[jnurmine]

"Draft changes would create new exceptions for AI companies that would allow them to legally process special categories of data (like a person’s religious or political beliefs, ethnicity or health data) to train and operate their tech."

Fully anonymized health data I can somehow understand, but what kind of AI needs to be trained with "a person's religious or political beliefs [or] ethnicity", anonymized or not?

[Muromec]

>but what kind of AI needs to be trained with "a person's religious or political beliefs [or] ethnicity", anonymized or not?

It's a category thing. There is personal data and then there is sensitive personal data, which is the enumerated list you quoted.

Snarky non-answer reply is of course -- the kind of ШІ that makes autistic roman salutes and accelerates from 0 to 1933.

[DocTomoe]

Ethnicity can correlate with certain genetic or health predispositions - for instance, the U.S. has long recognized that some conditions (like sickle-cell anemia or hypertension) appear more frequently in Black populations. If AI systems were forbidden from even considering such demographic factors, diagnostic accuracy could suffer.

[zrn900]

> but what kind of AI needs to be trained with "a person's religious or political beliefs [or] ethnicity"

Propaganda AI. The EU openly declared that it will create a ministry of truth. Officially called 'fighting disinformation' of course.

[port11]

These categories of data are already somewhat special under the GDPR. I think we've committed enough genocide that now they're enshrined into protected data types.