Automated edits are discouraged in OSM, and if they are going to be done one has to send a notice to the appropriate mailing list and create the edits using additional username (so it will be easier to revert the changes).
I wouldn't say discouraged, just they have to be registered. See the list of all the imports, planned or finished:
http://wiki.openstreetmap.org/wiki/Import/Catalogue. In a lot of places pretty much all of the data came from automated imports. People checking the map is good, but manual data entry is a pain and there aren't many doing that.
That's why it's only possible to use whitelisted data sources: http://wiki.openstreetmap.org/wiki/Aerial_imagery . Given the scope of this task, "find a license-compatible imagery" is a non-issue.
run some machine learning algorithms to identify roads
If it was that easy we'd already be doing that. 30cm imagery is still very low resolution for tasks like spotting dirt tracks under foliage, plus even the current state of the art in image recognition gives far too many false positives and negatives to be really useful without having someone manually double checking everything.
The actual issue for tracing paths is that they're hard to distinguish, even from hi-res aerial imagery; a bit of ground knowledge of the area helps a lot, so that you don't map nonexistent paths which are actually fences or such.
Dunno about easiest, but the most effective is probably to just buy the detailed satellite/aerial imagery, (http://microsites.digitalglobe.com/30cm/) run some machine learning algorithms to identify roads, and upload that. https://github.com/trailbehind/DeepOSM https://developmentseed.org/blog/2017/01/30/machine-learning...