You bring up great points and I didn’t mean to imply that prediction is the only way DL/ML is used. When I use the word ‘assumptions’ I am pointing to assumptions about what people need as a solution before you validate it (and build it).
When we have done prototyping we have tried out experiences that aided in assessment, bubbled up important information (say with notifications), solution recommendation, scheduling (type of optimization problem), prediction, etc.
These possible solutions we prototyped came from understanding the problems that people were trying to solve first. Then we looked at what would be an ideal solution without knowing how to implement it. Looking at how people interact in ways that are tedium is a great starting point. Rich picture is an exersize to map out these interactions that I think will be used more going forward.