It's measuring the angle of the plane relative to the air moving past it (i.e. the angle the plane is "attacking" the air). Knowing the angle of the plane in an absolute sense isn't as useful when it comes to detecting stalls, etc.

It's not as useful but it is far from useless. The ordinary artificial horizon indicator should have been able to show that the external sensor was far out. I don't know anything about the aircraft or Boeing but as a software designer and implementor I would have wanted to use the output of the artificial horizon as well as the angle of attack sensor in an attempt to gauge the quality of the readings. I'm guessing here but it seems likely the aircraft was equipped with a normal gyroscopic instrument in addition to the external angle of attack sensor.

Of course the Boeing engineers who worked on this could easily tell us exactly how it works and why. I don't suppose that will ever happen though.

I think the problem is that moving air (such as an updraft) can alter the angle of attack independently of the pitch angle. That is important during landing where low thrust levels are set and the air is turbulent. But in other phases of flight it should be possible to keep the aeroplane stable using know thrust and pitch settings and ignore air speed and AOA sensors completely. Of course that requires fully functional control surfaces and engines. Maybe it would be better to remove the erroneous sensor type completely and have procedures that can cope with that.

Thanks, make sense.

Angle of Attack is based on the angle between the plane and the surrounding airflow, not based on gravity/landscape/etc.