My niece, Martha, posted throwback pictures from her 2001 wedding on Facebook, and one of them is a picture of my dad and his mate at the time. The other morning I woke up to a Facebook notification that “Martha posted a picture of me.” When I clicked through, sure enough, it was the picture of my dad from 2001. Talk about foreshadowing! But all chuckles aside, it is worth noting that Facebook’s facial recognition algorithm confused me with my dad. He was 81 at the time and I am 63 — what cruelty lies in Facebook’s computing? I’ll admit there is a resemblance, but that is a significant error. While it is (to some) amusing that Facebook’s algorithm confused my dad and me, consider if the confusion was a computer comparing pigmented lesions in a dermatologist’s practice. There, a diagnostic error is no laughing matter.
Pieces about artificial intelligence (AI) in healthcare continue to clog the Internet. Recently there was an op-ed piece in The Boston Globe, referring to an article from Nature Machine Intelligence, where researchers studied interactions with AI algorithms. Researchers asked study participants to play cooperative games with a computer or a person. They were distrustful of their partner when they knew it was a computer and also ruder in their interactions. This fascinating research is evidence that we are not yet comfortable interacting with machines, even as they get more human-like in their design.
Since becoming Editor-in-Chief at npj Digital Medicine, I have been amazed by the number of submissions related in some way to machine learning/artificial intelligence. This is more validation of the growing interest in these technologies and how they will impact care. It’s great fun to learn from looking at a published article and commentary by the authors in the journal’s community. One great example is a recent piece by Myers et al. outlining ways to identify unreliable predictions in clinical risk models. The accompanying blog post is even more thoughtful and practical.
As the controversies and tensions around AI in healthcare swirl, I am comforted by Janelle Shane’s writing and TED talks, where she amusingly demystifies AI. One conclusion is that today’s AI has the brain of a snail.
Back to Facebook’s algorithm confusion of my image with my dad. Things could be worse. But it does suggest that as today’s AI rushes headlong into the healthcare market, caution is advised until we can get a better grip on which tasks to delegate to computers and which to keep to ourselves.