemporary Research
Issues, Industrial and Corporate Change, 5:3, pp. 653-698
Cyert, R.M., and March, J., (1963), A behavioural theory of the firm, Blackwell: Oxford.
Daft, R., and Weick K.E., (1984), "Toward a model of organizations as interpretation systems,"
Academy of Management Review, 9, 284-295
Dosi, G. and Egidi, M., (1991) “Substantive and procedural uncertainty: An exploration of economic
behaviours in changing environments”, Journal of Evolutionary Economics, 1, pp 145-168
Dosi G, Marengo, L., Bassanini, A., and Valente M., (1999), “Norms as emergent properties of
adaptive learning: the case of economic routines”, Journal of Evolutionary Economics, 9, pp 526
Duncan, R., Weiss, A. (1979) “Organizational learning: implications for organizational design”, in
Research in Organizational Behavior, ed. B. M. Staw, 1:75-123. Greenwich, CT: JAI Press
Feldman, M., and March, J., (1981), Information in Organizations as Signal and Symbol,
Administrative Science Quarterly, 26:2, 171-186
Fischhoff, B., (1975), “Hindsight or foresight: The effect of outcome knowledge on judgment under
uncertainty”, Journal of Experimental Psychology, 1:288-99
Garund, R., and Rappa, M., (1994), “A Socio-Cognitive Model of Technology Evolution: The Case of
Cochlear Implants”, Organization Science, 5: 3, 344-362
Gavetti, G. and Levinthal, D., (2000), “Looking Forward and Looking Backward: Cognitive and
Experimental Search”, Administrative Science Quarterly, 45, 113–137.
Hedberg, B., (1981) “How organizations learn and unlearn?” in P. C. Nystrom & W. H. Starbuck
(Eds.), Handbook of organizational design (pp. 8-27). London: Oxford University Press.
Heiner R (1983), “On the origins of predictable behaviour”, American Economic Review, 73, 560-595
Heiner, R. (1988), “Imperfect decisions and routinized production: implications for evolutionary
modelling and inertial technical change”, in Technical Change and Economic Theory (ed. G.
Dosi et al.), Pinter, London.
Holland J. H., (1986), “Escaping brittleness: The possibilities of general purpose learning algorithms
applied to parallel rule based systems”, in R.S. Michalski, J.G. Carbonell and T.M. Mitchell
(eds.), Machine Learning II, Los Altos, CA, Morgan Kaufmann.
Holland, J.H, (1975), Adaptation in Natural and Artificial Systems, Ann Arbor, University of
Michigan Press
Huber, G., (1991), Organizational Learning: The Contributing Processes and the Literatures,
Organization Science, 2:1, 88-115
20
Klein, J. I. (1989), "Parenthetic Learning in Organizations: Toward the Unlearning of the Unlearning
Model," Journal of Management Studies, 26, 291-308.
Kogut, B., and Zander, U, (1996), What Firms Do? Coordination, Identity, and Learning, Organization
Science, 7:5, pp. 502-518
Kunisch S., C.Wolf and J. Quodt (2010), “When forgetting is the key: the value of unlearning
activities during post-acquisition integration” , Performance , 3 , 5-12, Ernst & Young
Levinthal, D. (2000), Organizational Capabilities in Complex Worlds” in The Nature and Dynamics of
Organizational Capabilities, edited by G. Dosi, R. Nelson, and S. Winter. Oxford University
Press
Levinthal, D., (1997), “Adaptation on Rugged Landscapes”, Management Science, 43, 934-950
Levinthal, D., and March, J., (1981), “A model of adaptive organizational search”, in J. March (ed),
Decisions and Organizations, pp.187-218, Basil Blackwell: Oxford
Levitt, B., and J.G. March, (1988), “Organizational Learning”, Annual Review of Sociology, Vol. 14
pg. 319-340
March J., and Olsen, J.P., (1976), Ambiguity and Choice in Organizations, (2nd Ed.) Bergen:
Universitetsforlaget
March, J., (1981), “Footnotes to Organizational Change”, Administrative Science Quarterly, 26: 4,
563-577
March, J., and Simon, H., (1958), Organizations, Blackwell: Oxford
Marengo, L. (1996), “Structure, Competence and Learning in an Adaptive Model of the Firm”, in G.
