One of the theories regarding the cause of schizophrenia suggests that, due to an excessive release of dopamine, the brain remembers too many irrelevant things. Schizophrenics are then overwhelmed by the vast amounts of facts, thoughts and memories all crammed together in their heads, and start processing them into conclusions that aren't based in reality. It's called the hyperlearning hypothesis, and researchers at the University of Texas in Austin recently tried to see if they could simulate it – in a computer.
Uli Grasemann, a graduate student in the Department of Computer Science, used a synthetic neural network designed by his adviser, Professor Risto Miikkulainen. Called DISCERN, the network was designed to simulate the effects that different types of neurological dysfunction have on human language function.
Grasemann and Miikkulainen started by telling DISCERN a series of stories, which it remembered using a process similar to that of human memory. "With neural networks, you basically train them by showing them examples, over and over and over again," said Grasemann. "Every time you show it an example, you say, if this is the input, then this should be your output, and if this is the input, then that should be your output. You do it again and again thousands of times, and every time it adjusts a little bit more towards doing what you want. In the end, if you do it enough, the network has learned."
Next, they altered DISCERN in order to stop it from discarding as much extraneous information – in other words, they told it to forget less. The result was "fantastical, delusional stories," that mixed and matched elements of unrelated stories. In one case, the network went so far as to claim responsibility for a terrorist bombing. The system also exhibited "derailment", a condition in which the speaker delivers jumbled sentence fragments, abruptly digresses to other topics, and continuously changes from first- to third-person storytelling.
When analyzed by Ralph Hoffman, a professor of psychiatry at the Yale School of Medicine, the behavior of the network was found to be similar to that of human schizophrenics. While not proof of the hyperlearning hypothesis, Grasemann believes the experiment does show the potential for using computers in neurological research.
"Information processing in neural networks tends to be like information processing in the human brain in many ways," he said. "So the hope was that it would also break down in similar ways. And it did."
