Ecologies of Learning
First Part
Topic One: Learning Substitution
What is the basic idea of "learning substitution"?
A learning system can adapt using several mechanisms at different points and get the same effect. Thus one learning mechanism can substitute for another. But often using one mechanism inhibits the use of another mechanism. Adaptation in one part of the organization either reduces pressure for other parts to adapt or inhibits their adaptation. These learning substitutions may have similar short-term effects but different long-term effects (or have effects in space and structure).
How is it manifested in such standard organizational phenomena as specialization and division of labor?
Division of labor is a common approach to address complex problems. Separate departments can buffer and segment experience, making it easier to learn in one part by inhibiting similar learning in another part. The amount of "loose coupling" can improve experience interpretation, but "tight coupling" can improve detection of errors.
The organization can substitute learning in one specialized area for learning in another area. However, learning that leads to a comparitive advantage in one domain may decay adaptive capabilities in another domain. Specialization can improve performance but eventually can limit that performance.
Under what circumstances can power be substituted for learning?
Power allows an organization to change their environments rather than adapt to them, which causes them to lose their adapative capability. If they lose their control someday, they will become severely disadvantaged.
What else can be substituted for learning?
Adaptation by a fast learner in an organization can reduce the pressure for a slow learner to do the same, causing eventual differences in adaptive capabilities.
Changes in search, slack, and targets are also learning substitutes. Since we develop each simultaneously, we can adjust either one to reach an aspiration/performance equilibrium.
Dissatisfied people can also substitute leaving with speaking up, which robs the organization of a chance to learn from their dissatisfactions. Often organizations instil loyalty in their customers and employees to allow this learning to take place before particiapants exit the organization.
Learning at lower levels can also be substitued for learning at higher levels. If customers adapt to inefficiencies, the manufacturers don't have to.
What are the consequences?
Learning substitutions often have unwanted effects in time, space, and level of failure.
1. Temporal Myopia
Learning tends to sacrifice the long-run for the short-run. Adaptation or fast learning in one subunit may inhibit learning in another unit, which may have unwanted effects once the environment changes. Specialization in a niche area will cause problems once the niche becomes irrelevant.
2. Spatial Myopia
Learning favors things close to the oranization. Positive incentives for an action may cause the organization to overlook secondary effects. Downsizing can help the company but hurt the employees. Adaptation in one area may be also be disadvantageous to another area.
3. Failure Myopia
Oversampling of successul people may cause too much risky behavior (as will oversampling of unsuccessful people). One can experience either competency or failure traps.
Also, if your customers adapt to your inefficiencies, you won't learn to be responsive to them. If another competitor learns faster, you may lose your customers.
Topic 2: Network Externalities
What is the basic idea of "network externalities"?
Learning does not take place in a vacuum. It occurs not only as a result of one's own experiences, preferences, and aspirations but the experiences, preference, performance, and aspirations of others. Those other individuals, subunits, or organizations in one's network has a strong influence on learning.
Decisions by one person have to take into account the preferences, identities, and likely actions of others. Interactions between organizations can strongly affect learning.
How does it affect organizational learning with respect to technologies, strategies, organizational forms?
From institutional theory we know that issues of external legitimacy can help build mimetic isomorphism among organizations. They learn from the experience of others, and imitate the technologies, strategies, and forms of other organizations.
Network externalities like professional groups and governments can also impose rules and legitimating procedures that affect organizational learning.
Knowledge also diffuses across organizations, especially innovative technoligies.
What is the learning explanation for the greater number of lawyers in US business organizations than in Japanese business organizations?
The US has a stronger identity with individual rights (vs community rights), and has a longer history of resolving differences in the court system than those in Japan. Since work is embedded in a network of social relations much more strongly in Japan, resorting to legal means to resolve differences strains this social network and thus is less beneficial. Lawsuits are still seen as an "ungentlemanly" way to resolve differences in Japan. They are also less utilized in contracts between organizations as well-- instead long-term social relationships and reciprocal agreements are more common.
In the US a legal department has also become not only legitimate but essential. Adding lawyers to your staff not only helps you defend against lawsuits but improves your capability to use the court sytem to strategic advantage. Since other firms have legal departments, your organization should have them too.
How do network externalities affect the learning of preferences or identities in organizations?
Network externalities like professional groups can affect learning preferences or identities when these professionals are hired into the organization. The hiring of nuclear professionals into LILCO undoubtedly shaped the head-strong commitment to the Shoreham nuclear power plant, which cost $5 billion and was never operated.
What difference does it make to organizational stability and change?
In general, network externalities make the concept of autonomous learning somewhat misleading.
Also, basing aspirations on those of other organizations will increase the chance that one will be consistenly above or below the mean aspiration level. This will cause the bifurcation of the population of organizations into largely successful firms and largely unsuccessful firms, and learning may then become more superstitious (nothing you do seems to change one's level of performance).
Topic 3: Nested Learning
What does it mean to say that a hierarchical organization learns at the same time as its subunits learn and at the same time as individuals within the organization learn?
Organizations simultaneously learn which strategies to follow at the same time they learn how to follow them. This occurs at individual, sub-unit, and organizational leve. The collective learning of the individuals is combined via hierarchical structures into the collective action and learning of the organization.
What difference does it make?
