Want to pass a 1-word Turing Test? Say "poop" - Page 2

Hellbound · 25th September 2018, 06:39 AM

Originally Posted by Dave Rogers

I'm pretty sure it actually said:

"What is a word people tend to think an AI wouldn't randomly use?"

Dave

So, in other words, "How to Fake a Turing Test"?

Dave Rogers · 25th September 2018, 06:42 AM

Originally Posted by Hellbound

So, in other words, "How to Fake a Turing Test"?

Story of my life....

Dave

Myriad · 25th September 2018, 07:16 AM

It's actually "What word do people [the survey respondents] tend to think other people [the judge of the hypothetical contest] tend to think an AI wouldn't randomly use?"

The contest itself would never work; it fails as soon as it becomes "meta," which it inherently does almost immediately. Regardless of how much thinking goes into deciding upon a word, it's trivially easy to imitate that word, or substitute similar ones, subsequently. It's only interesting if the challenge is new and unexpected, and even then, if the AI were able to predict "what would a human tend to say in this novel situation?" it would do well enough in the contest, and if it weren't it would be easy to distinguish from a human most of the time.

The interest is in the what-people-think parts, as Dave and the original investigators point out. The results show (not particularly surprisingly) that people would expect an AI to stand out as cerebral and erudite (as they've been depicted in SF for decades) and overlook the simple or childish or crude.

Of the answers I've though of so far, my favorite is "rosebud." It shows a degree of creative association, to a cultural reference in which a single word has a cryptic meaning exposing the questioned human-ness of the speaker, that just might be beyond an AI. (Though, if Watson can win Jeopardy, that approach might not work at all. And, on the other side of the coin, the human judge might not understand it.)

Squeegee Beckenheim · 25th September 2018, 07:40 AM

Originally Posted by Myriad

The contest itself would never work; it fails as soon as it becomes "meta," which it inherently does almost immediately.

It doesn't even need to wait that long. There's no reasonable way to expect to be able to tell a human from an AI through their choice of one word. As you note, all you're really testing there is what the people who programmed it think would be the word most likely to fool the examiners.

It'd be better if it was a word said in answer to a question, but even then the fact that it's a single word would allow for too much interpretation on the part of the judges. Unless the answer is completely nonsensical, then something where you have to search for meaning might be seen as a sign of abstract thinking.

For example, if the question were "what do you think about Donald Trump's presidency?" and the answer was "bananas" then that could either be a sign that the AI picked a random word that doesn't fit or it could be a sign that it's being compared to a banana republic. Or that Trump is being said to be crazy. "Dope" could mean that the respondent thinks he's an idiot, or that it's great, or that it's addictive. And so on.

Porpoise of Life · 25th September 2018, 07:51 AM

I just fill in the Captcha.

Squeegee Beckenheim · 25th September 2018, 08:39 AM

Originally Posted by Porpoise of Life

I just fill in the Captcha.

The street sign ones are always a little questionable. Do they mean all the squares which are completely filled with the signs, or should I click on a square that has 3 pixels that are from a sign? Do I have to click on the poles as well?

I've never been able to tell.

Skeptical Greg · 25th September 2018, 09:37 AM

Of course, little of this has anything to do with an actual Turing test.

( Has anyone coined the rule of " Of course..." ? )

TragicMonkey · 25th September 2018, 10:51 AM

Originally Posted by Squeegee Beckenheim

The street sign ones are always a little questionable. Do they mean all the squares which are completely filled with the signs, or should I click on a square that has 3 pixels that are from a sign? Do I have to click on the poles as well?

I've never been able to tell.

And is any sign upon a vehicle-travelled path a 'street sign'? A stop sign clearly is, but what about 'Vote For Nancy Buttocks'? Is the distinction between street, road, avenue, boulevard, court, and way relevant?

Perhaps an indicator of humanness is that given a task the human will raise objections, complaints, and questions!

Elagabalus · 25th September 2018, 11:20 AM

Originally Posted by TragicMonkey

... Is the distinction between street, road, avenue, boulevard, court, and way relevant?

In France it is!

WhatRoughBeast · 2nd October 2018, 08:13 AM

Originally Posted by gerdbonk

And nobody said "supercalifragilisticexpialidocious"?

At least one did. Also, "moist" and "asystole".

Psi Baba · 2nd October 2018, 08:23 AM

Splunge!

autumn1971 · 8th October 2018, 08:23 AM

Originally Posted by Skeptical Greg

Of course, little of this has anything to do with an actual Turing test.

( Has anyone coined the rule of " Of course..." ? )

So, of course Hitler was just a Poe.

