The goal of developing "automated machine learning" (AutoML) is to enable AI systems to handle automation tasks without human guidance. This includes things like model selection, hyperparameter optimization, and neural architecture search.
AutoML aims to take the tedious, repetitive tasks involved in machine learning workflow and have the AI system figure those out on its own. This could greatly increase productivity of AI practitioners.
Current AutoML systems still operate within bounds and constraints set by human developers. They choose among options humans have designed, rather than designing truly novel architectures themselves.
While AutoML shows promise for increasing AI productivity, experts debate whether AI systems will ever reach full autonomy in designing and creating themselves.
The counterpoint is that human guidance and oversight will always be necessary to align systems to human values.
As promising capabilities like few-shot learning and self-supervised learning continue improving, we may see AI systems take larger roles in their own development cycles.
But full AI "self-programming" remains largely theoretical at this stage.