You can “teach” a neural net a series of simple stories, but if the neural net is set to “hyperlearn” from examples, you get a schizophrenic computer:
For ordinary brains, while there’s significant evidence that people do pretty much remember everything, your brain stores them differently. In particular, intense experiences, which are signaled to the brain by the presence of dopamine, are remembered differently than others. Which is why, for example, you probably can’t remember what you had for lunch last Tuesday, but you still have strong memories of your first kiss.
The hyperlearning hypothesis posits that for schizophrenics, this system of classifying experiences breaks down because of excessive levels of dopamine. Rather than classifying some memories as important and others as less essential, the brain classes everything as important. According to the hypothesis, this is what leads to schizophrenics getting trapped into seeing patterns that aren’t there, or simply drown in so many memories that they can’t focus on anything.
In order to simulate the hyperlearning hypothesis, the team put the DISCERN network back through the paces of learning, only this time, they increased its learning rate — in other words, it wasn’t forgetting as many things. They “taught” it several stories, then asked them to repeat them back. They then compared the computer’s result to the results of schizophrenic patients, as well as healthy controls.
What they discovered is that, like the schizophrenics, the DISCERN program had trouble remembering which story it was talking about, and got elements of the different stories confused with each other. The DISCERN program also showed other symptoms of schizophrenia, such as switching back and forth between third and first person, abruptly changing sentences, and just providing jumbled responses.