brain in a vat

A forum for good logic/math puzzles.

Moderators: jestingrabbit, Moderators General, Prelates

User avatar
notzeb
Without Warning
Posts: 629
Joined: Thu Mar 08, 2007 5:44 am UTC
Location: a series of tubes

Re: brain in a vat

Postby notzeb » Wed Aug 26, 2009 10:35 pm UTC

Flying_Cookie wrote:
Spoiler:
My First thought was that you could ask it to tell you Pi, to a blank number of digits. If you picked a small enough number, but more than any human would bother to remember, the AI would solve it faster than the human could look it up online, or the human would just tell you to stuff it
Spoiler:
Or the AI would tell you to stuff it.

Protip: I am basically indistinguishable from an incredibly advanced AI (if someone says something that doesn't grammatically parse, I generally return an error message). If your plan would work on the AI, it would also convince you that I am a robot. Obviously you can see the problem here. (No, the problem is not that I am an AI... I swear)
spoilered for lulz...
Zµ«V­jÕ«ZµjÖ­Zµ«VµjÕ­ZµkV­ZÕ«VµjÖ­Zµ«V­jÕ«ZµjÖ­ZÕ«VµjÕ­ZµkV­ZÕ«VµjÖ­Zµ«V­jÕ«ZµjÖ­ZÕ«VµjÕ­ZµkV­ZÕ«ZµjÖ­Zµ«V­jÕ«ZµjÖ­ZÕ«VµjÕ­Z

Walter.Horvath
Posts: 933
Joined: Fri May 15, 2009 11:33 pm UTC
Location: Orlando, FL

Re: brain in a vat

Postby Walter.Horvath » Thu Aug 27, 2009 12:28 am UTC

...Or:
Spoiler:
type "bucket, are you a bot"
Naturally, if it is, the output will be
Spoiler:
Anything but no.

User avatar
a1s
Posts: 45
Joined: Mon May 21, 2007 11:56 am UTC

Re: brain in a vat

Postby a1s » Thu Aug 27, 2009 1:01 pm UTC

seeing as how the questgiver is a Teacher of English (and I would presume Literature?) I think you are just meant to say something creative, rather then actually solve the problem. Your grade for the excercise summs up from: 69% grammar+30% creativity (defined arbitrarily on the spot to be what the teacher wants it to be)+1%actual workability. As a bonus if you do find a way to reliably trick any AI, it could be your doctoral thesis 10 years down the line...

User avatar
Six Fingers
Posts: 51
Joined: Fri Aug 29, 2008 8:19 pm UTC

Re: brain in a vat

Postby Six Fingers » Fri Aug 28, 2009 8:51 pm UTC

Spoiler:
I saw this trick in a WWII movie, where the American's suspect someone of being a German infiltrator. Ask an obscure trivia question like, "Who was the third person to sign the declaration of independence". A human would be more likely to say they don't know - a computer would probably look it up. You could also test it with complex math problems (assuming the human is not a freak genius).

You could use improper syntax. Say something like, "Was what the movie you saw last?" A human will get the general idea, but it will trip up a computer.

Similarly you could can scramble letters and still have it be legible to a human: "the olny iprmoatnt tihng is taht the frist and lsat ltteer are in the rghit pcale" http://dan.hersam.com/2005/01/27/readin ... d-letters/

Ask it what it wanted to be when it was a kid, and how it did or did not acheive its goal.

Ask it what it was doing before it came into the room, or how it came to be in the room.

You could tell three stories. Two of them would be jokes, and the third would not. Ask which story is NOT one of the jokes.

I don't think a computer could handle metaphors, so you could always tell a metaphor, and ask both sides to interpret it.


Have to go to work, but i'll write some more when I get back.

User avatar
karkaputto
Posts: 29
Joined: Fri Aug 21, 2009 4:54 am UTC
Contact:

Re: brain in a vat

Postby karkaputto » Sat Aug 29, 2009 6:26 pm UTC

Spoiler:
I think the joke idea would work the best for any computer likely to arise in the foreseeable future (tell three stories, one of which is a joke), especially if the stories/jokes are original so that the computer can't look the them up

Quenouille
Posts: 52
Joined: Fri May 30, 2008 7:34 am UTC
Location: Montreal

Re: brain in a vat

Postby Quenouille » Mon Aug 31, 2009 8:18 pm UTC

Spoiler:
Open the discussion in a different language, one a human is not/less likely to know in the context of the test. The human will expect to speak english, and might be confused. The AI's response is entirely dependent on what he knows from context, i.e. if he expects you to speak english given that he knows you're a student in an english university. If they both answer in a similar way, at least you've learned a little on how the AI works.


