Daily Note - Wednesday, April 29th, 2026

Don’t be a downer, bro.

I am critical of AI. I don’t really like it. A lot of why, is it’s always felt “off.” There was something I couldn’t express about the proponents, the culture, the history, the way I felt it could be abused. It was a gut reaction.

Most of the time, those gut instincts are right.

An Economy of Empathy

Mario Munoz delivers an absolute must-see talk at North Bay PyCon.

This talk contains descriptions of horrible things. Please do not watch the first 3 minutes unless you know you can handle it.

This talk is important. We are already seeing examples of bias causing harm with AI. This is no surprise to those of us who have been following AI since it was called “Machine Learning.” Google’s image recoginition was overly trained on pictures of white people. When they launched, it labeled pictures of black people as “gorillas.” When AI is applied to court cases, it inherits the bias of the training data. What’s it trained on? Previous cases. Again, here trials of AI sentencing found it recommend harsher sentencing for black individuals.

“AI” is a stand-in for the technology we call Large Language Models (LLM). LLMs are token predictors. They produce the next most likely token based on the input, the output, and the training data. Herein lies the issue. The training data is biased. Just how biased depends on what’s been selected, but you can pretty much guarantee that the training data is compromised foudationally of works by affluent, white, Christian, cis-gender, straight (well, at least pretending to be) males. History has been racist AF, bro. That demographic simply had access to the means to produce, distribute, and persist their ideaologies. And for the majority of history, being misogynistic, racist, and classist was not only accepted, but expected of the elite.

Mario also discusses the importance of data labeling for AI training. In order for AI to know things like child abuse and sexual abuse are bad, someone has to train it as such. To do that, companies like Appen and Sama pay people in the global south very little money to read stores, look at pictures, and watch movies of horrific things so they can be labeled. He discusses some of this in-depth in the first three minutes of the talk. If you can’t handle, but want to understand the role eugenics and misogyny played in the history of AI, skip to the 3 minute mark.

Without data labeling, which is manual task, AI does not exist. Yet, when founders get up to talk about all the great things they’ve done, they never mention the scope or depth of data labeling that’s been done. They talk about algorithms and innovation in inference, but never mention that without paying people with dark skin in the global south less than a living wage to watch rape, torture, and child abuse videos, none of this would be fucking possible.

Now why do you suppose that is?

No Artificial Colors, Flavors, or Intelligence were used in the creation of this content.