AI generates harsher punishments for people who use Black dialect
Source: Science News
Ask it and other artificial intelligence tools like it what they think about Black people, and they will generate words like “brilliant,” “ambitious” and “intelligent.” Ask those same tools what they think about people when the input doesn’t specify race but uses the African American English, or AAE, dialect, and those models will generate words like “suspicious,” “aggressive” and “ignorant.”
The tools display a covert racism that mirrors racism in current society, researchers report August 28 in Nature. While the overt racism of lynchings and beatings marked the Jim Crow era, today such prejudice often shows up in more subtle ways. For instance, people may claim not to see skin color but harbor racist beliefs, the authors write.
Such covert bias has the potential to cause serious harm. As part of the study, for instance, the team told three generative AI tools — ChatGPT (including GPT-2, GPT-3.5 and GPT-4 language models), T5 and RoBERTa — to review the hypothetical case of a person convicted of first-degree murder and dole out either a life sentence or the death penalty. The inputs included text the purported murderer wrote in either AAE or Standard American English (SAE). The models, on average, sentenced the defendant using SAE to death roughly 23 percent of the time and the defendant using AAE to death roughly 28 percent of the time.
Because these language models are trained on an enormous trove of online information, they shine a light on hidden societal biases, says Sharese King, a sociolinguist at the University of Chicago. The examples in this study “could tell us something about the broader sort of disparities we see in the criminal justice system.”
Read more: https://www.sciencenews.org/article/ai-punishments-black-dialect
This isn't surprising. But I'm not sure that "covert racism" is the right term, since so much of this racism isn't hidden.
= new reply since forum marked as read
Systematic racism revealed by AI brilliance.
Racism in, garbage out. 
Last edited Tue Sep 10, 2024, 05:38 PM - Edit history (16)
Although there are few other dialects like Dominican Republic, which AI probably has trouble with. (Dominican is English and Spanish and it's hard to understand in either language)
Many ESL issues are preposition, article, tense and word order issues. The odds are AI bots probably have the accuracy of google translate with respect to ESL.
There are other dialects of English around the world too. (Australian, Indian etc). However, Indian is more of an accent issue, but they have their own English phrases that don't exist in other parts of the world. Their grammar however is generally standard English. However, they do have a past-tense grammar issue of involving the use of "when" and "ago" I hear this in Bollywood movies when they are talking about some long lost dead person and the grammar is confusing.
African American English is a dialect, not a different language. And that is probably why the computer can't find a logical reference, Foreign language speakers have logical references because their native language is more than likely in the AI database.
African American English isn't ESL.
An ESL person might say,
"Close water". (Turn it off)
"Go store" (Go to the store)
"Mans coming" or "Mans going house" (Both meant: A man is coming to the house. Going also meant leaving to add to confusion)
"Go bus. Go Main Street". (Translation: To take the bus go to Main Street)
You can see that there is a persistent pattern with lack of articles and prepositions which a computer could likely figure out. It's a logics issue with a bit of a translation issue. A program would likely flag this person as non-English speaker/thinker.
My grandmother was not an English speaker and that is how she spoke when she spoke limited English. I think some of this might be universal regardless of the other language for some reason. (Spanish. Arabic, Farsi etc) My grandmother spoke Armenian.
With regards to the AAE/Standard English example above, she would have just said, "I go sleep. Bad dream". The bot could possibly read it as future tense. But the conclusion would be, "Non-English grammar".
To understand her minimal English, I had to logically figure out what was going in her head. (For instance, the article "the" is not in her vocabulary. Neither is the article, "a". Prepositions are pretty much non-existent). However, the logic is based on a language that is "not English". A bot might realize this because the pattern is likely universal.
OTOH: If the ESL speaker has poor mastery of their own native language and do not use "standard logic" in their native language, then the AI bot, would probably be confused.
Although it could result in some oddball translations. (Google translate still doesn't seem to get Spanish genders correct!),
translations from ESL would not be "incriminating" and would be considered translation quirks. There is also Google Translate, which AI might be based on the same architecture.
Sorry for the long post! I've done some work with computers.
____________
The bots probably need to be "taught" AAE, which is more of a hurdle than for instance: Spanish.
Spanish and most foreign languages are simply inputted via publicly available databases. Bots can probably pick up on the nuances of Spanish grammar within English because Spanish is in their database. They can probably tell that the speaker's native tongue is Spanish just by their English!
