A learning curve measures how much labor productivity (or output per labor
input) increases with cumulative experience. Thus, a Swedish ironworks increased
its output per worker-hour 2 percent per year despite no new investment and no
new production methods for 15 years. Likewise, U.S. Air Force engineers assume
a constant relative decline in labor required for an airplane body as the number of
airframes previously produced increases. A constant relative decline in labor requirements
as output expands means labor costs approach zero as cumulative production
tends to infinity – a nonsensical idea if output runs were long, but because in practice
they tend to be less than 20 years, economists can safely use this form of the learning
curve (Arrow 1962:154–194). Furthermore, the British scholars of technical progress
Charles Kennedy and A. P. Thirlwall (1972:38–39) argue that, even where learning
by doing from a good ends, where product types are constantly changing, we can
assume there is no aggregate limit to the learning process.