Science fiction is filled with frightening tales of futuristic artificial intelligence, but in at least one present case, it promises to do much more help than harm. Work by researchers from Harvard and Emory universities has found that machine learning -- a type of artificial intelligence that allows computers to learn through input -- is able to predict future psychosis among a group of at-risk youth with 93 percent success. Researchers have long known that how a person speaks can be helpful in determining how at-risk someone is for developing psychosis, but even trained and experienced doctors can miss some subtle signs. The machine learning technique focused on the young people's use of words relating to sound and their semantic density as predictors. Doing so allowed the research group to, in a sense, use "a microscope for warning signs of psychosis." Psychosis, which is common in schizophrenia and presents as a person not being able to tell the difference between reality and non-reality, occurs in approximately 3 percent of all people. Researchers are hoping techniques like machine learning will better allow early diagnosis and treatment.
Facts about psychosis:
- Psychosis typically begins in a person's late teens to mid-20s and affects about 100,000 young Americans every year.
- Despite common belief, people experiencing psychosis are not usually dangerous, except possibly to themselves.
- Psychosis is a symptom of an illness, not an illness itself, and can be caused by stress, trauma, and drug use.