Researchers have simulated the symptoms of schizophrenia using a language-learning computer program, in a recent study published in Biological Psychiatry. The computer started showing schizophrenia-like symptoms when it was set to learn too much and forget too little. This study lends support to the hyperlearning hypothesis, that the brains of people with schizophrenia have trouble forgetting or filtering out irrelevant information. How the Heck:
What's the News:
The researchers used a neural network---a computer program set up to mimic networks in the brain---called DISCERN, which can learn natural language and remember stories. The researchers trained the program with 28 stories, half of which were first-person narratives, and half of which were impersonal gangsters-and-lawmen crime stories.
Once DISCERN learned those stories, the researchers tested out eight different changes to the program, each meant to simulate one possible mechanism for schizophrenia, such as random noise in certain memory networks or distortions in language processing. One of these simulated the hyperlearning hypothesis by increasing the program's learning rate, telling it to learn more intensely and to forget less.
They then prompted the program to retell the stories it had heard, and compared the results to stories told by healthy controls and schizophrenic patients.
Only the hyperlearning program made DISCERN sound like human schizophrenics: jumping from one topic to another, confusing who did what, and adding in new information. The hyperlearning computer's tendency to confuse personal narratives and third-person crime stories, the researchers wrote, "could account for the bizarreness of fixed, self-referential delusions, e.g., a patient insisting that her father-in- law is Saddam Hussein or that she herself is the Virgin Mary."
What's the Context:
The neural underpinnings of schizophrenia are unknown. While some scientists hold that hyperlearning---likely caused by a flood of the neurotransmitter dopamine in certain neural circuits, which is thought to signal the brain that new information is important---is at the root of the disease, others have suggested roles for a variety of different neurotransmitters, receptors, and brain regions.
Working with artificial neural networks lets researchers poke, prod, and investigate in a way that's impossible to do with biological networks. "We have so much more control over neural networks than we could ever have over human subjects," said study author Uli Grasemann, a computer science graduate student, in a prepared statement.
Not So Fast:
Computers are, of course, not people. Seeing schizophrenia-like results from a particular change in a computer's neural network doesn't necessarily mean a similar change in a person's neural network underlies the disease. This study provides a theoretical test, but more studies must be done to see if the hypothesis holds true in human patients.
Reference: Ralph E. Hoffman, Uli Grasemann, Ralitza Gueorguieva, Donald Quinlan, Douglas Lane, & Risto Miikkulainen. "Using Computational Patients to Evaluate Illness Mechanisms in Schizophrenia." Biological Psychiatry, May 2011. DOI:10.1016/j.biopsych.2010.12.036
Image: Flickr / Justin Ruckman
