How Do You Govern Self-Educating AIs?

When an AI reaches a level of complexity in its data analysis that it is impossible to tell how it is making its decisions, how do you make rules about competition or discrimination?
https://www.nytimes.com/2019/01/20/technology/artificial-intelligence-policy-world.html ;

WhatsApp limits forwarding rules to combat fake news

Mind you, we have no data that it actually worked when deployed in India. I hope someone is tracking whether this has impact.
https://blog.whatsapp.com/10000647/More-changes-to-forwarding ;

Links for MN/Frontier Report

Tech Dirt article: https://www.techdirt.com/articles/20190108/08015841353/frontier-hammered-minnesota-ag-refusal-to-repair-broadband-network.shtml 

Actual MN Report: http://mn.gov/commerce-stat/pdfs/frontier-service-quality-report-final.pdf