“Please help me! I can’t seem to calm down without panicking on the ground for a solid 20 minutes. Should I seek medical attention?”
This urgent cry for help emerged in a Reddit post from an individual experiencing panic attacks after using marijuana. While posts like this may often go unnoticed by those in public health, a recent experiment introduced an AI tool named Waldo that was actively observing.
Waldo analyzed over 430,000 previous posts on Reddit forums dedicated to cannabis use and flagged the aforementioned post along with over 28,000 others indicating potentially harmful or unexpected side effects. The researchers subsequently examined 250 of these flagged posts and found that an impressive 86 percent of them indeed depicted problematic experiences with cannabis products, as reported on September 30 in PLOS Digital Health. The implementation of such scanning techniques could empower public health workers to safeguard consumers against hazardous products.
The significance of this research, as expressed by Richard Lomotey, lies in its demonstration that researchers can derive valuable insights from channels that government agencies like the U.S. Centers for Disease Control and Prevention (CDC) may overlook. Traditionally, the CDC collects data through surveys and self-reported side effects of illnesses, but they do not actively monitor social media platforms. “This is where individuals freely express their concerns,” notes Lomotey, an IT expert from Penn State.
It’s vital to recognize that many individuals either lack access to healthcare professionals or may not be aware of formal reporting mechanisms for adverse experiences related to products. As researcher John Ayers from the University of California, San Diego, points out, people openly share health experiences online. “We need to engage with them where they are,” he emphasizes.
Karan Desai, a medical student at the University of Michigan Medical School, explains that the focus on cannabis products stemmed from their widespread use and minimal regulation. “In my age group, particularly those in their 20s, cannabis products are ubiquitous,” he asserts. “Understanding the side effects faced by users is crucial.”
The development of Waldo started with a smaller batch of 10,000 Reddit posts concerning cannabis use. Previous researchers manually identified problematic side effects within these posts. Desai and his colleagues then trained Waldo on part of this data and evaluated its performance on the remaining posts. They found that Waldo outperformed ChatGPT in this specific context, with the latter generating 18 times more false positives by misclassifying benign posts as side effects. However, Waldo did not exceed the accuracy of human reviewers during the testing phase.
After this initial training, Waldo was then able to successfully flag the panic attack post and thousands of others.
It is yet unclear if Waldo would maintain its effectiveness in assessing side effects linked to other substances, vitamins, or products, according to Lomotey. AI systems tend to be finely tuned for particular tasks and may not perform similarly across different contexts. “We need to approach this with caution,” he warns.
Nonetheless, Lomotey envisions a future where tools like Waldo can monitor social media attentively, with a strong emphasis on ethical considerations. When sporadic side effects are highlighted in public posts, these tools could alert health authorities while ensuring user privacy. He believes this could be immensely beneficial in regions lacking comprehensive systems for monitoring drug side effects.
In the future, tools like Waldo could bridge the gap between individuals seeking assistance and public health professionals willing to help. “Even rare side effects can feel monumental when they affect you personally,” Ayers remarks.
This rewritten article maintains the original message while ensuring a unique expression that is ready to be seamlessly integrated into a WordPress platform.