I also thought of AI while reading this title. Many people assess that strong AI will not be possible until will figure out how the brain works. Yet it's very much possible that we create a machine that exhibits human-like behavior before we understand how it works.
Trial and error exists, it's called 'genetic programming'
The trial in genetic programming is too random, compared to what a human researcher would try.
Doesn't this tear down the entire premise of the article? "Just trial & error" dismisses all the thought that goes into designing experiments and analyzing the data.
Human trial and error is more like a hill-climbing algorithm or a localized graph search algorithm, things which computers can do quite well.
You'd be surprised at how directed forces applied to "random" values yields surprising solutions.
We only need a way to tell the computer "hey, this random result/behavior could be useful for this and/or that specific purpose!"...