Computer Faces Human Psychological Test
A reader writes: "
The test known as the MMPI-2 (Minnesota Multiphasic Personality Inventory) is going to be administered to a computer-based personality. The test subject is GAC - Generic Artifical Consciousness." This is the first time I've heard of GAC - but this test is quite different from the Turing Test.
From Mindpixel's website:
The Mindpixel Corpus is the world's first and largest database of validated consensus human knowledge. It is an ever improving approximation of the mind of an average Internet user, constructed and owned tens of thousands of people just like you from all over the world, speaking all of the world's languages.
Emphesis added by me.
So, this thing is going to be obsessed with eBay and pr0n.
--
"We don't know if it is possible to build a normal personality out of millions of little pieces. This experiment will tell us how reasonable the idea is," Epstein added.
I predict that the result will be about as useful as Congress.
Aah, change is good. -- Rafiki
Yeah, but it ain't easy. -- Simba
You've gotta love this stealth advertizing for Spielberg's AI!
Refrag
I have a website. It's about Macs.
Didn't we try this 30 years ago? Did it work then?
--
--Got Lists? | Top 95 Star Wars Line
That said, it's fucked up. One of my roommates had to take it once (aside: we moved him out the next semester:), and he let us read it. There are 567 "yes/no" questions, and the first one is "I enjoy reading sports car magazines." Others include such obviously gender biased items as "I sometimes wear women's clothing." Basically, my point is, it's a very US-centric test designed to "catch" people who think differently than our definition of "normal." I suspect that the AI device wouldn't even have a coherent answer to many of the questions, since it's never lived in American culture, and doesn't have a childhood to be analysed as "traumatic," etc etc.
I can't see anything useful coming from this except perhaps a false reassurance that "this system won't go nuts like HAL-9000."
(yes, this is posted under another message below, but it fits better here.)
---
I agree that giving this test to an AI system is totally marketroid hype. I'd like to clarify your view of the MMPI, having taken it twice.
Yes, some of the questions do seem a bit bizarre and it is often difficult to concieve either the purpose of the question or how to answer, but there is tremendous redundancy built into the test and it is scored on a curve. That is, answering yes to question 241 does not tag you a psychopath; ultimately the results are correlated against the sampling of individuals who have taken the test, as is the Rorshach. The tested individual is found to have tested within some standard deviation along nine (IIRC) scales. Individuals with particular patterns of results can be associated with others with similar patterns. Occasional questions are "hot buttons" designed to illicit responses which indicate that the subject has poor reality checking or exhibits dangerous behavior. Of course, the test is only useful in conjuction with clinical, therapeutic observation.
Finally judgements about normativity, while they strike some as distasteful, are really the only way to generalize about psychological phenomena. The range of normative behavior, while flexible, is the only standard available. The idea of universal sanity or a generalized idea of madness are simply absurd and possibly dangerous.
illegitimii non ingravare
Is fixed here.