by Thomas Lee
There has been growing concern that high-tech reinforces racism by embedding historical bias and harmful stereotypes in algorithms that pretty much run everything in the world today.
The emergence of artificial intelligence programs only heightens those dangers, especially given the growing popularity of advanced AI chatbots like OpenAI’s ChatGPT.
“AI systems have the potential to deepen existing systemic inequalities, particularly in industries like healthcare, employment, and criminal justice,” according to the Georgetown Security Studies Review. “This injustice could lead to growing distrust in private and governmental institutions as marginalized groups face further discrimination.”
That’s because the algorithms that power AI are themselves trained on information created by humans, with all our biases, both overt and covert.
However, even seemingly ordinary programs like autocorrect/spell check programs can be suspect.
I was previously working on a story for The Boston Globe about 1587, an Asian American sneaker company whose brand is synonymous with cultural pride.
I normally write the draft in Google Docs before cutting and pasting the file into the Globe system. As always, I ran a spelling and grammar check. Google flagged two instances where I wrote the word “message.”
At first, I thought I misspelled the word. I didn’t. I then noticed Google in both cases suggested the correct spelling was “massage.”
Here’s where it gets interesting. In each case, I wrote “message” after the phrase “unapologetically Asian,” which happens to be the company’s motto.
Google’s algorithms apparently detected the word “Asian” and assumed that the word that should naturally follow it was “massage.” (Never mind “unapologetically Asian massage” makes no sense, given the context of my article.)
Therefore, “message,” according to Google, must have been a typo.
Wow.
In case you don’t know, Asian women who work in massage parlors are often stereotyped as giving sexual favors to clients. The association is part of a long history in the Western world of fetishizing Asian females.
For Google to naturally associate the word “massage” with “Asian” instead of my intended word of “message” really says something.
“As our society comes to terms with this history and seeks to understand and dismantle the structure of American racism, it will be necessary to acknowledge the role of technologies such as autocorrect,” according to a paper published in the Boston University Law Review.
“Autocorrect provides the tools and template for structural racism and Anglo normativity to flourish today.”
AsAmNews is published by the non-profit, Asian American Media Inc.
We are supported through donations and such charitable organizations as the Robert Wood Johnson Foundation. All donations are tax deductible and can be made here.
Please follow us on Instagram, TikTok, Facebook, YouTube and X.
This is an important and well written article about a serious concern. A concern that is in fact very broad based and just like the author mentions, and a possible (likely??) propagator of a variety of misconceptions and prejudices.
I hope we can see more articles on this subject, with illustrations such as included herein
Thank you for documenting something that I have been wondering about ever since I learned where the AI ‘knowledge’ was actually coming from.
Having read this interesting and important article just yesterday, I find the below article in my morning news feed…… As AI takes the helm of decision making, signs of perpetuating historic biases emerge …….. https://missouriindependent.com/2024/10/11/as-ai-takes-the-helm-of-decision-making-signs-of-perpetuating-historic-biases-emerge/
This is very interesting to me, and intersects two areas I am working on — the use of AI in media and its ethical challenges (like bias); and the newly relaunched AAJA style guide. Your example shows that words really do matter. I would like to use this example in an upcoming webinar we are doing on the AAJA style guide.