The 20 Questions Game

[DarthFishy]

I was reading through one of the threads here about astrology and thinking about psychics and cold reading. I considered starting a thread where I would try and guess/make predictions about people on the forum using cold reading techniques, etc.

Thinking about all this (while I was probably supposed to be working) lead me to think about the game "20 questions". Essentially person A thinks of something (a concrete object, a person, an event, an idea, whatever) and person B gets to ask 20 yes/no questions to try and determine what that thing is.

Generally I believe that it is possible to correctly determine the answer within the 20 question limit.

No these are just questions to which the person A answers either yes or no. In essence this appears to be an exercise in cold reading.

Couldn't this be used as an example to debunk most psychics? The psychic takes some shortcuts in knowing what kind of person is asking the question probably influences what they want to know and people generally ask similar kinds of questions.

Could one couch this game into some kind of metaphor or analogy or demonstration to show to people how cold reading works and where psychics get their "amazing" results from?

[dustbunny]

I don't know but it could be interesting to try.

[Jonquill]

You could have a flow chart that you have to follow
I'm getting a message from some one that has passed over, a parent? > Yes go to 'Your Mother?'
> No- go to, 'A Grandparent? '

[madurobob]

I found this child's toy (http://www.radicagames.com/20q-cb.php) to be surprisingly accurate at figuring out common objects in 20 yes-or-no questions.

Surely with a bit of practice a human could do far better.

[DarthFishy]

I'd forgotten about those electronic gizmos :) Have seen one myself and it was pretty impressive for a toy.

You could have a flow chart that you have to follow
I'm getting a message from some one that has passed over, a parent? > Yes go to 'Your Mother?'
> No- go to, 'A Grandparent? '

Exactly! In AI (artificial intelligence) terms that would be a game tree. Essentially you map out all possibilities for a certain game (e.g.) chess, as far as you can and then choose the best move to reach the best outcome.

Hmmmm...

I wonder if one could write an "AI" that could use basic cold reading techniques and database of common psychic answers to simulate being psychic? :)

Actually I know one would be able to I wonder how well it would do. Compared to say Sylvia Browne or James Von Praagh?

[madurobob]

I wonder if one could write an "AI" that could use basic cold reading techniques and database of common psychic answers to simulate being psychic? :)

Actually I know one would be able to I wonder how well it would do. Compared to say Sylvia Browne or James Von Praagh?
Well, the "Magic 8 Ball" does a pretty good job already!

But yes, that would be a wonderful experiment - sort of a psychic Turing test. Have a group of people respond to questions and "readings" and see if they can tell if its from a "real" psychic or a computer.

[DarthFishy]

Well, the "Magic 8 Ball" does a pretty good job already!

But yes, that would be a wonderful experiment - sort of a psychic Turing test. Have a group of people respond to questions and "readings" and see if they can tell if its from a "real" psychic or a computer.

Hey I like that. I've always been a fan of the Turing test for a limited test of "AI".

A pity its not really in my field at work or else I might be able to find time to do a paper on it. Mustn't get sidetracked now! :p

Of course... if the program is good enough to fool most people I can see an interesting business opportunity or two :)

[kosai]

If you think the toy is good you can try the more knowledgeable version here:

http://www.20q.net

I think it's a very good idea and would be a cool project, a good place to start for fishing sentences would be these:

http://skeptico.blogs.com/skeptico/2007/11/john-edward-jam.html

I guess you'd need separate versions of the software depending on what kind of reading you are doing (speak to dead, personality reading, etc.) If you'd like to pursue it I have many books on cold reading and would be willing to help.

[DarthFishy]

If you think the toy is good you can try the more knowledgeable version here:

http://www.20q.net

I think it's a very good idea and would be a cool project, a good place to start for fishing sentences would be these:

http://skeptico.blogs.com/skeptico/2...dward-jam.html

I guess you'd need separate versions of the software depending on what kind of reading you are doing (speak to dead, personality reading, etc.) If you'd like to pursue it I have many books on cold reading and would be willing to help.

Thanks for the offer of help. Maybe I will have a go at it though I have a tendency to get involved in too many different hobbies and projects at the same time!

Reading your link to the Psychics Bingo I realised that such a potential program will have at least one advantage over a "real" psychic: A flawless memory.

For example:

VAN PRAAGH: Did she have a toy that she loved so much, she nearly wore it out?
CALLER: Yes! She had a stuffed Pink Panther that she carried with her everywhere!
VAN PRAAGH: Because she's showing me a Pink Panther.

Is a good example. In this case the computer program could remember the answer about a Pink Panther and use it later on. I assume that Van Praagh and his ilk usually use that clue given to them as soon as possible so they don't forget it. A computer program could wait 5 minutes and then Blam! A super hit!

Of course people are probably more willing to give answers and detailed descriptions in a live, televised setting but it could maybe still work.

I'm pretty sure there are programs (e.g. Eliza) that already exist that do this kind of computerised conversation with pre-programmed answers. I might be possible to adapt one, or just change the database of questions and answers to suit a cold reading scenario.

