So perhaps neural networks will need psychiatrists someday.
And perhaps the engineers that work on neural networks will be able to bring insights from experimentation performed in the domain of artificial intelligence, where things are more easily quantified and the space less ethically-fraught, and apply those insights to the field of human psychology.
If we continue to model the artificial mind after the human one, we can expect to witness the presentation of disorders -- though they may at first be thought of as "bugs" -- similar to the disorders that arise in human minds today: Autism, seizures, bipolar, Alzheimer's, etc. Some of these are apparently disorders of architecture (i.e. autism and other life-long disorders which are effectively untreatable.) Others are disorders of neurochemical modulation (i.e. seizures, as evidenced by their presentation upon alcohol withdrawal; bipolar disorder, etc.) Others are disorders of maintenance, of hardware degradation (i.e. Alzheimer's, dementia.)
Architecture refers to the number of neurons present and which are connected to which.
Neurochemical modulation refers to the way that neurochemicals (or their artificial counterparts, in a computer model) affect the function and behavior of neurons and the systems which they compose.
Maintenance and degradation refer to the upkeep of neural systems over time. To the extent that our inorganic components are subject to degradation and pollution (i.e. diffusion of impurities in circuit elements,) we will see maintenance disorders manifest in inorganic neural systems.
Of course, there will always be fundamental differences between organic and inorganic minds, and therefore we can expect to be presented with some entirely new problems, too. One such problem is the one of catastrophic forgetting. Biological brains retain knowledge and skills quite well in the face of new information, compared to artificial neural networks. ANNs have a problem of forgetting old skills quickly. In other words, if an ANN is trained with additional data, it is often then observed to perform poorly on a task at which it previously performed well.
Coming to understand the nature of human mental disorders may be the key to debugging our artificial intelligences, and vice versa. Our first attempts will be clumsy, and we will likely see many iterations of artificial minds which are deficient in one way or another -- one might have a good eidetic memory but be terrible at adapting to new situations; while another might very dynamic, but not very resilient in its "core beliefs."
Indeed, we will likely come to a deeper understanding of how architecture gives rise to consciousness and behavior as we create minds from schematics and observe their behavior. What architectures are capable of creating good music? Eloquent speech? Visual problem-solving? What about mathematical analysis? And motor control?
Perhaps we will find that the mind is more self-organizing than we ever suspected, and that our so-called knowledge of the structures of the human brain and their respective functions is all a fallacy, that these functions arise simply as a matter of physical proximity to the inputs and outputs of the respective sense-organs and musculatures.
But, perhaps not. Perhaps we will find that there is a strict genetic code that produces the particular circuits within the brain which make possible certain behaviors and yield certain propensities -- a talent for music, for speech, for mathematics, for spatiotemporal performance (i.e. athletics,) for competition, for theory-of-mind. Perhaps we will find that certain neuromodulations correspond to certain characteristics of personality. Does norepinephrine calibrate curiosity? Does prolactin calibrate extraversion? Does GABA calibrate conscientiousness?
