|
It seems clear that a dev team could whip up a product that tested well (using folks in office, family members, friends, etc.); was trained with datasets that - for whatever reasons - weren't sufficiently varied; and hit some mark of success and pushed it out the door to refine the rest later. It also seems clear that the resulting product could do poorly when recognizing black skin, due not to ill intent but lack of polish with the resources on hand. But something I always wonder when accusations like "white supremacy" are thrown around: is it falsifiable? What evidence would dissuade you from that? - What if both ends of the spectrum do poorly and extremely pale people have problems, too? - What if the threshold is dark black and lighter-skinned black people, Asians, Middle Easterners and other non-white people are able to use it successfully? - What if only a narrow band of light levels work, making it clear their testing range was generally too narrow, not just in skin color? - What if they took care to incorporate black models in testing, but the photo quality (and their own in-house cameras and lighting) overestimated the quality of most home users'? And what of the myriad other things that were done poorly in the software: limited OS support, bugs, excessive memory usage, overall intrusiveness, browser limitations, disallowed mobile devices, lack of multi-monitor support? Do they likewise arise from systematic oppression of some group? What if we dig in and find that white people are more likely to use iPads, Linux, and multiple displays? Most often these accusations flow in only one direction, and that all other flaws or problems are taken to be simply happenstance and noise. Certainly anything that impacts white people negatively will not be automatically seen as anti-white, although in a world with activist devs, such a result isn't incomprehensible. Claims of white supremacy (among other accusations of character) are thus, to my mind, wildly speculative and carry a very heavy burden of proof. |