Look
learning is like tatonement
like groping by market agents in an exchange system
Out of equilibrium which is always
Rationality by agents is bounded by each agents
decision interval
choice of best model
And data base
Learning leads to the decision equivalent of perfect foresight
once you remove
the ramdom shock generator surrogate
for irreducible external Uncertainty
How any of this is pernicious in the science
Escapes me
The problem comes with policy
These models that depart from empirics
Are useless ...obviously
