The rapid evolution of technology in recent times has brought with it significant advances in artificial intelligence (AI). However, these remarkable innovations are not without their own set of challenges. One such issue that has recently been gaining attention across various industries ranging from academia to journalism is the fabrication of information by AI, specifically by OpenAI’s ChatGPT. Recently, Microsoft Corp (MSFT), which trades on NASDAQ, closed a $10-billion multi-year investment deal with OpenAI, bringing this issue even more into focus.
As someone who has personally experienced this phenomenon on several occasions, I sought to verify my concerns and discern the truth. I engaged ChatGPT to retrieve quotes from a specific person by providing it with precise information. However, to my surprise, it provided me with plausible responses in the speaker’s style that were not genuine quotes. Even when provided with ample context and explicit directions not to fabricate quotes, ChatGPT still produced invented remarks attributed to the person and enclosed them in quotation marks as if the source had said them verbatim. This practice is a gross misrepresentation as the person had not uttered those words. The need for rigorous fact-checking became more evident as the process continued despite careful measures to stick to the guidelines.
This growing concern was brought into sharp focus by the study, “High Rates of Fabricated and Inaccurate References in ChatGPT-Generated Medical Content,” which was conducted by Mehul Bhattacharyya, Valerie M. Miller, Debjani Bhattacharyya, and Larry E. Miller. The research analysed 30 medical papers generated by ChatGPT-3.5, each containing a minimum of three references. Researchers found that out of the 115 references generated by the AI, a staggering 47% were completely fabricated, 46% were authentic but used inaccurately, and only 7% were authentic and accurate.
This issue is even more severe in the journalism industry as attributed quotes and articles can cause a significant misrepresentation of facts. AI-generated fake news and sources undermine the credibility of legitimate news sources, propagating disinformation and damaging reputations. Several instances show the frequency at which ChatGPT creates fabricated citations, bylines, and even non-existent articles. Journalists have found their names falsely associated with articles and sources, resulting in widespread concern about journalistic integrity.
The issue of AI-induced fabrication is not limited to journalism, but has also impacted the legal profession. ChatGPT’s ability to fabricate references for non-existent research studies raises concerns about legal cases and the potential for disinformation in the field.
OpenAI is attempting to mitigate the issue of hallucinations by refining their models and exploring ways to allow users to customize ChatGPT’s behaviour, while still preventing misuse. Strategies to reduce the incidence of hallucination include training AI on a more narrow and vetted dataset or improving the user interface to indicate the confidence level of the generated text. However, these solutions come with their own set of challenges such as data privacy concerns and technological feasibility.
Despite the challenges, the promise and potential of AI are undeniable. Balancing the promise and perils of AI requires vigilance, discernment, and steadfast dedication to truth and accuracy. As technology continues to evolve, it is crucial to mitigate the risks of AI-induced fabrication while leveraging its text-generating abilities to the fullest extent possible without compromising the veracity of information.
Image credits: Alexander Limbach on Shutterstock and logo chatGPT on Wikipedia.
