why does ethernet not use gnd?

ethernet (CAT 5) generally has T+/TX- RX+/RX-
why cant the '-' be replaced by a single gnd?

and also usb for that matter

The way its done, by revolving each pair differently and also revolving all pairs, ensures that RX / TX does not interfere with eachother. It also allows that the two ends can have different GND-levels (avoiding gnd-loops)
At each reveiving end its the difference voltage that counts - not the voltage referenced to GND.
All in all: No ground loops, allows for greater distance, resistant to noice along the line.

To prevent all the problems with ground loops.

Ethernet signals are send down a twisted pair differential transmission line - the symmetry
of the D+/D- wires is crucial to the system behaving as a transmission line (ie constant
impedance all the way along to the receiver). Such differential or balanced lines perform
loads better than unbalanced transmission lines because all common-mode interference
cancels out. With an unbalanced line like coax the common-mode interference sees
the outer and inner conductors as different (the outer is earthed, for instance), and different
amounts of energy are picked up by each. This means a differential signal has been
injected onto the transmission line from a common-mode source - ie you cannot eliminate
common-mode interference.

Its the symmetry of differential twisted pair that gives it the immunity to common-mode
interference, so the wiring has to be symmetric and the driver has to drive the wires
in anti-phase.

Typically small transformers are used in ethernet to allow for varying system grounds
at each end, so its possible for the signal wires to be many volts away from local
ground.

Forgive me, if this was already clear to all readers....

With Ethernet... and RS-485, for the same reasons. And others no doubt... the design specifies the two states we need (to send "1"s and "0"s down the line) thus...

Take two wires. We'll call them A and B, because the usual names (D+ and D-) might draw your brain down WRONG paths....

(I may have the following "backwards", by the way, but even if I do, it explains the principle...)

If when you measure the voltage between A and B, you find that A is "above" B, then you call that a "1". And, now it should be obvious, if you measure the voltage between A and B and B is "above" A, you call that a "0".

WHY you do it this way? Already covered in previous posts. I hope WHAT is being done may now be more clear, if you didn't understand previously.

All in all, knut_ny pegged it in one. It is because it is not a matter of connecting wires, it is a matter of transmission lines and the cheapest transmission line technology is balanced pairs.

The only other technology for transmission lines is coaxial cable and that has proven too expensive and impractical for Ethernet (but was in fact, the first technology used). Strictly speaking, there are two other technologies available, waveguide and optical waveguide which is now used(, called "optical fibre").

And for completeness surface-acoustic waves too, but that only works on the surface
of an appropriate kind of material, as in SAW-filters.

Tin-can and string toy "telephones" are actually a transmission line too, thinking about it!

The key property of transmission lines is that they have no frequency dependence of
their properties (to a good approximation and within limits), so high frequency signals
propagate efficiently and without distortion.