Dosi and F. Malerba (eds.), Organization and Strategy in the Evolution of the Enterprise.
London: MacMillan
Marengo, L. and G. Dosi (2005), “Division of Labor, Organizational Coordination and Market
Mechanisms in Collective Problem-solving", Journal of Economic Behavior and Organization,
58, 303-326
Marengo, L., (1992), “Coordination and organizational learning in the firm”, Journal of Evolutionary
Economics, Vol. 2, pp. 213-226
McGrath, J. E, &, Argote, L. (2000), Group processes in organizational contexts, in M. A. Hogg & R.
S. Tindale, (Eds.), Blackwell handbook of social psychology: Group processes (Vol. 3),
Oxford, UK, Blackwell
Nelson R., and Winter, S., (1982), An evolutionary theory of economic change, Belknap Press/Harvard
University Press: Cambridge MA
Nystrom, P.C. and Starbuck, W.H. (1984), “To Avoid Organizational Crises, Unlearn”, Organizational
dynamics, Vol. 12, pp. 53-65
21
Pentlant, B., and Rueter, H., (1994), “Organizational routines as grammars of action”, Administrative
Science Quarterly, 39, pp.484-510
Pettigrew, A. M, (1985), The Awakening Giant: Continuity and Change in Imperial Chemical
Industries, Oxford: Blackwell
Rantakari, H., (2008), “Governing Adaptation”, Review of Economic Studies, 75, 1257–1285
Rumelt, R., (1995), Inertia and Transformation, in Montgomery (1995) Montgomery, C.A. (ed.),
(1995), Resource-based and Evolutionary Theories of the Firm, Boston, Kluwer
Simon, H. A (1962), “The Architecture of Complexity”, Proceedings of the American Philosophical
Society, 106, 467-482.
Simon, H. A. (1981), The Science of Artificial, 2nd Ed.. Cambridge, MA: MIT Press
Tripsas, M., and G. Gavetti, (2000), “Capabilities, cognition, and inertia: Evidence from digital
imaging”, Strategic Management Journal, Vol. 1: 10/11, pp. 1147-1161
Tushman, M. L., Romanelli, E., (1985), “Organizational evolution: a metamorphosis model of
convergence and reorientation”, in Research in Organizational Behavior, ed. L. L. Cummings,
B. M. Staw, 7:171-222. Greenwich, CT: JAI Press
Tushman, M.L., and Rosenkopf, L., (1996), “Executive Succession, Strategic Reorientation and
Performance Growth: A Longitudinal Study in the U.S. Cement Industry”, Management
Science, 42: 7, 939-953
Walsh, J.P. and G.R. Ungson, (1991), “Organizational Memory”, The Academy of Management
Review, Vol. 16: 1, pp. 57-91
Walsh, J.P., (1995), “Managerial and Organizational Cognition: Notes from a Trip Down Memory
Lane”, Organization Science, 6:3, 280-321
Winter, S. (1987), “Knowledge and competence as strategic assets”, in D. Teece (Ed.), The
competitive challenge: 159-184. Cambridge, MA: Ballinger.
Winter, S., (1986), “The research program of the behavioural theory of the firm: orthodox critique and
evolutionary perspective”, in B. Gilad and S., Kaish (eds), Handbook of Behavioural
Economics, vol.A, pp. 151-188, JAI Press: Greenwich, CT
Winter, S., (1996) in Cohen, M. D., Burkhart, R., Dosi, G., Egidi, M., Marengo, L., Lassimo, W., &
Winter, S (1996), “Routines and other recurring action patterns of organizations:
Contemporary research issues”, Industrial and Corporate Change, 5:3, 653-698
22