• Quick note - the problem with Youtube videos not embedding on the forum appears to have been fixed, thanks to ZiprHead. If you do still see problems let me know.

An interesting little excercise.

NFG

New Blood
Joined
Jul 8, 2007
Messages
17
Hello,

I was browsing around reading things on the internet today and, as often happens when surfing, I found myself looking at something completely unexpected. In this case it was a web site that provided psychics with a channel to sell their wares through per-minute charges, presumably on the telephone or in a chat room.

I started thinking about an old experiment, a bit of software called ELIZA that used a lot of stock phrases to do a kind of parody of a psychiatrist. It asked questions, you answered. It would re-use the words you used and feed them back to you in a way that smelled conversational. The idea was that humans would perceive intelligence from the program output despite the absence of such intelligence.

Perhaps you see where I'm going with this.

It wouldn't be hard to write a form of ELIZA in Javascript or PHP or whatever that basically collected information about people from a chat dialog and form generalized apparently psychic predictions of their future. The process would be a bit like an interview that is used to fill out a form letter.

The software could be a little tricky but it wouldn't really be that hard to do, particularly if a first-run version were based on ELIZA. You could also modify a config file for an AIML based chatbot. There are a few that one can download for free.

So you put this thing on the web and you see who buys it. Maybe you encourage people with a few "readings" and then ask them to click a button to pay for more - the old 'fulfillment'.

Click the button and the hoax is revealed. "Boo! It's software. You're an idiot. Ha ha."

So here are my questions:

1-
To be convincing a digital psychic would need a good script. What would a canned digital psychic say?

2-
I'm quite sure that someone else has done this, but just in case they haven't ... would creating something like this effectively make the situation worse by providing tools for people to engage in real fraud?

3-
Given people's aptitude for percieving deeper meaning where there is none. What are the chances that someone will come along to insist that the machine really is psychic, or a channel for otherworldly beings, and has a better than 50% accuracy rate on predictions?


Here's a good version of ELIZA. The script is well devised so that the oddities of the conversation are at least explainable. It's written in Java.

chayden.net/eliza/Eliza.html

NFG
 
1-To be convincing a digital psychic would need a good script. What would a canned digital psychic say?

"I am not a psychic" (Stock phrase).
"I am a conglomerate of emotional and psychological elements with an entrepreneurial spirit!" (Quoted from "Wapsi Square" by Paul Taylor).
"Hmm ... yes ... I see ..." (Freudian silence filler).

2-I'm quite sure that someone else has done this, but just in case they haven't ... would creating something like this effectively make the situation worse by providing tools for people to engage in real fraud?

Definately! Then again, it would also provide competition for the 'real' psychics, especially if your 'readings' were just as good as theirs!

3-Given people's aptitude for percieving deeper meaning where there is none. What are the chances that someone will come along to insist that the machine really is psychic, or a channel for otherworldly beings, and has a better than 50% accuracy rate on predictions?

Pay me NOT to, and I won't! I used to do readings, and even though I claimed to have no special 'powers' there were those who instisted that I did, and that I was just being modest.

Yeah ... me, modest ... LOL! :rolleyes:
 
It would be a but tricky to be believable. ELIZA and other response-bots still aren't really believable. Convincing automated language of any type is very hard to do.

I don't think an effective "psychic reader" program would help (or hinder or affect) the fraudsters much. And, yes, some people would believe that the computer is psychic.

In the most simple form, this would simply be a horoscope or the I Ching. To make it a bit fancier, you could make it like a cheap psychic reading where you just hear what you want. So just kick back fortune-cookie-isms, or and sentence from a horoscope, or just any general advice, or general direction, or just anything nice at all. "Plan for many pleasures ahead." "Your hard work will payoff soon." "Good luck is the result of good planning." "It is time to make some changes in your life." "Everything will turn out for the best, even if it doesn't seem to be going that way at first."

It would be best if you had pointed questions that would result in the types of questions where these types of answers would fit.

