Sophia looks and behaves friendly. Language remains her drawback, when she promises to destroy humanity: she mistakes a joke for a command.
Science-fiction has not had much sentiment for an idea to have human minds judged by artificial circuits; but science fiction has been to sell, and such evaluation would be excess criticism. This does not sell.
Favorable or faultfinding the picture on human intellect, literary projections on human future would place robotics in contexts for potential totalitarianism.
RT News is a non-profit by Russian TV Novosti.
RT News: Computer predicts psychosis 100%
Predilection for “mental technology” has found a way also to American columns. These are pages from the USA PubMed Gov.
PubMed.Gov: Computer-aided diagnosis of schizophrenia
The power of ELM to provide quick but near optimal solutions to the training of Single Layer Feedforward Networks (SLFN) allows extensive exploration of discriminative power of feature spaces in affordable time with off-the-shelf computational resources.
In simple words, the software can use a single-layer artificial network, to diagnose a human being with schizophrenia, in a nick of a time, and for a nickel.
All you need is a signature.
As Wikipedia informs on the gov extension, the domain name gov is derived from the word government.
It was folk lore, some time ago, you scared children into good behavior. There had to be reasons to abandon the habit, as modern upbringing is observably more permissive, which I mean in a favorable sense.
Therefore, I will not elaborate for grown-ups, on potential horrors of technological assessment on human wits. I have tried a spellchecker on Thomas Paine’s Common Sense: to take out all that artificial intelligence condemns, you’d have to become another Elm Street slasher.
Wikipedia, A Nightmare on Elm Street
One good way to deal with a bogey is to have a look at it. Single layer feedforward means there is one network to boss the system, in the machine.
The machine is much unlike the human brain, a phylogenetic advantage. To see this, let us try Google translation on Emily Dickinson’s poetry, for Polish.
Of all the sounds despatched abroad,
There’s not a charge to me…
Click to enlarge.
We can try back-translation. With natural languages, the human translator receives the original and translated text, to verify the the translation via an analysis of word senses and meanings the translation would bring in the language of the original.
Click to enlarge.
Although the context clearly is not crime, the computer interprets the word “charge” (a sound) for an “accusation” (zarzut in Polish), which it re-translates into “objection” in English. Links here connect with the Translatica network.
The Google machine looks single-layer, in its treatment of the Polish zarzut. A human translator would be unlikely to have an objection for synonymous with an accusation.
Let us compare one more piece, Lost.
I lost a world the other day.
Has anybody found?
You’ll know it by the row of stars
Around its forehead bound.
Click to enlarge.
Here is the back-translation.
Click to enlarge.
Well, the pixie would curse the world entire, having made the world all-male, with “his” forehead.
Evidently, computers are not good enough for translation. How could they be good enough for language diagnostics?
The matter is not in the people who fix the options to need a lot of language expertise. The options being fixed is an error in itself.
Even if you made another “Alexandrian library” to avoid events as a spellchecker on Thomas Paine, you could not expect of people to “talk as by the book”. Most importantly: is language a good diagnostic for mental disturbance?
Mental disorders do not result from language variance; they are not primarily speech disorders; they do not follow language defects.
As Irena Styczek noted within her holistic approach to language and therapy, schizophrenia may develop into schizophasia.
Schizophasia is a kind of phonologically driven discourse. The person would be “following speech sounds”, and not word sense.
The example of “word salad” here would show a recourse to the hard palate (sadness-outage):
sadness cups coffee printer power outage.
Schizophasia, if it occurs, comes in late stages of the illness. It does not cause the disorder; it results from it. Diagnosis on the illness would most often come much earlier than the language deficit.
Examples of “word salad”
As Vander and others published in 1985, neuro-imaging showed differentiative biological structuring of altered function, in schizophrenia patient brains. The studies implied a developmental background, as a viral infection, malnutrition, or other, during fetal life or early after birth.
The findings were not to encourage preemptive diagnoses. The computer was to provide neuro-imaging. It was not to diagnose or predict illness.
A requirement for fMRI diagnostics might limit probability, for political abuse in mental care. Well, to mind the gov extension.
Political abuse of psychiatry, Wikipedia
Recurring to Sophia, I do not understand why make humanoid robots. For assistance with groceries, you would look desperate, to use a thing with an inevitably fake smile. To have a machine therapist, you would look desolate, utter dejection. And well, regarding other, implicated purposes — why try to have it for artificial, yet intellect?