Cover Image


View/Hide Left Panel
FIGURE 3.1 Price of passenger transportation in cost per passenger kilometer (km)-hour.

FIGURE 3.1 Price of passenger transportation in cost per passenger kilometer (km)-hour.

issues embodied in learning curves include understanding the specific processes that lurk behind the black box of technological improvement over time and, more precisely, the question of “who learns what?”

At the most general level, technological progress results from cumulative experience, but the magnitude of this progress for an individual technology or service is hugely uncertain, and there is almost nothing deterministic about the learning phenomenon. A wide range of examples shows a fairly consistent set of results indicating that cost reductions of 10 to 30 percent for a technology might be expected from a doubling of cumulative production. However, Nakicenovic reminded the workshop audience that the deterministic appearance of many of the learning curves is deceptive and that we are essentially dealing with a probabilistic phenomenon. One can find many examples of negative learning and cost escalations, including the case of the Lockheed Tristar aircraft, as well as U.S. and French nuclear reactors. In exploring learning for specific technologies, he noted that for solar photovoltaics in Japan, cost reductions were very marginal during the basic research and development phase, and costs declined rapidly only when significant funding went into applied research. Analysis of other renewables technologies shows that increasing the scale of production, the size of the manufacturing facilities, the size of devices, and the size of installations contributes to cost reductions.

In his talk William Nordhaus of Yale University focused on the perils of the learning model for representing endogenous technological change in energy-economic models. He discussed the question of the mechanisms of learning, who learns, and how learning is transmitted from one generation to the next. He stated a belief that learning is driven by cumulative production, and noted the inherent difficulties in disentangling the effects of learning from other sources of productivity growth such as research and development; economies of scale; and technologies that are imported from outside the boundaries of the firm, the industry, or even the country. Nordhaus also discussed a study of the semiconductor industry by Irwin and Klenow (1994) that showed learning was three times more powerful within firms than across firms and that also found insignificant learning effects from one generation of a technology to the next; if a technology grew rapidly in one generation or slowly in one generation, the effect on the next generation of the product was insignificant.

Nordhaus expressed his concern about using learning in models. He noted that learning has become a favorite tool for representing technological change in many models of the energy sector and global warming. He attributes this to its being one of the few “theories” of technological change that can be included easily in models because of its simple specification. Nordhaus concluded that the modeling of learning is a dangerous technique, however,

The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement