Just before the weekend, I stumbled over this article: Algorithms That Adjust for Worker Race, Gender Still Show Biases.

There are a couple of obvious questions that the title alone provokes:

  • Is it even possible to create technical systems without bias? Wouldn’t this require either the existence of a »ground truth« or a societal agreement on specific values that then can help specify a substitute ground truth?

  • Can or will we ever agree on what a bias-free system is? What is bias when there is not ground truth? A bias-free technical system has to reflect a reality that does not exist yet. Can we, as a society, agree on how this reality should be?

  • Isn’t one of the sources of bias the fact that technical systems are inevitable based on categorizations, i.e. the need to assign any non-continuous values into predefined categories so they become processable?

I don’t want to disqualify the endeavor for bias-free systems; whenever bias is visible, we have to work to remove it. But ultimately, we have to ask whether this is a goal we can ever reach, just like Achilles will never catch the turtle.

image created with midjourney from the prompt »engineers trying to build a bias-free machine, in the style of star trek TOS«

Comment