I gasped as the image became clear: A young ballerina with six fingers, backwards ears, and teeth like a carnivorous piranha. I was trying to use an Artificial Intelligence (AI) program to create an illustration of a young ballerina for my daughter’s book and I had felt that I was quite close to the goal. The lighting and coloration were stunning, but the human figure was distorted like a Picasso depicting the horrors of war -- not quite the feeling I was striving for in a cute children’s book.
You see, AI is limited by the scope of its training and that means it can sometimes miss the simple observations -- just like us.
A human example became clear when I was reviewing medical documentation. It was not unusual to find examples where all but one nurse documented a bilateral leg amputation -- yet that one nurse documented that they had successfully applied TED hose to prevent blood clots in the patient’s legs.
How could the patient have a risk for blood clots to their legs if they had no legs?
Even though applying TED hose would technically be more work (not to mention miraculous for an amputee), the TED hose documentation was the most common, habitual response. If someone was distracted by other details, it was actually an easy mistake to make.
When it comes to clinical observation, we’ve known about the risks of inattentional blindness for a while. Focus is an advantage in so many ways, but it can also become a risk to accuracy.
By Gayle Porter