Dave Rogers · 8th October 2018, 08:37 AM

Originally Posted by Skeptical Greg

( Has anyone coined the rule of " Of course..." ? )

No, but I think it's pretty well established that it's a pre-emptive claim that anybody disagreeing with what follows is either ignorant, stupid or both.

Sorry, let me rephrase that.

No, but of course it's pretty well established that it's a pre-emptive claim that anybody disagreeing with what follows is either ignorant, stupid or both.

Dave

Sideroxylon · 8th October 2018, 02:15 PM

Originally Posted by Squeegee Beckenheim

The street sign ones are always a little questionable. Do they mean all the squares which are completely filled with the signs, or should I click on a square that has 3 pixels that are from a sign? Do I have to click on the poles as well?

I've never been able to tell.

In the future a driverless bus is going to miss the “Bridge Out” sign and it will be all your fault.

25th September 2018, 06:42 AM	#42
Dave Rogers Bandaged ice that stampedes inexpensively through a scribbled morning waving necessary ankles Join Date: Jan 2007 Location: Cair Paravel, according to XKCD Posts: 28,485	Originally Posted by Hellbound So, in other words, "How to Fake a Turing Test"? Story of my life.... Dave
	__________________ Me: So what you're saying is that, if the load carrying ability of the lower structure is reduced to the point where it can no longer support the load above it, it will collapse without a jolt, right? Tony Szamboti: That is right

25th September 2018, 07:16 AM	#43
Myriad Hyperthetical Join Date: Nov 2006 Location: A pocket paradise between the sewage treatment plant and the railroad Posts: 14,477	It's actually "What word do people [the survey respondents] tend to think other people [the judge of the hypothetical contest] tend to think an AI wouldn't randomly use?" The contest itself would never work; it fails as soon as it becomes "meta," which it inherently does almost immediately. Regardless of how much thinking goes into deciding upon a word, it's trivially easy to imitate that word, or substitute similar ones, subsequently. It's only interesting if the challenge is new and unexpected, and even then, if the AI were able to predict "what would a human tend to say in this novel situation?" it would do well enough in the contest, and if it weren't it would be easy to distinguish from a human most of the time. The interest is in the what-people-think parts, as Dave and the original investigators point out. The results show (not particularly surprisingly) that people would expect an AI to stand out as cerebral and erudite (as they've been depicted in SF for decades) and overlook the simple or childish or crude. Of the answers I've though of so far, my favorite is "rosebud." It shows a degree of creative association, to a cultural reference in which a single word has a cryptic meaning exposing the questioned human-ness of the speaker, that just might be beyond an AI. (Though, if Watson can win Jeopardy, that approach might not work at all. And, on the other side of the coin, the human judge might not understand it.)
	__________________ A zømbie once bit my sister...

25th September 2018, 07:40 AM	#44
Squeegee Beckenheim Penultimate Amazing Join Date: Dec 2010 Posts: 23,129	Originally Posted by Myriad The contest itself would never work; it fails as soon as it becomes "meta," which it inherently does almost immediately. It doesn't even need to wait that long. There's no reasonable way to expect to be able to tell a human from an AI through their choice of one word. As you note, all you're really testing there is what the people who programmed it think would be the word most likely to fool the examiners. It'd be better if it was a word said in answer to a question, but even then the fact that it's a single word would allow for too much interpretation on the part of the judges. Unless the answer is completely nonsensical, then something where you have to search for meaning might be seen as a sign of abstract thinking. For example, if the question were "what do you think about Donald Trump's presidency?" and the answer was "bananas" then that could either be a sign that the AI picked a random word that doesn't fit or it could be a sign that it's being compared to a banana republic. Or that Trump is being said to be crazy. "Dope" could mean that the respondent thinks he's an idiot, or that it's great, or that it's addictive. And so on.
	__________________ I don't trust atoms. They make up everything.

25th September 2018, 08:39 AM	#46
Squeegee Beckenheim Penultimate Amazing Join Date: Dec 2010 Posts: 23,129	Originally Posted by Porpoise of Life I just fill in the Captcha. The street sign ones are always a little questionable. Do they mean all the squares which are completely filled with the signs, or should I click on a square that has 3 pixels that are from a sign? Do I have to click on the poles as well? I've never been able to tell.
	__________________ I don't trust atoms. They make up everything.