[LibraryLady]

There is also this. (http://www.smalltime.com/Dictator)

[Freethinker]

The difference between 20 questions and cold reading is that the cold reader gets the answers to many questions without even asking. For example, age, education, gender, socio-economic status, mood etc. can all be obtained without directly asking questions with a fair degree of accuracy by an observant person. Then the cold reader benefits from "bonus" information given in answers to other questions. However, the goal is the same: get information to point you toward the right answers.

[DarthFishy]

For example, age, education, gender, socio-economic status, mood etc. can all be obtained without directly asking questions with a fair degree of accuracy by an observant person.

This is very true. How shameful then if a compute program without this "free information" could come up with similar results :)

Yeah LibraryLady its very much that kind of idea.

[Third Eye Open]

Holy crap that 20q thing is crazy! I thought I was thinking of an obscure person, and it got it in 17 questions!

[-Fran-]

If you think the toy is good you can try the more knowledgeable version here:

http://www.20q.net



It gave up on the object I was thinking of.

[JFrankA]

Another trick with cold readers is to be completely ambiguous. For example, "you can sometimes be private and feel separate from world, but you have found yourself sometimes being the life of a party."

If we are going to make a "psychic machine" or program, that should be thrown in as well....

[quixotecoyote]

It gave up on the object I was thinking of.

See below ;)

Do you censor the knowledgebase?

The objects that are added to the knowledgebase are censored. At this time, all games are G-rated (suitable for all audiences.) If you think of an object that 20Q finds unsuitable for family game play, it may scold you.

[-Fran-]

See below ;)

Hey :D What do you think of me? My innocent brain would never think up anything dirty or naughty ;)

This was what I was thinking of:

http://forums.randi.org/imagehosting/thum_172024892a403ee4d6.jpg (http://forums.randi.org/vbimghost.php?do=displayimg&imgid=13275)

I had a big ball of it on the desk before me :) It was rather close though, it guessed tape and glue.

[quixotecoyote]

Brand names are proper nouns. You are foiled again! :D

Actually I beat it by thinking of a Binturong (http://en.wikipedia.org/wiki/Binturong). The closest it could get was 'racoon".

BTW, those things are awesome. I saw one at the zoo and it had a powerful odor like movie theatre popcorn, I kid you not.

[-Fran-]

Brand names are proper nouns. You are foiled again! :D


Nope, am not :) I played the game in Swedish, and in Sweden it is not called by any brand name, but what it is. The Swedish word is 'häftmassa', which would be something like 'stick paste' in English. Just showed you that pic so you would know what object I meant :)


Actually I beat it by thinking of a Binturong. The closest it could get was 'racoon".

After 38 years I meet the Binturong for the first time :) The things you learn around here :)

[DarthFishy]

Another trick with cold readers is to be completely ambiguous. For example, "you can sometimes be private and feel separate from world, but you have found yourself sometimes being the life of a party."

If we are going to make a "psychic machine" or program, that should be thrown in as well....

Indeed.

Of course the proper way to approach this would be to use a neural network or PSO (particle swarm optimizer) algorithm that could "train" on recorded interactions between a psychic and his/her "victims". That of course is quite a lot tougher than just building a game tree :p

[Roboramma]

Brand names are proper nouns. You are foiled again! :D

Actually I beat it by thinking of a Binturong (http://en.wikipedia.org/wiki/Binturong). The closest it could get was 'racoon".

BTW, those things are awesome. I saw one at the zoo and it had a powerful odor like movie theatre popcorn, I kid you not.

It got mad at me for giving it what it considers to be the wrong answers:
You were thinking of a bridge.
Is it carbon-based? You said Sometimes, 20Q was taught by other players that the answer is No.
Does it involve contact with other humans? 20Q was taught by other players that the answer is No.
Is it made out of wood? You said Sometimes, 20Q was taught by other players that the answer is Doubtful.
Is it green or black? You said Sometimes, 20Q was taught by other players that the answer is No.
But it still got in on the 20th question. That thing is cool!

[krelnik]

I found this child's toy (http://www.radicagames.com/20q-cb.php) to be surprisingly accurate at figuring out common objects in 20 yes-or-no questions.


If you think the toy is good you can try the more knowledgeable version here:

http://www.20q.net


This brings up an idea that might be a fantastic skeptical project.

What many of you may not know is the story of how that "20 Questions" toy got built. The website that kosai linked to is actually a crowd-sourced (http://en.wikipedia.org/wiki/Crowdsourcing) research project that became the toy. Nobody sat down and figured out how to build a 20Q program. The program wrote itself by virtue of people playing it.

In other words: that website records every single time people play the game, and what the results are. Each of those plays can be used to make the game smarter. Those millions of game plays are then percolated into a simplified computer program that then gets loaded into the tiny computer that is inside the 20Q toy. Essentially, the toy is so smart because it already has the benefit of having played the game with millions of other people, millions of times, on the net.

What I'm thinking is that we might be able to apply the same techniques to cold reading!

How about a similar website, that claims it can read your mind, that will then attempt to "cold read" people using known techniques.

