(...)
IMO you shouldn't be amazed at all, most of the retail bots are pure garbage. They are just a bunch of ifs and else statements that utilize the same ol' useless Indicators and shipped as AI. The issue is these "bots" are merely "tuned" to the "most profitable" settings they have found through brute forcing through historical data.
(....)
Maybe the one you are mentioning about garbage bots or normal ifs and else statements are those script kiddies bots that are available online.
You forgot that there are some entities or big traders that have bot some different purposes and those bots are not ordinary, some of them already have machine learning (AI) integrated which over time, collects data or learn from past price actions and more accurate somehow or more precise but I believe it's still not 100% accurate all the time.
Machine Learning is not magic,
The data you teach the machine will be the output of his learning:
good data in -> good data out
garbage in -> garbage out
there is no such thing as garbage in -> good data out
There are too many cases in the software industry where AI (prefer to call it neural network) is used as a marketing buzz-world just to show that someone uses the latest technology.
The real interesting topic here is the difference between the Baeysian method of learning and the classical scientific learning.