I will describe a new technique for deriving thermospheric vector wind fields from Doppler shifts of airglow/aurora that are remotely sensed using an array of optical spectrometers in Alaska. Deriving these winds is a classic geophysical inverse problem. Such problems can be difficult, because free parameters are often non-linearly related to the model's goodness-of-fit and, in many cases, the observations are too sparse to uniquely constrain the model. Nevertheless, clever state-of-the art techniques can often still generate rigorous, stable, and well optimized models. (E.g. Bill Bristow's Sep 19 Journal Club on SuperDARN ion velocity measurements.) While such approaches would likely work well in this application, they require expertise and effort to implement. Instead, taking a cue from nature (and motivated by apathy), I am letting evolution do the hard work for me. The approach starts with an initial four-dimensional (lon, lat, altitude, time) three-component vector wind field model that is easily derived from the data but is too crude to be realistic or useful. The algorithm then applies random "mutations" to this model, retaining those that improve the model's goodness of fit to the data, and rejecting those that don't. Results after many millions of mutations are remarkably good - detailed and realistic flow structures emerge, and the overall fields match winds derived in other ways. While I am astonished by how effective dumb randomness (and laziness!) can be, the algorithm is not entirely trivial. In particular, I will describe the way the mutations are generated, and how the "survival of the fittest" test is implemented. Some example results will be included.