We know what schizophrenia looks like in humans. We think we know what schizophrenia looks like in mice. Now we may know what it looks like in a computer.
Researchers at the University of Texas have modeled the disease in a natural language parser, called DISCERN, as a way to test competing theories of the neural mechanisms that cause schizophrenia.
DISCERN was built to process and recall simple narratives. Through training, the system learns what words mean and how they work in sentences in a way that mimics activation patterns in the brain. With the original settings, DISCERN is able to digest a narrative, retain it, and reproduce the story in its "own" words. The system also identifies negative and positive aspects of the story to change the likelihood that it will remember a specific detail. In this way, DISCERN models the brain both as a semantic and emotional processor.
When the researchers at UT changed the rate at which DISCERN learned, they noticed drastically different results, After learning a third-person narrative about a mobster, DISCERN retold fragments of the story in the first person, adopting a wild, criminal autobiography. Mistaking who did what to whom, an error called "agent-slotting," is common among people with schizophrenic symptoms of delusion, and among the human patients used in controls for the study.
When diagnosing and treating schizophrenics, doctors often look for disturbances in storytelling and semantics. Finding those same symptoms in DISCERN has strengthened a theory that hyper-learning causes schizophrenia.
But the real triumph of this work could be that it validates the computer as a neural model, proving some relevance to clinical research.
We know very well what schizophrenia looks like. But, so far, we don't know really know what causes it. This has been the big, unsolvable problem with neuro-psychiatric disorders, where clinical ethics stand in the way (and rightly so) of the scientific method. If people were not people, then neuroscientists could study the phenomenon with reverse engineering, gradually coming to understand it by learning how to cause it. Instead we try to create mice with schizophrenia.
A model that communicates with words but does not breathe or feel would be a breakthrough indeed.
You can find the study results here.