ChatGPT’s responses are still littered with inaccuracies and OpenAI has not done enough to remedy the issue to make the platform GDPR compliant, according to EU regulators.
According to a task force focused specifically on examining the use of ChatGPT, OpenAI’s flagship chatbot is problematic in that the content it generates is often likely to be taken as fact despite inaccurate answers to user queries.
“The outputs provided by ChatGPT are likely to be taken as factually accurate by end users, including information relating to individuals, regardless of their actual accuracy,” the task force said.
Lawmakers advised OpenAI to take the necessary steps to ensure that users understand that ChatGPT’s “generated text, although syntactically correct, may be biased or made up”.
Though OpenAI has taken some measures, according to the watchdog, it has not done enough to make the platform fully compliant with GDPR standards and principles.
“Although the measures taken in order to comply with the transparency principle are beneficial to avoid misinterpretation of the output of ChatGPT, they are not sufficient to comply with the data accuracy principle, as recalled above,” the task force said.
The ChatGPT task force was set up last year by the European Data Protection Board (EDPB) “to foster cooperation and to exchange information on possible enforcement actions conducted by data protection authorities”.
The move followed enforcement action undertaken by the Italian data protection authority in April 2023 which saw the Italian government ban ChatGPT. Italy has since lifted its ban on the platform.
Accuracy is an ongoing concern in generative AI
CPO and CTO of xDesign Jeff Watkins told ITPro that accuracy is a difficult issue for generative AI platforms and large language models (LLMs).
Under GDPR, reasonable measures must be taken to remove inaccurate data, and regulators have grown concerned that some generative AI platforms are in breach of the legislation.
“If the model itself is generating inaccurate data due to hallucinations, trying to fix these errors is like playing whack-a-mole with a probability engine,” Watkins said.
ChatGPT has found itself susceptible to issues of inaccuracy, with the platform having been found to provide inaccurate answers to coding questions over 50% (52%) of the time in 2023.
While OpenAI’s CEO has described inaccuracies and hallucinations as part of the “magic” of generative AI, there is still a clear need for many to consider the practical implications of inaccuracy, especially regarding GDPR.
While notable industry experts such as Dell CTO John Roese have stated recently that hallucinations are no longer much of an issue, platforms such as GitHub Copilot are still returning problematically erroneous code.
According to Watkins, generative AI developers will need to prove their credentials going forward when it comes to data accuracy, showing users “they’re improving on inaccuracies and providing sources”.
