The artificial intelligence that shot down human pilots in simulation after simulation at Wright-Patterson Air Force Base can also help treat bipolar disorder. The model originally used in air-to-air combat predicted who will respond to lithium 100 percent of the time, using “genetic fuzzy trees,” which mimic the natural evolution of genetics.
The LITHium Intelligent Agent (LITHIA) model, run by University of Cincinnati researchers, is the first to manage such accuracy, as even the best of eight common models used in treating bipolar disorder predict who will respond to lithium treatment with only 75 percent accuracy.
Lithium is a salt which treats both mania and depression in a lucky thirty percent of people with bipolar disorder. But prior to the discovery of a new method to predict who will respond to lithium, people were playing roulette. Scientists at the Salk Institute are able to predict with 92 percent accuracy who will respond to lithium with a blood test. But these results were blown out of the water by the UC model.
Fuzzy logic, the system the artificial intelligence is based on, relies on generalizations to make correct choices. The logic is labeled “genetic fuzzy” because the AI continually refines its answer, choosing the best inputs in a process similar to Darwinian natural selection. And unlike other types of AI, systems based on fuzzy logic can explain why it made its choices.
Teaching the AI is similar to getting a child to recognize an apple. Green apples and red apples may look different, but there’s a correct way to identify them each as the specific fruit after being given a few examples. In this way, the UC model was able to compensate for “noise” in the data, and narrow down its options to predict the correct route.
The UC findings were published in June 2017 in the Bipolar Disorders scientific journal. The researchers are also using the model to study concussions, a topic which has often confused scientists.