|AI so appealing because it can scan through reams of data and pick out valuable variables / Photo Credit: Tamakhin Mykhailo (via Shutterstock)|
This month, Google made headlines for a study that claimed their AI system outperformed health professionals at finding breast cancers on mammograms, wrote Christie Aschwanden of Wired, an American magazine that focuses on how emerging technologies affect culture, the economy, and politics. Does that mean there will be fewer false positives? While this might be a revolutionary breakthrough in the AI field, what if doctors asked the wrong questions? Or what if they let the AI pursue faulty premises? The technology “will be a bust.” With Google’s research, the researchers are trying to replicate and outperform human performance “on what is at its core a deeply flawed medical intervention.”
However, AI systems like the one from Google promise to combine humans and machines to better facilitate cancer diagnosis. Sadly, they also have the potential to escalate pre-existing problems such as overtesting, overdiagnosis, and overtreatment. We don’t know if the improvements in false-positive and false-negative rates will be applicable in the real world. What makes AI so appealing is its ability to scan through reams of data and pick out variables that humans never saw as valuable. In principle, this can help us diagnose any early-stage disease, akin to how a subtle movement of a seismograph can give us early signs of an earthquake.
In Aschwanden’s perspective, sometimes those variables are not that important. For example, your data set might be drawing information from a cancer screening clinic that is open for lung cancer tests every Friday. Hence, it’s possible that the AI might think that those scans taken on Fridays are more likely to be lung cancer. Some uses of algorithms and machine learning might introduce puzzling problems for clinicians. Let’s take Apple watch’s feature to detect atrial fibrillation, a type of heart arrhythmia considered as a risk factor for stroke. What if atrial fibrillation was detected by a person’s smartwatch?
Traditionally, this is diagnosed when a patient complains about their systems to a doctor. But with smartwatches like Apple Watch, it monitors healthy people without symptoms and finds cases that may otherwise not be present in clinics. Before we can adopt AI in medicine, we should consider answering these two questions: “What problem is the technology trying to address, and how will it improve patient outcomes?”