25th September 2018, 09:37 AM	#47
Skeptical Greg Agave Wine Connoisseur Join Date: Jul 2002 Location: Just past ' Resume Speed ' . Posts: 15,317	Of course, little of this has anything to do with an actual Turing test. ( Has anyone coined the rule of " Of course..." ? )
	__________________ " The main problem I have with the idea of heaven, is the thought of spending eternity with most of the people who claim to be going there. "

25th September 2018, 06:39 AM	#41
Hellbound Merchant of Doom Join Date: Sep 2002 Location: Not in Hell, but I can see it from here on a clear day... Posts: 13,134	Originally Posted by Dave Rogers I'm pretty sure it actually said: "What is a word people tend to think an AI wouldn't randomly use?" Dave So, in other words, "How to Fake a Turing Test"?

25th September 2018, 07:51 AM	#45
Porpoise of Life Illuminator Join Date: Oct 2014 Posts: 4,740	I just fill in the Captcha.

25th September 2018, 10:51 AM	#48
TragicMonkey Poisoned Waffles PITH Winner Join Date: Jun 2004 Location: Monkey Posts: 47,376	Originally Posted by Squeegee Beckenheim The street sign ones are always a little questionable. Do they mean all the squares which are completely filled with the signs, or should I click on a square that has 3 pixels that are from a sign? Do I have to click on the poles as well? I've never been able to tell. And is any sign upon a vehicle-travelled path a 'street sign'? A stop sign clearly is, but what about 'Vote For Nancy Buttocks'? Is the distinction between street, road, avenue, boulevard, court, and way relevant? Perhaps an indicator of humanness is that given a task the human will raise objections, complaints, and questions!
	__________________ You added nothing to that conversation, Barbara.

25th September 2018, 11:20 AM	#49
Elagabalus Illuminator Join Date: Dec 2013 Posts: 3,846	Originally Posted by TragicMonkey ... Is the distinction between street, road, avenue, boulevard, court, and way relevant? In France it is!
Elagabalus Illuminator Join Date: Dec 2013 Posts: 3,846

2nd October 2018, 08:13 AM	#50
WhatRoughBeast Graduate Poster Join Date: Apr 2011 Posts: 1,357	Originally Posted by gerdbonk And nobody said "supercalifragilisticexpialidocious"? At least one did. Also, "moist" and "asystole".

2nd October 2018, 08:23 AM	#51
Psi Baba Homo Skepticalis Join Date: Aug 2001 Posts: 4,006	Splunge!
	__________________ We have reached a point in time where reality and satire have intersected and now you can't tell the difference. -- Lewis Black There is a cult of ignorance . . . nurtured by the false notion that democracy means that my ignorance is just as good as your knowledge. -- Isaac Asimov

8th October 2018, 08:23 AM	#52
autumn1971 Master Poster Join Date: Jan 2007 Posts: 2,621	Originally Posted by Skeptical Greg Of course, little of this has anything to do with an actual Turing test. ( Has anyone coined the rule of " Of course..." ? ) So, of course Hitler was just a Poe.
	__________________ 'A knave; a rascal; an eater of broken meats; a base, proud, shallow, beggardly, three-suited, hundred-pound, filthy, worsted-stocking knave; a lily-livered, action-taking knave, a whoreson, glass-gazing, superservicable, finical rogue;... the son and heir of a mongral bitch: one whom I will beat into clamorous whining, if thou deniest the least syllable of thy addition."' -The Bard

8th October 2018, 08:37 AM	#53
Dave Rogers Bandaged ice that stampedes inexpensively through a scribbled morning waving necessary ankles Join Date: Jan 2007 Location: Cair Paravel, according to XKCD Posts: 28,485	Originally Posted by Skeptical Greg ( Has anyone coined the rule of " Of course..." ? ) No, but I think it's pretty well established that it's a pre-emptive claim that anybody disagreeing with what follows is either ignorant, stupid or both. Sorry, let me rephrase that. No, but of course it's pretty well established that it's a pre-emptive claim that anybody disagreeing with what follows is either ignorant, stupid or both. Dave
	__________________ Me: So what you're saying is that, if the load carrying ability of the lower structure is reduced to the point where it can no longer support the load above it, it will collapse without a jolt, right? Tony Szamboti: That is right

8th October 2018, 02:15 PM	#54
Sideroxylon Featherless biped Join Date: Aug 2008 Location: Aporia Posts: 20,031	Originally Posted by Squeegee Beckenheim The street sign ones are always a little questionable. Do they mean all the squares which are completely filled with the signs, or should I click on a square that has 3 pixels that are from a sign? Do I have to click on the poles as well? I've never been able to tell. In the future a driverless bus is going to miss the “Bridge Out” sign and it will be all your fault.
	__________________ 'The first principle is that you must not fool yourself - and you are the easiest person to fool.' - Richard Feynman