AI detection programs discriminate against non-native English speakers, study finds
There have been instances in the past where humans have been discriminated against in society, but a new study has revealed that we might not be the only ones to do so. Generative AI has seen its popularity soaring, especially since the launch of ChatGPT, and measures to mitigate its misuse, such as cheating in exams, have also been developed in the form of AI detection programs. These programs can examine the content and reveal whether it was written by a human or an AI program. However, now, these programs have been accused of shocking discrimination against non-native English speakers.
Yes, Generative AI has previously been accused of exhibiting biases and now a new study has shed light on its detection programs also being capable of discrimination.
Discrimination by AI detection programs
According to a study led by James Zou, a biomedical data science assistant professor at Stanford University, computer programs that are used to detect the involvement of AI in papers, exams, and job applications can discriminate against non-native English speakers. The study, published in Cell Press, was carried out by screening 91 English essays written by non-native English speakers through 7 different programs that are used to detect GPT, and the conclusions might shock you.
As many as 61.3 percent of the essays that were originally written for the TOEFL exam were flagged as AI-generated. Shockingly, one program even flagged 98 percent of the essays as the creation of an AI program.
On the other hand, essays written by native English-speaking eighth graders were also submitted to the program, and nearly 90 percent of them came back as human-generated.
How do these programs work?
To detect the involvement of AI, these programs examine the text perplexity, which is the statistical measure of how a generative AI model predicts the text. It is considered low perplexity if the LLM is able to predict the next word in a sentence easily. Programs like ChatGPT generate content that is low perplexity, meaning it uses simpler words. Since non-native English speakers also tend to use simpler words, their written content is susceptible to being falsely flagged as AI-generated.
The researchers said, “Therefore, practitioners should exercise caution when using low perplexity as an indicator of AI-generated text, as such an approach could unintentionally exacerbate systemic biases against non-native authors within the academic community.”
There have been instances in the past where humans have been discriminated against in society, but a new study has revealed that we might not be the only ones to do so. Generative AI has seen its popularity soaring, especially since the launch of ChatGPT, and measures to mitigate its misuse, such as cheating in exams, have also been developed in the form of AI detection programs. These programs can examine the content and reveal whether it was written by a human or an AI program. However, now, these programs have been accused of shocking discrimination against non-native English speakers.
Yes, Generative AI has previously been accused of exhibiting biases and now a new study has shed light on its detection programs also being capable of discrimination.
Discrimination by AI detection programs
According to a study led by James Zou, a biomedical data science assistant professor at Stanford University, computer programs that are used to detect the involvement of AI in papers, exams, and job applications can discriminate against non-native English speakers. The study, published in Cell Press, was carried out by screening 91 English essays written by non-native English speakers through 7 different programs that are used to detect GPT, and the conclusions might shock you.
As many as 61.3 percent of the essays that were originally written for the TOEFL exam were flagged as AI-generated. Shockingly, one program even flagged 98 percent of the essays as the creation of an AI program.
On the other hand, essays written by native English-speaking eighth graders were also submitted to the program, and nearly 90 percent of them came back as human-generated.
How do these programs work?
To detect the involvement of AI, these programs examine the text perplexity, which is the statistical measure of how a generative AI model predicts the text. It is considered low perplexity if the LLM is able to predict the next word in a sentence easily. Programs like ChatGPT generate content that is low perplexity, meaning it uses simpler words. Since non-native English speakers also tend to use simpler words, their written content is susceptible to being falsely flagged as AI-generated.
The researchers said, “Therefore, practitioners should exercise caution when using low perplexity as an indicator of AI-generated text, as such an approach could unintentionally exacerbate systemic biases against non-native authors within the academic community.”
Denial of responsibility! Techno Blender is an automatic aggregator of the all world’s media. In each content, the hyperlink to the primary source is specified. All trademarks belong to their rightful owners, all materials to their authors. If you are the owner of the content and do not want us to publish your materials, please contact us by email – [email protected]. The content will be deleted within 24 hours.