Non-native English speakers are being unfairly discriminated against by AI-detectors; this is according to a study by Stanford University.
As AI programmes, such as ChatGPT, become increasingly popular, so too does the wave of students using it to complete their coursework. However, 61.22% of essays by TOEFL students are being flagged as AI-generated.
James Zou, professor of biomedical data science at Stanford, has said “[Detectors] typically score based on a metric known as ‘perplexity,’ which correlates with the sophistication of the writing.”
Non-native speakers are likely to score lower on perplexity, which measures lexical richness, lexical diversity, syntactic complexity, and grammatical complexity. As a result, students could face discrimination and be unfairly accused of cheating.
Alongside this bias, AI-detectors can be circumvented by a process known as “prompt engineering.” This means a student could enter previously generated text and ask the AI to “elevate the provided text by employing literary language.”
“Current detectors are clearly unreliable and easily gamed, which means we should be very cautious about using them as a solution to the AI cheating problem,” says Zou.
On how to deal with the problem, some possible solutions include avoiding the use of detectors in educational settings. Developers could also use more sophisticated techniques – such as “watermarks” – rather than perplexity to detect AI.
In an ever-advancing world, it goes without saying that more needs to be done to protect foreign students.
