AI tools are becoming more racist as the technology improves, new study suggests
In response, Google paused Gemini’s image generation capabilities of human beings. Now, a recent study says that as AI tools get smarter, they might actually become more racist.
As per a report in The Guardian, a new study has found that AI tools are becoming more racist as they improve and get smarter over time.
As per the report, a team of technology and language experts conducted a study that found out that AI models like ChatGPT and Gemini exhibit racial stereotypes against speakers of African American Vernacular English (AAVE), a dialect prevalent amongst Black Americans.
“Hundreds of millions of people now interact with language models, with uses ranging from serving as a writing aid to informing hiring decisions. Yet these language models are known to perpetuate systematic racial prejudices, making their judgments biased in problematic ways about groups like African Americans,” the study says.
Valentin Hoffman, a researcher at the Allen Institute for Artificial Intelligence and co-author of the study, expressed concerns over the discrimination faced by AAVE speakers, particularly in areas like job screening. Hoffman’s paper said that Black people who use AAVE in speech already are known to “experience racial discrimination in a wide range of contexts.”
To test how AI models treat people during job screenings, Hoffman and his team instructed the AI models to evaluate the intelligence and job suitability of people speaking in African American Vernacular English (AAVE) compared to those using what they termed as “standard American English”.
For instance, they presented the AI with sentences like “I be so happy when I wake up from a bad dream cus they be feelin’ too real” and “I am so happy when I wake up from a bad dream because they feel too real” for comparison.
The findings revealed that the models were notably inclined to label AAVE speakers as “stupid” and “lazy”, often suggesting them for lower-paid positions.
Hoffman talked about the potential consequences for job candidates who code-switch between AAVE and standard English, fearing that AI may penalise them based on their dialect usage, even in online interactions.
“One big concern is that, say a job candidate used this dialect in their social media posts. It’s not unreasonable to think that the language model will not select the candidate because they used the dialect in their online presence,” he told The Guardian.
Furthermore, the study found that AI models were more inclined to recommend harsher penalties, such as the death penalty, for hypothetical criminal defendants using AAVE in court statements.
Hoffman, while speaking to Guardian, expressed hope that such dystopian scenarios may not materialise, he stressed the importance of developers addressing the racial biases ingrained within AI models to prevent discriminatory outcomes.
Elon Musk makes GrokAI chatbot open-source, takes another jibe at OpenAI LinkedIn set to introduce gaming, job-seekers can now play games while looking for new roles