Schizophrenia from a computer that acts like a human with schizophrenia

Scientists at Yale University and the University of Texas at Austin (UT) created a computer brain, then tweaked it to trigger schizophrenic behavior that included claiming responsibility for a bombing. Grad student Uli Grasemann and Professor Risto Miikkulainen and colleagues built an electronic brain called a neural network and then got it to behave like a real brain flooded with a chemical called dopamine. Excess dopamine is linked to the behavioral and speech symptoms of schizophrenia. With the simulated dopamine flood, the computer brain spat out unrelated sentences and told fantastic or nonsensical stories. At one point, it declared that it had committed a terrorist bombing. The group published their findings in the May 15, 2011, issue of Biological Psychiatry.

The researchers were testing an idea about how schizophrenia might develop called the “hyperlearning hypothesis.” Based on this thinking, people with schizophrenia lose the ability to filter out irrelevant information. Think about all of the inputs your senses are receiving right now: colors, sounds, touches, tastes, smells. Think how very few of them you register when someone hasn’t just asked you to think about them. In schizophrenia, according to the hyperlearning idea, the brain doesn’t filter all that out. All the information comes through as relevant and overwhelms the circuitry. In an effort to bring order to the chaos, the brain tries to form a story from it all, even though no linear narrative is possible. The result is the garbled or incoherent manifestations of schizophrenia.

The neural network that claimed a terrorist bombing is named DISCERN. This network of software “neurons” or simulated brain cells processes human language. Scientists have taught DISCERN stories that DISCERN then can summarize, including crime stories with an emotional element. In the computer version of reading comprehension testing, DISCERN even learned to answer questions about the stories.

Of course, a key to understanding any story is filtering out what’s not relevant and keeping in the important parts. The original version of DISCERN was great at that.

But then, researchers simulated flooding the neural network with dopamine, reflecting what they think happens in a schizophrenic brain. To do this, they lowered the rate at which DISCERN forgets. Suddenly, DISCERN became a hyperlearner. It amped up its learning rate, taking in information faster and holding onto it. It could no longer tell “important” from “not important” in the stories. As it created narratives from the crime stories in its network, DISCERN inserted itself as a central character in some of the fantastical webs (PDF) it spun. It also lost the capacity to recall specific memories clearly, instead leaping around from phrase to phrase, sometimes using “I” and sometimes referring to itself in the third person. And it thought it was a terrorist.

The researchers compared DISCERN’s behaviors under the dopamine simulation with those of a group of patients with schizophrenia. They concluded that the hyperlearning that DISCERN showed was the best fit for what the patients did in mixing up their memories of stories and their roles in those stories.

What’s the point of using an electronic brain with schizophrenia to understand a real human brain with the disorder? “We have so much more control over neural networks,” Grasemann said. “The hope is that this kind of modeling will help clinical research.” A computer brain that responds to something like dopamine the same way a human brain does may offer a model for the effects of other factors in such disorders. According to Grasemann, the work he, Miikkulainen, and their colleagues have completed with DISCERN offers support for the hyperlearning hypothesis and shows how researchers can use neural networks to mimic how our brains process information.

A cluttered mind doesn’t remember

Meditation may help the brain turn down volume on distractions