With some more work, you could probably do a Sylvia Brown. Especially if you just do medical questions. Create a list of diseases, ailments, etc. Create a list of the dozen or so diseases that Sylvia likes to talk about and a bunch of rants about each disease. When someone mentions a disease/ailment, tell them they don't have it and select a "Sylvia" disease. When they reply, match keywords to a rant and send back the rant. For anything else, just reply "Yes, Sweetie", or "No, absolutely not.". It doesn't matter. Sylvia does the same thing and even she knows it doesn't matter as long as it is delivered with confidence.

A John Edward would seem more applicable to automation, but would actually be a bit trickier. It is sort of like the Grateful Dead--you have to see it live. A program going: Is there an F? No. Is there a P? No. A John? No. A Jack? No. A Joe? No. A Joseph? No. A Joan? No. An older woman? No. Someone close to you? Yes. A woman? Yes. A sister? No. And so on. This obviously doesn't work. The user has to enter "no" so many times it is obvious it is just guessing. But John can toss out all these fish hooks in less than a second: "I see an F or P. A John/Jack or Joe/Joseph or Joan. An older woman. Some one close. A woman. A sister." You can say all of that in less than one second. Of course if you have time to sit there and read it, it looks like fishing. So you would have to create sentences that throw out lots of possibilities without looking like fishing. But it could be done.

I thought about doing something like this a few years ago. I analyzed transcripts from psychics on medical predictions. As you might expect, their guesses are fairly consistent with a most common diseases. And, of course, you can start with a "chest" problem (where most of the vital organs are) and then move to the "head" if you are wrong and relate it back to the chest in some way (use charkas if you need to).

Statistics-wise, the psychics did rather well. I expect there was some editing of what did and did not make the shows. There were also obvious cases where you have something like a 60 year old woman who weighs 800 pounds being told she might have heart problems.

So, you could build in these gimmicks. Sylvia’s gimmick is t revert to the few diseases she knows and go with them. John Edward has some nicer tricks of starting general and moving to more specific. He also has a gimmick to sometimes guess very weird things. It paid off—at least in the transcripts. He had one (I think it was him) where he guessed that the deceased person was killed by being kicked by a horse! He missed. He tried to tap dance back. But that is a great idea.

Every now and then, throw in that the deceased person was hit by lightning, or bit by a poisonous snake, or ate bad spinach. Over thousands and thousand of guesses, you might just get one right. Than everyone can say, “How could that computer program really actually know that? That is an amazing psychic ability!!!!!”
 
Is there a such thing as a Turing Test applied by someone who wants it to be a person?
 
Eliza by phone would be rough. The versions I know of have to have all communication typed in. Like this primitive version

Granted.
Voice would be rough for a compelling convincer.

I was thinking mostly text chat style as in the eliza program and others like it.
All of these eliza type programs are very thick headed.

Mostly I suppose the trick is to make up for that in the script.
 
Hello,
1-
To be convincing a digital psychic would need a good script. What would a canned digital psychic say?

NFG


Some of Sylvia Browne's favorite responses...

"Yeah."

"Yeah."

"You see what I'm saying?"

"Yeah."

ETA: You don't even need a script. Just have a random number generator select a response from a list of about 6-8 generic ones.

BTW, welcome to the board.

Steve S.
 
Last edited:
It would be a but tricky to be believable. ELIZA and other response-bots still aren't really believable. Convincing automated language of any type is very hard to do.

That's a long post. I'll need to read and process, but in general I agreed with you when I posted at first.

give me a minute.
 
Granted.
Voice would be rough for a compelling convincer.

I was thinking mostly text chat style as in the eliza program and others like it.
All of these eliza type programs are very thick headed...
So? The people who I have seen convinced by John Edward are not on the top of the bell curve. This is from personal experience, which we all know is as good as an antelope, er anecdote.
 
It could be more convincing than some of the posts I've read here.
 
Do you understand? / Does that make sense?

I'm getting a 'G' or a 'J' sound - who's that? ... Yes, cause s/he's saying "Tell him/her its [reply]"

Above only with 'an older gentleman' etc

He's near trees, rocks or water.
 
It would be a but tricky to be believable. ELIZA and other response-bots still aren't really believable. Convincing automated language of any type is very hard to do.

