Switching to modal screen-based interfaces happens when there isn't enough room on the panel in question to supply a separate physical control per thing that needs controlling.
Once you've had to do that, though, every remaining knob is a cost that could be removed by adding a few lines of code to the interface software.
A single touchscreen for everything is the natural endpoint of this process.
Because 'ooh, shiny!' That's my best guess, anyway.
Although there is a benefit to being able to deploy a new control configuration with a software change, I think it's far outweighed by having something physical to hold when operating the controls.
… and one wouldn't want the controls one is used to changing anyway …
Plus, fixed touchscreens on cars, planes, equipment, etc. tend to be resistive ones, and those can be very annoying to use even when they're perfectly still.
I was worried about the issue, but in light turbulence, it’s no issue IME and in moderate, it’s maybe less certain than knobs, but I bet even there that I could still enter an accurate flight plan update, whether a new full route and just a direct-to shortcut) faster on the touch GTN 750 than on an old (multi-knob) GNS 530. (That’s not just touch; it’s partially the discoverability and access to “more than the -D-> key” functions.)
I am glad the critical autopilot inputs (HSI heading bug) and other flight controls are not touch, though.
Also, knobs in turbulence are fun. "Turn left 15 degrees" can become a random amount on your heading bug with an unfortunately timed thump!
Anyway, bottom line is that I find it easier to see the relative angles and trends of the pointers in old analog gauges in turbulence than read the tapes. The angles meant something easily interpreted as being unexpected at a glance even if the absolute value might take another moment to read.
Imagine if our AI were two tapes, like the EFIS horizontal and vertical deviation indicators!!