Gavroche’s recent discovery of the The T.W.I.N.K.I.E.S. Project conducted at Rice University in 1995 stirred some memories. I recall seeing this web site a few years ago, while looking for references to “Peep Jousting” and other “Peep Research“. On the one hand, I’m surprised the site is still around if it’s not been receiving any maintenance. On the other hand, it simply must at least as popular as the Peep sites, so I guess it’s not too surprising.
Giving the site a quick look, I noticed a possible flaw in the conclusions of the the Turing Test they administered. (Yes, they apparently administered a variation of the Turing Test to a Twinkie.) Given the length of the critique, I felt it would perhaps be better to comment here rather than on John’s site.
In their description of their testing methodology, the researchers note a procedural error.
When asked to assign himself and the Twinkie the designations A & B without telling us which was which, the human promptly replied “I’ll be A.” However, we decided to continue the test.
(Under normal circumstances, the researchers should have abandoned the test and attempted the experiment again with another set of subjects. Their willingness to overlook the error may be forgiven as they had already abandoned a previous attempt and the test was taking place at a relatively late hour during the final exams period. As an additional consideration, had they abandoned this test, the procedural flaw may not have been discovered and important data lost for all time.)
Note that according to the researchers, it was the human who replied to that instruction. Because both subjects were behind a sheet, it’s not clear how they determined the reply originated with the human, but we’ll have to assume they somehow knew this to be the case.
Next, examine the pattern of the responses:
Q (cg): What would you describe as the purpose of your existence?
Subject A: (no answer)
Subject B: To woo women.
Q (ts): Describe where the other subject is, relative to you.
Subject B: On a chair.
Subject A: (no answer)
Q (cg): Describe the last meal you ate.
Subject A: (no answer)
Subject B: These chicken chunks (after joking about eating subject A)
Q (ts): How do you feel about your mother?
Subject B: She gives me money, I like her.
Subject A: (no answer)
Subject B: (ostensibly the Twinkie) is the only one to respond! The same pattern occurs during the free association portion of the test, again, only the Twinkie responds.
After examining the test data, the researchers reported their conclusions:
After careful study of all responses, we determined that subject A was the Twinkie, and B was the human.
This conclusion however completely contradicts their earlier observations!
This leads me to some rather startling conclusions of my own.
If Subject A was indeed the Twinkie, then we have to face up to the reality that Twinkies are capable of speech. The ability to speak is a sign of intelligence. The fact that said speech took place in a manner which initially led the researchers to conclude they were speaking to a human means the Twinkie in fact passed the Turing test.
If Subject B was the Twinkie, then we have to face up to the reality that the Twinkie was intelligent enough to make the researchers believe they were conversing with a human. Again, this means that the Twinkie passed the Turing test.
Either way, the social, religious, moral and dietary implications are quite staggering. To be blunt, Eating Twinkies constitutes murder. Consider too that Twinkies do not occur in nature, coming instead from industrial bakeries. Apparently the Hostess company has been playing god all these years.
I would offer one closing thought: The ability to distinguish a human from a Twinkie should be at least as significant as the ability to hold a conversation when determining whether a subject is sentient.
Although it’s dangerous to draw generalities from a single data point, the conclusion of the T.W.I.N.K.I.E.S. project’s Turing test would seem to suggest that Rice University students don’t qualify as sentient life forms.
Good Catch.
There is the unfortunate possibility that the human was the one who said “I’ll be A”, and then responded as B. In which case the description of the events, and the conclusion, are accurate.
However, that raises the point that since humans are known for their ability to lie, someone should have been observing human and twinkie on the other side of the sheet to detect any deception. And to verify whether those conducting the experiment made the correct conclusion. As it is they ‘proved’ that B was Sentient and A wasn’t, and that’s really all they proved.
Finally…the R comes before the O in my nickname.
Fixed spelling. (Spell-check didn’t recognize either version.)
Well, John, when you Google “gavorche,” you may find a whole new line of Hugo research records: those who have misspelled, for some reason, one or more terms (names, etc.)
How does this apply to AI tests? Possibly, this way, remotely –>>
What if Twinkies don’t speak English?
In that case, the test procedure was flawed from the get-go.
But, even if they speak English, they may choose a variant of it, because of their super secret society’s rules — where all writings of names follow a rule of reversing a letter in any proper name, perhaps.
Hence, I warn you to beware of the Twinkies among us — you never know who is only pretending to be human, but is actually a Twinkie in human disguise. Given a Twinkie’s advanced mind-control powers, disguises are quite easily. I think Rod Sterling tried to do an episode on this one time, but it was lost in the twilight days of the show, when the Twinkie conspiracy was extending its soft touch farther into world domination.