Economics, privacy, and natural monopolies

Classical liberalism is well-supported by a first class in economics: when economics assumes competitive markets, equilibrium outcomes have a number of desirable characteristics.   Classically, this was a reasonable expectation:   where patents might have offered a monopoly, it was temporary and the market was likely to eventually compete away the innovator's profits.

To the extent to which ideas may be patented and then their innovations exploited, this  premise was successful.  Modern machine learning, however, makes use of extremely large data-sets and has outputs which cannot be easily reverse-engineered.  It may well be that the reason that the methods are so widely published is that  researchers in this space realize that the data itself will be the source of future monopoly.

A natural monopoly occurs when successive units of a good are increasingly cheap.  In data-driven goods, this is super-charged: additional goods may actually reduce the cost of all other goods, as new data improves the algorithm across the system.  

Data-based monopolies are likely to be extremely durable.  Where safety is a concern, public policy is likely to guarantee their persistence.    Any upstart attempting to create a similar function will not have the data-set -and the underlying costs mean that any data owner may respond with lower prices, higher availability, or some other add-on to quash the competition.