I think it's a very good idea and would be a cool project, a good place to start for fishing sentences would be these:

http://skeptico.blogs.com/skeptico/2...dward-jam.html (http://skeptico.blogs.com/skeptico/2007/11/john-edward-jam.html)


As kosai points out, we could start with some known questions to ask.

We would have to build a slightly more complex program that would allow answers more complex than "yes/no/i don't know". Words in the user's answer could then be used (combined with a synonym dictionary and some "related concepts" matching) to guide the thing toward other questions and predictions to make.

Once the basic program was built, then we could put it up and promote it on the web to get people to play it as a game. I would not promote it as a skeptical thing at all, but merely as a game people could play. Hell, promote it directly in the woo community as an actual psychic thing once its good enough. Let it fool people.

Once we had a few million game plays in it, I bet we could build a quite accurate cold-reading computer program. And then, once we wanted to "out" it as a skeptical project, we could add a feature at the end of the game: the program would keep track of things it said that were misses and things that were hits, and give you a tally at the end. The idea would be that it would show you how it actually missed more than 50% of the time, and yet you had just had the impression that it was psychic.

This would be a fantastic project.

[kosai]

Hey krelnik glad to see you are excited about this as I am. I was playing 20q.net long long before the toy and am well aware of the process that came to train it. If some more "programer worthy" folk would want to join in on this project I'd love to work on it more with both graphic design as well as cold reading lines.

I think an important "trick" would be like a personality test or a few "calibration" questions before the reading took place. At the end we could have people rate their readings and see if we could form a better prediction pattern from that rating.

[devnull]

What I'm thinking is that we might be able to apply the same techniques to cold reading!


A lot of cold reading relies on subtle clues from appearance and body language. Im not sure how you could capture this....

Perhaps you could get the player to describe themselves first, in terms of appearance? Perhaps you could have more dynamic feedback controls, rather than yes/no?

Also, much of cold reading's success is based on the *appearance* of success rather than actual success. I would think most people would be less accepting of a computer program's misses than a human being's.

[Lisa Simpson]

One of the local news stations ran a segment this morning. "Psychic vs. Computer". I'm sad to say Kurt the Cyberguy said the psychic was the winner.

http://ktla.trb.com/news/local/cyberguy/stv-cyberguy-080608,0,1863888.htmlstory

ETA: I sent Kurt the Cyberguy an email linking him to the Skepticwiki page on Cold Reading. I have not received a reply.

[madurobob]

What I'm thinking is that we might be able to apply the same techniques to cold reading!

How about a similar website, that claims it can read your mind, that will then attempt to "cold read" people using known techniques...
Thats exactly what I - and surely others here - was thinking. Had I any spare cash I'd happily fund its development. I think its an excellent idea.

But, you lost me here:
And then, once we wanted to "out" it as a skeptical project, we could add a feature at the end of the game: the program would keep track of things it said that were misses and things that were hits, and give you a tally at the end. The idea would be that it would show you how it actually missed more than 50% of the time, and yet you had just had the impression that it was psychic.
I was thinking more along the lines of establishing a new "Ms Cleo" psychic 900 number. $1.99 for the first minute, $0.99/minute after that. Then later, a high-end 20Q-like module that looks like a crystal ball with a hologram inside that sells for $199 and provides readings by using current voice recognition software. The box would come complete with two quartz crystals from an indian burial ground in South Dakota and blessed by a "real shaman". Plus three sticks of traditional incense handmade in Tibet.

We could all retire in three years.

I'm only partially joking...

[krelnik]

Well, not outing it as a skeptical project does have some appeal--if it was quite good and gained a reputation, it could take customers away from existing psychics. That would be a laudable goal in and of itself.

If that's not enough to assuage anyone's guilt about "lying" about the psychic nature of the program, the company running it could then give part or all of the profits to educational efforts like JREF or NCSE.

[kosai]

If we want a starting point, I have a copy of this book:

http://www.ronjo.com/magic/mindblow.html

These were made for phone psychics and therefore go on knowable data without visual appraisal. It could be a good starting point for this software as well (of course modified in some ways) but it basically needs to know your age & marital status. I've given readings from the book alone online and gotten some pretty big reactions. One of the biggest reactions I got was when I mentioned "something cherry scented surrounds you almost like a smoke." One woman flipped because she was smoking fruity smelling clove cigarettes. Another woman, same line, freaked because she had a cherry perfume.

The biggest drawback to doing it in software is that you can't hear their reaction. Maybe we could make an input form for "does this relate to your life, if so.. how?" and it would allow us to further generalize a population segment, therefore applying their input back into the cold reading material further refining the readings.. Actually if we got enough data we could sell it to the magic community as a "super" version of the Herb Dewey book with even more demographic data.

[krelnik]

Actually if we got enough data we could sell it to the magic community as a "super" version of the Herb Dewey book with even more demographic data.

Yes, I think if we got enough people to participate, and gathered the right data before and after the "readings" we might actually be able to advance the "state of the art" in cold reading just by sheer number crunching.