Different strokes. Industrial sites remain just as noisy as they used to be; the din of mechanical contacts has been oft replaced by poorly- or un-filtered VFDs. These should be used in shielded conduit, but often the ground quality suffers at HF, and you'll never get a stable trace on the scope, on site, again. Not to mention various other sorts of process equipment, from material handling (ersatz van de Graaff generator, say?) to specialty processes like EM heating or impulse systems. In these environments, the ~10V noise margin of RS's 232, 485, etc. remain as beneficial today as they were then; or even more basic, signaling like 4-20mA loop or, going even further back, teletype loop (~100V 60mA?).
These have always been industrial standards; they're a bit overkill for commercial products, but can certainly be used, and the fact that PCs have always been somewhat interoperable with telecoms stuff -- hence RS-232 UART for example -- makes them a welcome combination. They're also not expensive to implement, 232 for example powered by the obligatory* +/-12V supplies in PCs back then, or 422/485 from 5V and who cares; it's no wonder Apple went with 422 for their serial ports.
*Almost obviated these days, but still provided nonetheless, lol.
PCs certainly didn't need to use them, and logic levels have been in use for a long time -- albeit mainly for internal signals (ribbon cables and such), and either slowish things (keyboard/mouse, unshielded), or any speed with shields (e.g. parallel ports like IEEE-488 or -1284). At TTL levels, higher speeds, and with potentially joining between grounded equipment, shielded cabling is definitely recommended, and so it was done.
I would suppose that PC serial would've been TTL if there was never a need (or expectation) to connect to grounded equipment; or that attaching an interface module would be expected when signaling compatibility is required. (But that would be dumb for something as big as a desktop; perhaps the budget at-homes e.g. Speccy, C64, etc. took such a route though?) Perhaps IBM PCjr is a better representative here -- though it probably can't be called a commercial success, and, I don't really know what all the connections, pinouts and signals are, I'm just assuming they went with bare TTL (or CMOS equivalents).
Today's high-speed interfaces of course are obligatory shielded, as the differential receivers only have 0 to 3.3V compliance range. USB itself isn't even [fully] differential.
But that said, a fair and direct comparison still exists: a classic serial mouse at 9600,n,8,1 can transfer up to ~1kB/s, or 60kB/min; the requirement in this thread is 50kB/10min so it's a fair order-of-magnitude comparison -- if of course we set aside the bursty requirement. But if we upgrade to a USB HID standard, that's in the 500kbps range, which will still do the job in some seconds, is a pendant device with no local ground/earth connection just the cable, and is ballpark similar cable length (usually a bit shorter than 2m, but that just scales the upper cutoff for noise immunity).
...Actually if you have a USB low-speed driver in that MCU (
can even be bit-bang, though you might not have enough CPU bandwidth otherwise to actually use it this way?), this might be a very promising way to go.
Tim