How does the principle of learning substitution apply?
As a result, learning at one level in the organization can substitute for another level. Fast adaptation at one level leads to slow adaptation at another level. A company can find new markets for their products rather than expend R&D resources to develop better ones for the existing markets.
What are the implications for understanding the mutual learning of decision makers and decision implementers?
Decision makers focus on deciding which strategies to pursue, while implementers focus on how to pursue them. If these groups learn at different rates, desire may eventually exceed capability or vice versa.
Why might there be different delays in feedback for different levels of an orgnization?
Different levels of the organization may have different environmental feedback rates, which can cause either learning rate differences between organizational levels or interpretive differences.
For example, management may buffer R&D from the environment, which may improve exploration but reduce R&D's ability to adapt to sudden environmental changes.
What complications would differences in feedback delays pose for organizational learning?
Feedback delays are the primary cause of Platt's "social traps" and effects like "tradegy of the commons". Positive short term feedback will reinforce organizational behavior to act in ways that have long-term negative consequences.
Reducing R&D spending may quickly improve the bottom line for this year, but may ultimately sacrifice the viability of the firm. Since the negative feedback has a longer delay, it is less salient and thus doesn't inhibit short-term behavior.
Part 2:
What is a competency trap?
The more a particular behavior is used, the more it will be reinforced by experience and thus used more frequently. This positive feedback loop reduces the possibility of substituting another, potentially more suitable alternative.
Competency traps mean people who are successful with one technology may continue to exploit the technology and reinforce their focus and reliance upon it to the exclusion of other potential ideas. Competency can increase exploitation and reduce exploration, which may cause future difficulties if the environment changes.
Why might it be called a "lock-in" or a case of "increasing returns"?
When continued improvements to an existing technology bring increasing returns, there is little incentive to devote research funds to exploration activities that have only a small chance at major success. Organizations then get locked-in" to a specific technology, and lose adaptive capabilities in their pursuit for maximum returns.
Harriot, Levinthal, & March Model
This model explores how individuals or organizations change their budget allocations to various alternatives based on experience with these alternatives. Total performance is the sum of potential * allocation * competence. Comptencies and goals adapt at the same time as allocations. Other learners can also have an effect on either outcomes, aspirations, or competencies.
Key model assumptions:
1. Competence decreases with time and increases with allocation
2. Propensity to increase or decrease allocations depends on the success or failure with a decrease/increase strategy in the last period. Learners will continue what is successful.
3. Goals adapt to past performance
4. People will incorporate the allocation behavior of others into their experience (this model uses a mean allocation effect).
5. People will adapt their competencies to the highest level of competence within the population of actors (diffusion of knowledge)
6. Goals will also adapt based on the mean performance of others plus one's own.
7. In competitive situations, there may be less rewards if everyone allocates alot to the same alternative.
In the Herriott, Levinthal, & March model, what affects the likelihood of a trap on an inferior alternative in the case of an isolated learner? Why?
If the learning parameter b1 is fixed, the learner will specialize in one of the two alternatives, but not always to the best one. It largely depends on the initial conditions -- if initial conditions favor the inferior alternative, they are more likely to go that way.
This occurs because in general the performance in each period usually exceeds the goals, which increases allocation to the alternative. Fast learners may quickly devote all their resources to one alternative based on "increasing returns" without noticing it is the inferior alternative.
It can be avoided by either:
* slow adjustment of allocations and rapid adjustment of goals (which will produce more frequent failures)
* slow learning and fast forgetting
What are the effects of the diffusion of behavior and goals in a population of independent learners? Why?
The diffusion of allocation behavior tends to decrease the mean and variance of allocations, driving all actors toward the same set of allocations. People tend to imitate each other until homogeniety is complete. This favors fast learners who change allocations quickly and goals slowly (so they don't experience failures).
Goal diffusion leads to a division of actors into a "largely successful" group and a "largely unsuccessful" group. Those below the mean performance will consistently underperform and continue to engage in "failing" behavior. Those above the mean will consistently overperform and engage in "success" behavior. Some actors also oscillate between the two.
What general lessons can be learned?
If one also considers competition with diffusion, one finds that competitors who are successful tend to specialize. Fast learners specialize quickly but often on the inferior alternative.
The variation of performance decreases with allocation diffusion but increases via competition and goal diffusion.
All of this suggests that interdependent learning makes rational behavior much more unreliable and confusing. This often creates trial-and-error learning and random performances.
The paper's final conclusions were:
1. Learning leads reliably to optimal choices in some situations but not others.
2. The final equilibrium is often determined by adaptation rates and initial conditions.
3. The best performers can be fast or slow learners (depending on the situation)
4. Fine-tuning learning can help find an equilibrium quicker but doesn't necessarily mean improved performance
To what extent do they depend on relatively minor asumptions and to what extent might they extend to a larger population of learning situations?
Allocation of resources to different alternatives is certainly an appropriate assumption for organizations -- this is similar to priority setting based on experience or typical budgeting rules (the successful organizations in the past get more resources).
Firms probably engage in more exploratory behavior than the model tends to allow. It is rare that a firm will completely abandon other alternatives in their pursuit for success, especially if they see other competitors pursuing them.
Also, firms may adjust their goals to something other than the mean performance,
especially if they are far above or far below the mean.