Ruler

Summer time medical shadowing is one way for high school students to venture in the area of practicing doctors. This gives students a better idea what the career of a medical doctor is like. 

After a busy clinic with high school students sitting next to me today, I had a far less hard-pressed dinner time with them. We started off talking about dermatology specialty. And then the conversation moved on and they asked me about if and how artificial intelligence is going to eventually phase out the medical doctors. 

I don't know the answer but their questions remind us that the virtual brains are hitherto the size of a worm's. That means artificial intelligence works well with one narrow task at a time.  Deep convolutional neural networks, for example, can be trained by a dataset of hundreds of thousands of clinical images to achieve performance on par with board-certified dermatologists. That's a way to use artificial intelligence to tell the difference between healthy skin and skin cancer. So promising are those algorithms that even the scientific journal Nature published a proof-of-concept study on this topic.   

Wait a minute, you might be thinking. Did the pea-sized virtual brain make mistake? 

Yup. Sometimes, they did. One day the research team noticed something odd about the results: the machine was inadvertently trained to be a ruler detector instead. The rules-based algorithm was "smart" enough to find out skin cancer in the training data had been frequently photographed next to rulers for scale. 

Alas, the program turned out to call out loud "cancer" whenever there is a ruler.