Will Crichton

limitation of example-driven approaches to mixed-initiative co-creativity: you can't infer user intent purely from concrete examples. examples communicate both too little (missing general rules that guided example selection) & too much (extraneous features irrelevant to intent)

program synthesis is promising here bc if we can synthesize a bunch of simple candidate programs that match user-specified examples, we can then present these programs to the user directly ("did you mean X?"), let them pick the best program & refine it to clarify their intent

if you execute the interaction design correctly here, it starts to look a lot like explicit support for the process of intent formation! example-driven synthesis helps the user discover what concepts the system understands & how they can be combinedhttps://twitter.com/maxkreminski/status/1129122168871170048 …

nice, thanks for the ref! also planning to do a short paper-of-the-week thread on your Human-Centric Program Synthesis paper soon too, since that one anticipates a lot of the things i’m working on to some extent