Life,  Tech

AI Predicts Death with 90% Accuracy: Is This A Black Box We Don’t Need?

  In 2016, researchers at Stanford University trained an artificial intelligence system that could predict with relative accuracy whether a patient would die within the next three to 12 months. The team used data from 170,000 patients who died from cancer, heart and neurological diseases. The team input information including patient diagnosis results, medical procedures, and prescribed drugs to teach the artificial intelligence system. Of those expected to die within 12 months, 90% did die within this time frame, and 95% of patients predicted to live beyond 12 months did live longer. Don’t be scared by the 90% accuracy rate. Most of these patients are terminally ill, and doctors and ordinary people can make similar predictions. However, this prediction system is still quite scary, and we don’t know what to do with it.
  Doctor Siddharth Mukherjee wrote an article in the media. He said: “The death prediction system can indeed learn knowledge, but it cannot tell us why it learns knowledge. It can predict probabilities but cannot express them. The reasoning process behind the prediction. It’s like a child who learns to ride a bicycle by trial and error. When asked to describe how to ride a bicycle, he just shrugs and runs away. When we ask the algorithm ‘why’ , it can only look at us blankly. Like death, this is another black box.” Research on black boxes continues. In 2019, Google researchers said that in the field of predicting human lifespan, deep learning is used in hospitals to A large amount of patient information can achieve better results than traditional statistical models.
  Never mind the system. Let’s look at a real case. In the summer of 2001, a 58-year-old British man suddenly found that he could not pee. He was worried that he had cancer. The initial diagnosis was prostatic hypertrophy. Taking medicine could solve the peeing problem, but I have to go to the hospital for a blood test every three months. On October 16, 2002, he had a blood test. The doctor informed him a week later that he was suspected of suffering from chronic leukemia and that he needed to undergo a bone marrow test and spleen test. The doctor told him, don’t be afraid, even if you have leukemia, you can live for another ten years. This is a prediction made by a doctor. For this patient, ten years is not enough. His father lived to be in his nineties, and he also wanted to live to be in his nineties. Subsequent tests showed that he did not have leukemia. His disease was called myelofibrosis, a type of anemia in which the patient’s spleen will become enlarged. The patient’s family members breathed a sigh of relief at first. It sounded like myelofibrosis was milder than leukemia. However, after searching online, the family members discovered that this disease is also very dangerous. One website said that the patient can survive for four years, and another website said, Patients can live five to ten years. Although we do not yet have a recognized and authoritative prediction system, survival rate and survival time can always be found online.
  The diagnosis and treatment plan provided by the doctor to this patient was bone marrow transplantation. The doctor said that without bone marrow transplantation, the patient would only have two years of healthy life. Surgery also comes with various predictable and unpredictable risks, and according to the doctor’s “comprehensive” surgical plan, he will spend a long time in the hospital for the rest of his life. David’s wife suggested to him that he should not undergo any treatment. After two years of living a healthy life, let the rest go. David was very depressed after hearing this. He was not satisfied with living for two years. He frequently went in and out of the hospital for prostate examination and spleen examination. The doctor finally decided to remove the spleen first, then do a bone marrow transplant three months later, and finally consider prostate cancer. David followed this procedure, but after the bone marrow transplant, David passed away. He did not live to the ten-year lifespan predicted by doctors based on leukemia, nor did he live with bone marrow fibrosis according to the most pessimistic statement on the Internet. Four years later, I still haven’t lived long enough for the doctor to say I can live a healthy life for two years without surgery.
  To predict how long a patient will live, Stanford University’s intelligent system includes 13,654 dimensions of input. It sounds complicated, but its principle is not that complicated. It is conceivable that if there is enough data, the intelligent prediction system will probably become more and more accurate. . But does a patient need this prediction? Does society or government need such predictions? If we know that a patient is in critical condition, how many medical resources will be available to rescue him? So, Siddharth Mukherjee chose to ignore this black box.

error: Content is protected !!