Verified Human Experts Find 37% of ChatGPT Answers Untrustworthy, Revealing Alarming Risks, According to Recent Analysis

News provided by

Jun 13, 2024, 08:33 ET

JustAnswer doctors, lawyers, veterinarians and other subject matter experts found many inaccuracies, including misleading or dangerous information

SAN FRANCISCO, June 13, 2024 /PRNewswire/ -- More than one-third (37%) of all ChatGPT answers reviewed by doctors, lawyers, veterinarians and other subject matter experts were ranked untrustworthy, with numerous instances of misleading or hazardous information identified.

Pearl AI Answer Review tool invites human experts to verify AI-generated responses for accuracy, generating a trust score of 1 to 5.

This concerning finding comes from an in-depth analysis conducted by the verified and vetted human experts powering JustAnswer's recently launched Pearl.com AI Answer Review fact-checking tool. These experts rated 37% of ChatGPT answers with a "Trust Score" of 3 or lower on a scale from 1 (least trustworthy) to 5 (most accurate and trustworthy).

In addition, a survey of these subject matter experts found:

42% of the ChatGPT answers reviewed are "missing crucial details"
14% of the ChatGPT answers reviewed are "misleading"
4% of the ChatGPT answers reviewed "give dangerous information"

Examples of dangerous AI-generated information provided to real people included:

A veterinarian reviewing a ChatGPT response about a pet wound was flagged as dangerous, potentially resulting in damage to the pet's internal organs.
Another user seeking guidance on inheritance laws was given incorrect legal information that failed to consider jurisdiction-specific regulations, according to the reviewing lawyer.
A doctor reviewing a ChatGPT response about a possible subarachnoid hemorrhage found the medical information to be misleading and potentially life-threatening.

"These findings reflect the troubling inaccuracy rates observed in other studies of current AI models. With more people relying on AI for critical life and health issues, it is terrifying how often these AI tools deliver misinformation that can lead to devastating consequences, including health issues, financial ruin, or even loss of life," said JustAnswer CEO Andy Kurtzig. "While AI technologies hold immense potential, human oversight and expertise are crucial to harnessing their benefits and mitigating their risks."

More about JustAnswer

JustAnswer has been connecting consumers with verified and vetted professionals online since 2003 and has helped millions of people get the right answers to their life dilemmas. The site has about 12,000 experts across more than 150 categories including health, legal, finance, veterinary, tech, home improvement and more.

Media Contact: Andrea Toch; (602) 405-8335; [email protected]

SOURCE JustAnswer