It is not the "dispassionate dividing of the world into rigid categories" necessary in order to implement software that is morally or ethically problematic, it is the imposition of those rigid categories upon society at large, perhaps by itself or perhaps merely just the conflation of the separation between the categories that gets incorporated into the design as a subconscious or even conscious choice to reify the categorical differences that is the problem. It is not the knowledge itself that is the problem, it is how that knowledge gets incorporated into an overly simplistic and heavy-handed approach to act. Indeed that is the very thesis of the
Seeing like a State book.
To take the first example of scientific forestry that replaced the local control over the land in which the categories were the species of plants as an example, it was not the factual knowledge about the species of plants, which is an observation about the true nature of reality, it was the decision to use that fact about the distinction between the plants and decide to separate the plants geographically to uniform monoculture plots rather than allowing them to be mixed together (and, in the process, in fact destroying things that they did not understand, the ecosystem) that is argued to have been the cause of the scientific forestry failure. The plots were just fine before and after the knowledge about what plants were where and what the different types of plants there were became known; it was the act of separating the plants into the monoculture plots. The knowledge was a necessary prerequisite to do so, but it was not sufficient, and I don't know if I could believe that accurate knowledge of actual pre-existing categories that correspond to the actual universe could by itself be wrong. Certain plants really are different than other plants, species actually exist. What you decide to do with that information is up to you, but I don't know if I could support the suppression of the acquisition of knowledge merely because it might (but also might not) be used in ways we disagree.
If they had merely studied the plants, that would not have been wrong. If they had created a map of where the plants were, that also would not have been wrong in my view. It was when they decided to change reality that the moral implications attach, in my view. They could have mapped out the plants and divided the map into categories: "there are oak trees here and not here", or "there are strawberry bushes here, and not here", and that is "dividing the world into categories" (namely: presence or absence of certain plants in certain locations), but that is merely the true expression of factual reality. It was how they decided as a separate step, "let's make all the strawberry bushes be in the same location" which incorporated many smaller actions like "let's move this strawberry bush from here to here" or "let's remove all the plants we have not specifically identified, as we (falsely) believe that they are not essential" in which reality was actually changed, and thus the moral or ethical implication attaches in my view. Those who act have a duty to act responsibly, to do no harm, and so on, and the duty scales with the scope of the action, and perhaps there is a scope of action which by itself is too large, the standard or duty of care too impossible to satisfy due to the sheer complexity, for any one person or collection of people to rightfully possess what would be obligatory in my view to have the permission to act.
I would also not view the mere publication of software as action per se either. In your example, it is the utilization of the marketing channels to push the software onto users with or without their knowledge, to exercise editorial control over what software may or may not be used on one's supposedly own device, and so on that the duty applies. If someone publishes a program that divides the world into categories, but no one uses the program, it doesn't matter how wrong the program is. The program can have a field for skin color that is literally black and white for the purpose of calibrating an algorithm that divides pixels into skin-or-not-skin, for example to cut out the background, and yet if no one actually uses, much less is forced to use the program then the mechanism of distinguishing between skin-and-not-skin being incorrect and not for example a scalar or some other model does not harm anyone. It's just an overly simplistic program, a piece of text, an incomplete idea, a reflection of a categorical distinction that does not exist in reality but only in the mind of the programmer and in the software, but if no one ever uses the software let alone is forced to use it, who possibly can be harmed by it?