It’s only a flaw insofar as it’s used in ways in which the property of the tool is problematic. Stereotypes are use for good and bad all the time, let’s not pretend that we have to attack every problem with a funky shaped hammer because we can’t admit that it’s okay to have specialized tools in the tool belt.
I guess I’m saying that the specific application of stereotypes may be a feature. I don’t think we’ll see a single prevailing winner takes all model, so there is diversity in that respect too. And I think you will even see diversity from a single model. In other words, I don’t think Daniel Craig is the only thing a model will return for “suave english spy”. Just a cheap and easy one.
I don't follow your analogy. Is the "specialized tool" the AI or the way that it returns "problematic" results? Because I'm not saying the system is bad for using negative stereotypes. I'm saying the system is bad because it removes natural variety from the results making them misleading. The reliance on stereotypes are just one example of this phenomenon with another example being "suave English spy" only returning Daniel Craig.