But not if your audience is extremely motivated to believe. I remember playing with ELIZA a long time ago, and if you didn't try to trip it up, but sort of played along, it could be fairly convincing.

I think the idea of a cold-reading ELIZA is pretty cool. And then set up a test with believers where they don't know that one of the "psychics" is a machine. Let them rate the accuracy of the readings, and take note how few of them catch that it's not a person.

On the other hand, cold-reading through a typed interface would be a bit of a challenge. Sylvia's "yeah" suggesting that she already knew what someone just told her doesn't have the same effect in print.
 
To be convincing a digital psychic would need a good script. What would a canned digital psychic say?

Start the reading with a modified Forer Effect profile of the person. Instead of the typical "You like people but are sometimes shy" type of thing, tweak it to, "I'm getting that you like people but are sometimes shy". Then, maybe have it ask the person if that sounds like them, to have them fill in some info.

Then, ask the person to ask a question. It wouldn't take much alteration to change the counselor-Eliza into pyschic-Eliza. Instead of simply rephrasing questions back, just dress it up with psychic sounding stuff.

Subject: "Will I get the promotion at work?"

psychic-ELIZA: "I know because my spirit guide is telling me that the promotion at work is important to you. Are you feeling any anxiety about the promotion at work?"

IF you limited the reading to missing person cases, it'd be a breeze. Just have the client feed in the known info, and spit back with psychic certainty the statistically most likely turn of events with a few unverifiable details added in for color. (If person missing more than 6 months, tell them the person is dead, etc.)
 
Sylvia’s gimmick is t revert to the few diseases she knows and go with them. John Edward has some nicer tricks of starting general and moving to more specific. He also has a gimmick to sometimes guess very weird things. It paid off—at least in the transcripts. He had one (I think it was him) where he guessed that the deceased person was killed by being kicked by a horse! He missed. He tried to tap dance back. But that is a great idea.

See now this is where I think this can really work.
Somebody like Sylvia Brown appears to be genuinely cruel, a distanced personality is perfect. Most people try to script their bots to either have a set character or to learn from what other chatters say. They often tout the quality of their bot based upon its ability to collect phrases. The results are often nonsense, and all of these bots are hard to talk to - assuming that the user is in charge.

Lets assume a few things.

First: This is a genuinely cruel thing to do to another person.

Second: People who want to believe will go to some effort. How many of these people are there?

Third: Without suspension of disbelief it is not possible to do this.


I contend that there is plenty of said suspension and a desire for it.

Is it good to enforce this idea on the suspended?
 
At first I was thinking of this more as a controlled experiment or at least a one-time demonstration, but I think you're talking about an on-going website.

I'm sure chat bots are much more sophisticated than ELIZA was, but I still suspect that the illusion will only last as long as the user is, as you said, suspending disbelief (or, suspending their innate skepticism) in order to play along. Once the illusion is broken, these things fail pretty miserably. So my thought was, if you do this as a website (low-cost psychic readings--actually bill their credit card and refund it afterward--without the money, there's less motivation for them to deceive themselves), you'll have some period of time where the believers will be duped. Before long though, word will get around that it's a program. Then, psychics and their supporters will say, "See, it's such a clumsy imitation, no one would fall for it. Therefore psychics are real."

HOWEVER, from that time on, anytime one of these folks goes for an on-line psychic reading, they'll be a bit more suspicious--wary of falling for another chat bot.

You will have un-suspended their skepticism.

I think it's a worthwhile project.
 
To all negative responses, say, "I see why you think that way. But you need to understand that the universe is domineering in our lives, and it gives the truth all the time. You must believe it."

or

"It's ok that you aren't in tune with the wisdom of the universe. Just suspend all of your skepticism and stay with me."
 
Sorry I left a part out. Then keep chunking the poor guy with cold reading, fortune cookie variations, and things like: "I get the image of a fish... does that remind you of anything?" and if they say no, use the anti-negative response. In this case it would be: "Don't worry if you can't remember. Let's try something else that you may remember."

BLAME THE CLIENT FOR NOT HAVING THE BOT WORK!!!
 
Don't forget to get access to a list of the most popular names, to increase the likelihood of getting hits.
 

Back
Top Bottom