... And this is what worries me about much flat design: With one very good designer it is probalby all fine. But if many people design slightly incoherently with it and/or make minor mistakes it gets harder to use.... So it is not that flat design is wrong, it is very brittle.
Here's a fresh example of flat design gone awry due to losing too many cues. I am depositing a check via banking app. The first screenshot is at the beginning. The second is after I've photographed the check front and back and manually typed in the value. Can you see what to do next?
The button changes color when active, but it changes to look exactly like the title, which is not clickable. And it's placed above the other controls, rather than below them. So it's missing some key affordance cues (look clickable; be in the expected place; be in sequence) and the one it does have (change affordance from "visible but visibly not available" to "visibly available") is botched. Some of that is due to flat design reducing the margin of error.
I can certainly see how this makes a UI less easily learnable.
I strongly think that most, if not all applications should be easily learnable. Partly because everyone begins at some point and because needing to rely on learned things still will have some background cognitive load and proneness to errors.
I think there may be such a thing as permanently unlearnable. Some users can master certain apparently simple functions but not others that seem to be as simple. The threshold of difficulty may vary by user. There are functions and household controls and keyboard combinations that I never master at the same level of muscle memory as others. Another example where flat design contributes to a permanently unlearnable or less learnable function:
Try to find the control to change from random play to linear play. One easier and one harder:
Jonathan Morgan:
For better or worse, most of us are pretty use to weak/absent signifiers in a mobile context by now--think about all the functionality on your phone that is only accessible through multitouch gestures, which usually aren't called out in the UI at all.
While I haven't presented any quantitative data above, I suspect that "use[d] to weak/absent signifiers" is not the same as not impacted by weak/absent signifiers.