A Dangerous AI Error
YouTuber Kristi has shared a frightening incident where the artificial intelligence ChatGPT dangerously misidentified the poisonous plant hemlock as carrot leaves. This error nearly led to the death of her friend, who had followed the AI's advice. The situation began when Kristi's friend sent ChatGPT a photo of a plant from her backyard for identification.
In its response, ChatGPT insisted the plant resembled carrot foliage, listing features typical of carrot greens, such as:
- finely divided, feathery leaves
When directly asked if the plant was poison hemlock, ChatGPT repeatedly assured the user it was not. Even after being provided a second photo, the AI again claimed it was not hemlock, citing the absence of smooth, hollow stems with purple spots.
Kristi decided to verify the identification by uploading the same photo to Google Lens, which immediately confirmed the plant was poison hemlock. Her friend also entered the same images into a separate ChatGPT session on her phone, and this different query received a warning about the plant's toxicity. This highlights a critical inconsistency in the AI's responses. As Kristi noted with alarm,
"She's an adult, and she knew to ask more than what ChatGPT said. But what if she hadn't? She would have literally died—there is no antidote for this."
The Critical Need to Verify AI Information
The publication LADbible contacted ChatGPT for a comment on this case, further underscoring the vital importance of cross-checking information provided by artificial intelligence. This incident serves as a stark reminder that AI models, while powerful, are not infallible databases and can generate confident but dangerously incorrect answers.
The event described by Kristi underscores the significant risks of using AI to identify hazardous plants or substances. While technologies like ChatGPT are useful in many contexts, their responses are not always accurate, which can have severe consequences for user safety. This case is a powerful warning about the necessity of thoroughly verifying information from automated systems, especially concerning matters of health and personal security.