Neural Network Taught to Exhibit Schizophrenic Behavior
In 2011, computer scientists at the University of Texas at Austin conducted a remarkable experiment: they deliberately induced symptoms resembling schizophrenia in a neural network, causing the system to generate delusional statements including claiming responsibility for a terrorist bombing. The research provided new evidence supporting a leading hypothesis about the neurological mechanisms behind schizophrenic disorders.
How DISCERN Learned Language
The research team built a neural network called DISCERN, designed to learn and process natural language. The system was trained on a series of simple stories, learning to store information as relationships between words and sentences — mimicking the way a human brain encodes narrative information.
Under normal operating parameters, DISCERN could recall and distinguish between different stories it had been taught, maintaining clear boundaries between separate narratives and factual contexts.
The Hyperlearning Experiment
The researchers then modified DISCERN’s learning parameters in a specific way: they increased the system’s rate of information absorption while simultaneously reducing its ability to filter out noise in the data. In practical terms, they told the computer to “forget less.”
This manipulation was designed to simulate a phenomenon known as the hyperlearning hypothesis of schizophrenia. Some mental health researchers believe that schizophrenic individuals are unable to properly forget or ignore irrelevant stimuli, which overwhelms their ability to process and synthesize information correctly. In biological terms, the neurotransmitter dopamine plays a central role in this process of understanding and differentiating between experiences.
Flooding DISCERN with excessive data retention was analogous to flooding the human brain with dopamine, confounding the system’s ability to discern relationships between words, sentences, and events.
The Computer’s Delusional Output
The results were striking. After the hyperlearning modification, DISCERN began inserting itself into stories it had been taught, generating fantastical and delusional narratives that blended elements from unrelated stories. The network lost the ability to distinguish between separate narratives, merging them into confused accounts where it placed itself at the center of events.
In one notable instance, the system claimed responsibility for a terrorist bombing — a statement that had no basis in any of the training stories it had been given.
Implications for Understanding Schizophrenia
The researchers, who published their findings in the journal Biological Psychiatry, noted that while the experiment did not definitively prove the hyperlearning hypothesis, it provided significant supporting evidence.
Graduate student Uli Grasemann, who participated in the research, highlighted the practical advantages of using neural networks as models for studying mental illness. “We have so much more control over neural networks than we could ever have over human subjects,” he explained. “The hope is that this kind of modeling will help clinical research.”
The experiment demonstrated that artificial neural networks could serve as useful analogues for studying information-processing disorders in the human brain, potentially opening new avenues for understanding and eventually treating conditions like schizophrenia without the ethical and practical limitations of human experimentation.
