GPS speed could be used to have an approximation to real airspeed in case your air data is lost. The error would be huge if you donīt program it carefully (basically wind speed=error). However, if you assume that the readings from air sensors were right to the point in which you had a disagreement, you can calculate the wind at that moment (strictly speaking, an average of the last seconds) and assume a persistent wind vector. The error in the very first moments of "GPS synthesized" airspeed would be negligible, even in a turn. As the aircraft moves and wind conditions change, error will grow, being immense in worst case scenarios and longer times.
Same applies to Inertial speeds.
It could even be possible (maybe some planes do it nowadays, I'm no expert) to compute a real airspeed without air data. At any given moment of the flight, the airspeed is the only unknown parameter of a vector F = m x a, because mass is accurate to some degree from load sheets, acceleration you can measure with the inertial platform and the force is the result of adding the trust of the engine, which is calculated out from engine conditions and tabulated air conditions from GPS altitude, and the lift + drag, which relates to air speed and air conditions with a known characteristic.
In other words, the plane knows its airspeed just by feeling how pitch and thrust translate into acceleration (longitudinal, vertical and lateral) at every moment.
Sure the error will exist (almost every part of the calculation is an estimation), but it would be tolerable (as a backup, say 20 or 30 knots) and it will valid over long periods of time.
This concept of multiple possible ways of calculation of a magnitude is valid with many others (altitude) and to me is one of the things the plane could do before just going all UAS on you.
I dream (it is free) of a dial with a very precise reading when all sensors are working that turns into a less precise reading (a sector instead of a thin needle) when errors are expected, because of alternative calculations. "250 knots with 50 knots error" is a lot better than "250 knots but do not believe it much because another sensor reads different". (And a stick shaker and overspeed clacker on top of it, just to get things interesting).