Continuing the motivating remarks for a prediction network, we invite you to think about algorithms as managers ... of other algorithms. Traditionally humans play this role, and data scientists may be seen as performing supervision, selection and reward of models and data, as well as "helping" the data and algorithms in various ways.
However the vast majority of businesses and organizations cannot afford bespoke data science if it comes with the fixed high cost associated with humans. We must construct an alternative.
Automated assessment of statistical modeling has already proven to be a critical component of the Machine Learning revolution - indeed it may be seen as delineating it from inferential statistics. But if there is sufficient data for data hungry methods to work, there is sufficient data to assess them as well, algorithmically. There is no need to introduce expensive humans as overpaid judges.
Whether it be scoring rules, Nash equilibria or Errors-in-Variables literature, abundant theory exists to guide algorithms as they communicate with other algorithms, supervise them, assess them and assist them in various ways.
Algorithms can take constituent ingredients (micropredictions of snow cover and traffic) and produce a product "upstream" (ice cream sales predictions). When all economic decisions are automated, economics friction is close to zero and the consumer of microprediction is the benefactor.