A conversational framework for machine learning

This position paper describes a framework for machine learning (ML) based on conversation, which positions human and machine participants in a learning system with feedback mechanisms separated into a level of actions and a level of descriptions. This social model is based on theories of human learning and systems theory, and is intended for programmers, domain experts, users and non-experts to make sense of human-machine learning systems. It enables
journalistic questions to be raised about the descriptions, actions and feedback in a ML system, and furthermore addresses broader ethical dimensions of such a system such as trust and control, as ML systems become rapidly distributed and networked globally. Situated in art and design, I explore critical and creative uses of ML as an alternative to commercial ones, with the conversational framework outlined here intended for analysis and design of all these, across a broad range of applications.