AI Bias: An Unsolvable Problem

Every time I hear discussions about preventing AI bias I am wondering if this is actually possible. Of course, what is meant in these discussions is the need to develop systems with less harmful outputs. As positive as this is, I think that most discussions are rooted in a problematic assumption of how we relate to reality. What is often considered as AI bias, is not an error in misinterpreting how the world is, it is a feature of specific artificial reality constructions that emerge from the way a machine experiences the world. What is usually meant when people speak about AI bias is that the worldview of an AI system is not aligned with a specific human worldview. If we treat AI bias as a solvable problem we tend to solve it once by aligning this worldview gap and then consider the work as done. But values change over time and are never resolved, which means that an unbiased AI still holds biases. We would just not call them biases because artificial and human worldviews are aligned. Hence, we need to acknowledge that a problem-solution logic cannot be applied to AI bias, because a solved problem gets pushed out of focus. And this narrows the window for further worldview transformation and thereby stabilizes existing worldviews. This might be good in some cases but can also be very problematic. Instead, I think that bias has to be discussed as what it is: an unsolvable problem that needs constant transformation. From a less anthropocentric perspective this does not just refer to AI but also means that human value systems often need alignment with other existing and emerging value systems.

Nov 14, 2025