The Turing Test was introduced by the famous English mathematician, cryptanalyst, and computer scientist, Alan Turing in his 1950 paper entitled “Computing Machinery and Intelligence”. During his stint at the University of Manchester, Turing wanted to know “Can machines think?” 

However, as “thinking” is too ambiguous to be verified, Turing replaced it with a three-person “imitation game” and changed his question to “Are there imaginable digital computers which would do well in the imitation game?” Over here, a human confederate, will determine, via a series of conversation between two parties, which is a computer and which is a human. 

An alternative Turing Test 

Last week, Dr. Michael J. Joyner, an anesthesiologist and physiologist of Mayo Clinic wrote an op-ed for STAT. He suggested perhaps a new form of “Turing Test” is required to determine whether artificial intelligence (AI) is capable of undertaking complex medical responsibilities. 

In his article, Dr. Joyner acknowledged the power of AI, machine learning, and data. He believed technologies could improve workflow and enhance physician-patient interactions as the former is now free of certain duties. Nevertheless, he also warned medicine is a field “lack fixed rules and stereotypical features” – two important attributes for machines to perform their jobs. As such, AI may face challenges that human counterparts do not and affect their overall potential. 

The new Turing Test proposed by Dr. Joyner involved creating a weight loss plan for patients who are severely obese (i.e., with a body-mass index of 40 and above). In order for a machine to out win a human, this new weight loss plan has to be as convincing as bariatric surgery (i.e., procedures to decrease the amount of food intake by reducing the size of one’s stomach) or in another word, a non-inferiority trial (i.e., the new treatment is not less effective than the ones already in use). 

There may not be a winner 

Obesity is chosen because, as noted by Dr. Joyner, the condition has a measurable outcome and it is generally treatable. Furthermore, there are sufficient data coming from one’s genetic profile, diet, lifestyle, exercise patterns and so on, to be fed into an AI and create algorithms that “would clearly tip the scales and show the skeptics what medical AI can do”. 

The Turing Test has been around for near 70 years yet Alan Turing’s question: “Are there imaginable digital computers which would do well in the imitation game?” remained unanswered. It’s hard to foresee whether there will be an answer to a Turing Test for medical AI. 

Besides, as Dr. Joyner mentioned earlier, medicine is diverse and complex and there is no one size fit all. Hence, even if an AI is able to come up with a weight loss plan just as effective or better than those created by human, it does not mean it will do well in other areas. Chances are there need to be different Turing Tests for different medical AI to judge which will “go big or go home”.  

Author Bio
synthetic gene empathy chinese artificial intelligence data medicine healthcare ai

Hazel Tang

A science writer with data background and an interest in the current affair, culture, and arts; a no-med from an (almost) all-med family. Follow on Twitter.