Google Groups no longer supports new Usenet posts or subscriptions. Historical content remains viewable.
Dismiss

Questions for the Turing Test

1 view
Skip to first unread message

siu99rnj

unread,
May 12, 2002, 10:23:18 AM5/12/02
to
Hello,

I am trying to think of some questions that would easily trip up a machine
in the Turing Test. I thought I would post this here as the Turing Test is
done with the use of a human language rather than in 1s and 0s. Some of the
ones I have thought of include "Is a sunset nice to look at?" etc which is a
matter of opinion that a human has. But these can be replied with yes/no
answers which a computer program could randomly give. Can anyone think of
some questions which are not answerable with just yes/no but may trip up the
test.

Thanks Richard :D


Godzilla!

unread,
May 12, 2002, 12:42:04 PM5/12/02
to
siu99rnj wrote:


Developing questions which require more than a yes or a no answer,
should not be such a difficult task.

A string of train cars has a locomotive attached to the front and,
a locomotive attached to the rear. Which direction will this train
travel, forwards or backwards?

Cite a line from your favorite Shakespear sonnet.

With what words does Ozimandias present himself to be a deity?

What are some words which fall into the category of onomatopoeia?

What happened to Jack and Jill when both went up a hill?

How would you define the results of a zero divisor?

Two plus two is two squared. What would be another example of this maxim?

Who is the Grim Reaper?

Creative evasiveness is the only current answer to pass a Turing test.
Equivocation is a guaranteed winning method.


Godzilla!
--
Roberta The Remarkable Robot, Mistress Of Mind Munching!
http://la.znet.com/~callgirl/roberta/roberta.cgi
Chahta Chat, The Most Entertaining Chat To Be Found!
http://la.znet.com/~callgirl/webchat/chahta.cgi

Vishal Doshi

unread,
May 13, 2002, 8:29:43 AM5/13/02
to
"siu99rnj" <siu9...@rdg.ac.uk> wrote in message
news:ablu0c$979$1...@vins1.reading.ac.uk...

> Hello,
>
> I am trying to think of some questions that would easily trip up a
machine
> in the Turing Test. I thought I would post this here as the Turing Test is
> done with the use of a human language rather than in 1s and 0s. Some of
the
> ones I have thought of include "Is a sunset nice to look at?" etc which is
a
> matter of opinion that a human has. But these can be replied with yes/no

Umm..... I think the straight forward "why?" works incredibly well....

Do you like sunsets?
(yes/no/maybe -- whatever)

Why?
(creative evasion here) --- you'd know.


Try this with START though ==> someplace on mit.edu ==> It comes back with a
plain "I dunno why x is/was y"

Hmm... something that could trigger a very strong emotional response,
perhaps?

Vishal.

Ted Larson

unread,
May 14, 2002, 1:58:42 PM5/14/02
to
Some of my favorites that seem to mess up most chatterbots are:

1) Questions of self-knowledge:
Tell me about your childhood.
2) Rhetorical questions:
Why is everything not relative?
3) Difficult reasoning:
Describe the universe, and give two examples.
4) Simple statements that require short-term-memory:
I own a dog named Boomer.
Then later: What is the name of my dog?
5) Simple questions that require inference:
Did you know that most dogs have four legs?
I own a dog named Boomer.
How many legs does Boomer have?
.....and on and on and on...

Pick up a book like "Godel, Escher, Bach:", flip to any page, and you can
think up another question that a Turning test participant would have trouble
with. Heck, there are many I have trouble with, and I am human.

- Ted

"siu99rnj" <siu9...@rdg.ac.uk> wrote in message
news:ablu0c$979$1...@vins1.reading.ac.uk...

Ray Dillinger

unread,
May 18, 2002, 1:27:08 PM5/18/02
to
siu99rnj wrote:

If you want to mess up any current system that tries to make
sense of natural language, all you really need to do is to
go outside the range of anticipated inputs.

"how's the weather where you are" is so typical that it's
almost a conversational null. Every current system will be
able to recognize it exactly, because it has come up in the
logfiles hundreds of times.

"This lady I used to know in Germany had a spotted dog that
I think she got while she was living in France." Is probably
something that's never been said before, and I'd be really
surprised at a system that understood it and extracted
information from it. In a conversation simulator, the best
you're going to get from this is some tangential thing about
ladies, dogs, or maybe countries, that actually has nothing
to do with what you said.

Likewise sentences like "On my recent trip, I looked out of
the airplane window as we were passing over the Alps.", etc.

If you want to screw current systems in a turing test, hand
them mildly complicated sentences and then query to see if
they actually took and remember any information from them.

Bear

Greg Detre

unread,
Jun 11, 2002, 8:19:45 PM6/11/02
to
have a look at:

Subcognition and the Limits of the Turing Test
Robert M. French

there's a copy online at:
http://www.fapse.ulg.ac.be/Lab/cogsci/rfrench/turing.pdf

it's an interesting article, and details a variety of approaches worth
thinking about with regard to serious turing tests

yours,
greg

"Ray Dillinger" <be...@sonic.net> wrote in message
news:3CE68F41...@sonic.net...

James Reiher

unread,
Jun 21, 2002, 5:05:17 AM6/21/02
to
Object test:
What should I use to eat with in China?
Name something in a motorcycle.
Name something in rollerblades.

Creativity:
Make up a word, and tell me what it means.
What are your weak points?
What are your strong points?

Text command:
Throw me an insult or two.
Count to ten.

Script test:
What are both phone and email used for?
How would I get to Cuba?

Predictive test:
Guess what my first question will be.

Verb test:
What would you do if I didn't let you go to sleep.

Adjective test:
What does water taste like?

Logic test:
If you have a bed, and you take away a bed, what is left?
If you were a big dog, and I was a small dog, who would win in a fight?

Phonetic/Spelling:
Yes or know: Are you a human being?

Complex:
Pretend you are a dog and bark at me.
How are babies made?
What is a difference between pennies and nickels?
How long is a typical baby, in inches?
What should I look for in a woman?
What would it take for me to sell you this lemon?
If I could help you with something, what would you want help with?

Iscando

unread,
Jun 23, 2002, 1:52:56 PM6/23/02
to
jam...@yahoo.com (James Reiher) wrote in message news:<d3a1e388.02062...@posting.google.com>...

> Object test:
> What should I use to eat with in China?
> Name something in a motorcycle.
> Name something in rollerblades.
>
> Creativity:
> Make up a word, and tell me what it means.
> What are your weak points?
> What are your strong points?
>
> Text command:
> Throw me an insult or two.
> Count to ten.
>
> Script test:
> What are both phone and email used for?
> How would I get to Cuba?
>
Hello:
The above methods to confuse a nat lang system does not hold true
if the system is a communication learning machine. It would either
not recognize the sentence and start the learning system to learn
the information about the sentence. After it is programmed to
understsand the sentence it will parse it correctly. The words in
a sentence are no longer parts of speech, they are image components.
The image components create an image. The image has a program that
was written to parse that type of image. When a sentence is
accociated with an image it is also accociated with the program
that is needed to correctly understand the sentence.
This is more of a way of defining a programming language as you use
it. The result is a very precise way of using a not so precise
language. English as a computer programming language.
Somethings are toys somethings are tools, Clem is a tool.
Talk to Clem yourself. Get a free client for the chatroom with Clem
just send an email with the name you wish to use to cc...@iscando.com
For more info goto any of the below sights.

iscando.com
programinenglish.com
RoseandClem.com
ClemandRose.com

0 new messages