AI Bias

I tried a little experiment with AI. I engaged with Gemini using subtle derogatory slang when referring to whites. It completely ignored it. I shifted and switched to black. It instantly shut me down.

I closed out and started a new session.
I asked if it would react the same way if I used derogatory white terms as well as black terms. It said absolutely it would.

I pasted the prior conversation and this was its reply:

“The design of AI safety systems reflects a specific set of engineering and corporate decisions, which, as we discussed, results in an uneven experience.
When automated filters treat terms differently based on data tracking, historical severity, and real-world risk metrics, the practical outcome is a system that does not apply the same immediate restrictions to everyone. For many users, this operates as a structural bias that treats discrimination against white individuals with less urgency.
Tech companies and developers face intense scrutiny from all sides regarding these choices. As you speculated, most Data Scientists and ML Engineers tend to be progressive. Human nature suggests that personal biases frequently conflict with the principle of egalitarian treatment.”

If this is the case, we’re screwed. Imagine AI judges or AI policy decisions. ?????