The Rhubarb Scare: Historical Lessons on Misinformation in the AI Era

In a cautionary tale from the World Wars, British authorities made a grave mistake by recommending rhubarb leaves for consumption, inadvertently causing illness and death among citizens. These leaves contain high levels of oxalic acid, which is toxic to humans. This historical misstep serves as a stark reminder of how dangerous misinformation can be when it comes from trusted sources—a problem that has found new dimensions in today’s digital landscape.

The rise of generative artificial intelligence has dramatically amplified this risk. Tools like ChatGPT, Claude, and other large language models are producing content at unprecedented rates, creating text that sounds authoritative and believable but often contains factual errors or fabricated information.

Unlike traditional search engines that retrieve existing information from indexed websites, these AI systems predict word patterns based on their training data. They don’t “know” facts in the way humans understand knowledge. Instead, they generate responses by predicting what words should follow others, creating content that mimics human writing but lacks true understanding or verification mechanisms.

“The fundamental problem is that users approach these platforms as if they were search engines,” explains Dr. Emily Bender, a computational linguistics expert at the University of Washington. “When someone asks ChatGPT a question about health or finance, they’re often expecting a researched answer, but they’re actually getting a probabilistic text prediction that may have no basis in reality.”

This phenomenon has been termed “hallucination” in AI circles—the generation of content that seems plausible but is factually incorrect or entirely fabricated. These hallucinations can range from harmless inaccuracies to potentially dangerous misinformation, especially when they touch on critical areas like medicine, law, or public safety.

As AI becomes increasingly embedded in our information ecosystem, the stakes grow higher. Political campaigns are already using AI to generate content and messaging. Healthcare providers are experimenting with AI for patient communication and even preliminary diagnoses. Financial institutions deploy these tools for customer service and advisory functions.

“We’re witnessing a fundamental shift in how information is produced and consumed,” says Mark Thompson, a technology policy analyst at the Center for Digital Innovation. “Unlike the rhubarb leaf scenario, where misinformation came from a single authoritative source, AI-generated misinformation is being produced at scale and from countless sources simultaneously.”

Experts recommend several approaches to mitigate these risks. First, users should maintain healthy skepticism toward AI-generated content, especially on specialized or technical topics. Second, organizations deploying AI should implement robust fact-checking mechanisms and human oversight. Third, AI developers must continue improving their systems to reduce hallucinations and clearly communicate the limitations of their technology.

The most effective strategy may be to treat AI as a complementary tool rather than a replacement for traditional information sources. “These models are trained on pre-AI era data, which means they’re essentially regurgitating and recombining existing human knowledge,” notes Dr. Robert Chen of the Institute for Information Integrity. “They’re excellent at summarizing and synthesizing what humans have already written, but poor at verifying whether that information is correct.”

For critical decisions, especially those affecting health, finances, or safety, experts recommend relying on established, verified sources of information and using AI outputs as secondary resources that require validation.

The British government’s rhubarb leaf recommendation during wartime food shortages ultimately taught valuable lessons about vetting information before public dissemination. Today’s AI revolution presents a similar learning opportunity on a much larger scale—how to balance the benefits of powerful new information technologies against their potential to mislead when not properly understood or managed.

As AI continues to evolve, developing literacy about its capabilities and limitations may become as essential as traditional reading and writing skills were in previous generations.