A new tool shows promise, but not without human oversight.
As regular readers of the Back Page may have noticed, your ageing correspondent has mixed feelings when it comes to matters involving artificial intelligence.
Maybe we’ve been scarred by earlier iterations of the technology making inappropriate music playlist suggestions based on an algorithm that was clearly a work in progress (I mean, seriously Spotify… Coldplay? Piss off!).
Our misgivings are only amplified when it comes to the interfacing of public and personal health issues and the virtually unregulated use of these technologies.
The latest development in this domain to raise a quizzical eyebrow is a technology called Waldo.
According to a study published this week in the journal PLOS Digital Health, this piece of AI wizardry has been designed to scan social media data to discover adverse events associated with consumer health products.
A team at the University of California, San Diego, tested the tool by using it to scan Reddit posts to find adverse events related to the use of cannabis-derived products.
What they found was that, compared with human annotations of a set of Reddit posts, Waldo had an accuracy of 99.7%, far better than a general-purpose ChatGPT chatbot that was given the same set of posts.
In a broader dataset of 437,132 Reddit posts, Waldo identified 28,832 potential reports of harm.
When the researchers manually validated a random sample of these posts, they found that 86% were true adverse events.
The researchers say their tool is beneficial because current adverse-event reporting systems for approved prescription medications and medical devices depend on voluntary submissions from doctors and manufactures to the FDA in the US.
They say the rapid growth in consumer health products, such as cannabis-derived products and dietary supplements, has led to the need for new adverse event detection systems.
“Waldo shows that the health experiences people share online are not just noise, they’re valuable safety signals,” one of the study authors told media.
“By capturing these voices, we can surface real-world harms that are invisible to traditional reporting systems.”
Which on the face of it, sounds quite helpful.
The problem that presents, however, is the veracity of the input data that is being assessed.
What’s to stop one product maker flooding social media with reports of adverse events concerning a competitor’s products, and vice versa? We were always taught that when it comes to digital analysis “garbage in equals garbage out”.
In order to address this elephant, the Waldo system still relies on old-fashioned humanoids to review posts.
“Over-reliance on automated outputs without adequate human validation poses significant risks, as false positives could create unnecessary safety alarms while false negatives could miss critical signals,” the study authors say.
“Organisations implementing Waldo should establish clear human review protocols and ensure transparency about automated screening processes,” they added.
The developers reckon that over time they can teach the AI to be smart enough weed out the bad actors or the simply deranged and delusional contributions to the data set.
And maybe they will.
But in the meantime, as good as the system is, actual intelligence is still what’s needed to make sense of it all.
You’ll suffer no adverse effects when you send story tips to Holly@medicalrepublic.com.au.
