The purpose of this paper is to demonstrate that cybernetic theory explains learning curves and sets the curves as legitimate and efficient tools for a pro-active energy technology policy.

The learning system is a non-trivial machine that is kept in non-equilibrium steady state at minimum entropy production by competitive, equilibrium markets. The system has operational closure and the learning curve expresses its eigenbehaviour. This eigenbehaviour is analysed not in calendar time but in the characteristic time of the system, i.e., its eigentime. Measured in eigentime, the minimum entropy production in the steady-state learning system is constant. The double closure mechanism described by Heinz von Förster makes it possible for the learning system to change (adapt) its eigenbehaviour without compromising its operational closure.

By obeying basic laws of second order cybernetics and of non-equilibrium thermodynamics the learning system self-organises its learning to follow an optimal path described by the learning curve. The learning rates are obtained through an operator formalism and the results explain observed distributions. Application to solar cell (photo-voltaic) modules indicates that the silicon scarcity bubble 2005-2008 produced excess entropy corresponding to costs of the order of 100 billion US dollars.

Grounding technology learning and learning curves in cybernetics and non-equilibrium thermodynamics open up new possibilities to understand technology shifts through radical innovations or paradigm changes.

Learning curves are legitimate and efficient tools for energy policy and industrial strategy.

Grounding of technology learning and learning curves in cybernetic and thermodynamic theory provides a stable theoretical basis for applications in industry and policy.