Source
March (1991), LIT-729, read in NOTE-581.
What was actually shown
March (LIT-729) argues the first half and simulates the second.
- The trade-off, argued. Compared with exploitation, exploration’s returns are “systematically less certain, more remote in time, and organizationally more distant from the locus of action and adaption” (p. 73). Learning from feedback therefore favours exploitation, the advantages cumulate, and competence in an inferior activity can exclude a superior one. Balance is “a primary factor in system survival and prosperity” (p. 71); March gives no rule for it.
- Mutual learning, simulated. Members adopt the code’s beliefs; the code adopts the majority view of members who know more than it does. Without outside input both converge on a shared, stable, not necessarily accurate belief. Equilibrium knowledge is highest when the code learns fast and members are socialized slowly; a mix of fast and slow learners beats a uniform population at the same average; moderate turnover improves the code because recruits, though less knowledgeable, are diverse.
- Under environmental change. Without turnover, the converged beliefs stay put while reality moves, and knowledge falls to chance. Moderate turnover sustains it, provided recruits are not chosen to fit the code.
- Who pays. Gains from slow learners accrue to fast ones, so no member has an incentive to learn slowly; under turbulence tenure is valuable to individuals in an organization with turnover.
What could have come out otherwise: the simulations. Slow socialization might have lowered equilibrium knowledge, and turnover might only have diluted it. Both results depend on the code learning only from members who outperform it.
What this does not say
- Not that learning is bad. March says the models do not “overturn the renaissance”; knowledge remains important.
- Not an optimal balance. He says determining one is “a nontrivial exercise”.
- Not about members choosing to leave. Turnover is an exogenous rate, and recruits are random with respect to the code.
- Not shown in real organizations. The dynamics are properties of one stylized model in which nobody observes reality and members learn only through the code.
Connections
- Transient diversity (THEORY-148) and the Zollman effect (THEORY-143). Those accounts find, in models of scientific communities, that diversity of pursuit must last long enough to survive misleading early evidence, and that dense communication can end it too soon. March’s model reaches a parallel result for an organization: fast socialization ends diversity too soon for the code to learn from it. The mechanisms differ, a central code here and peer communication there, so no relation is declared.
- Drift (THEORY-144). Rasmussen’s drift moves practice toward a safety boundary under cost pressure. March’s lock-in keeps belief fixed while the boundary moves. Both make an organization worse while its local adjustments seem sound; the routes differ.
- The owner’s essay. For §5.4 it supplies lock-in without a decision to lock in; for §8.2, an organizational case of a community’s knowledge depending on sustained dissent.