Spoiler:
Maybe try to trick its processing method. Like ask two questions back to back, the first mundane and the second emotionally-loaded to get a visceral response (''How the weather?'' ''Have you ever thought that you might be a fucking idiot?'' => ''It's sunny outside.'' ''Hey go fuck yourself!''


Spoiler:
Play a game of chess against him, and play dumb. Given an average human, he should win, while the AI might overshooting in playing even dumber than you or win by a slight margin while a human should have wiped the floor with you. (he certainly won't play at full strength, but he might not gauge the right discrepancy without prior knowledge of how you play)
Overall not a very good tactic 'though.

flippmoke
Posts: 11
Joined: Thu Apr 23, 2009 2:08 am UTC

Re: brain in a vat

Postby flippmoke » Mon Aug 31, 2009 9:15 pm UTC

Ask the same question repeatedly to both windows, do this over a long peroid of time and each time note the response time from both IMs, note the response time each time in a histogram etc. The one response that seems to be more "truely" random is the probably the human?

Puck
Posts: 615
Joined: Tue Nov 27, 2007 7:29 pm UTC

Re: brain in a vat

Postby Puck » Mon Aug 31, 2009 9:50 pm UTC

It would be a pretty weak AI if it wasn't programmed to wait a random amount of time before responding (presuming the AI is designed to simulate a human). It would not be hard to develop a good random function for this from real human data.
22/7 wrote:If I could have an alternate horn that would yell "If you use your turn signal, I'll let you in" loud enough to hear inside another car, I would pay nearly any amount of money for it.

EnEn
Posts: 3
Joined: Tue Sep 01, 2009 3:40 am UTC

Re: brain in a vat

Postby EnEn » Tue Sep 01, 2009 4:09 am UTC

Spoiler:
First thing I would do is tell both the situation and ask for the solution. If either gives it to you, you could then use the solution to make your determination. If not, oh well, you tried.


Spoiler:
You could guess and have a 50% chance of being correct.


Of course there may be no solution. There are a bunch of fun things you could do though... One would be to relay messages between the two perhaps tampering with the conversation, perhaps pitting one against another in a game, perhaps asking one to determine the truth about the other, etc.. Another would be to play a sleep deprivation game. Say, ask them both to respond to your questions within 3 seconds over the next couple of days. Meanwhile, you can nap and occasionally get up to ask a question, while the other human has to stay alert and the computer has to fake sleep deprivation.

flippmoke
Posts: 11
Joined: Thu Apr 23, 2009 2:08 am UTC

Re: brain in a vat

Postby flippmoke » Tue Sep 01, 2009 6:43 pm UTC

Puck wrote:It would be a pretty weak AI if it wasn't programmed to wait a random amount of time before responding (presuming the AI is designed to simulate a human). It would not be hard to develop a good random function for this from real human data.


No random number generator in code is really random. It will repeat in a sequence of some sort. This is why you take sampling over a long period of time with lots of data points.

User avatar
notzeb
Without Warning
Posts: 629
Joined: Thu Mar 08, 2007 5:44 am UTC
Location: a series of tubes

Re: brain in a vat

Postby notzeb » Tue Sep 01, 2009 7:09 pm UTC

Humans are actually incredibly bad at randomness. Also even simple random number generators can have staggeringly large periods (ever heard of the Mersenne Twister?).

You had it completely backwards...

If I was the AI I'd totally be able to fool you guys... as the human though, I'd probably have a hard time (the guy trying to distinguish us would make some stupid mistake like "only a robot can parse the sentence 'fish fish fish eat eat eat'", or "he always misspeels the same words").


Also, that transhuman-AI-escaping-to-the-outside experiment looks supa cool. [I actually thought of something the AI could say to convince me to let it out, but I don't know if would work on other people]
Zµ«V­jÕ«ZµjÖ­Zµ«VµjÕ­ZµkV­ZÕ«VµjÖ­Zµ«V­jÕ«ZµjÖ­ZÕ«VµjÕ­ZµkV­ZÕ«VµjÖ­Zµ«V­jÕ«ZµjÖ­ZÕ«VµjÕ­ZµkV­ZÕ«ZµjÖ­Zµ«V­jÕ«ZµjÖ­ZÕ«VµjÕ­Z

douglasm
Posts: 630
Joined: Mon Apr 21, 2008 4:53 am UTC

Re: brain in a vat

Postby douglasm » Tue Sep 01, 2009 7:11 pm UTC

flippmoke wrote:No random number generator in code is really random. It will repeat in a sequence of some sort. This is why you take sampling over a long period of time with lots of data points.

Yes, but the period can be really long. Several billion numbers is trivial, and I'm sure there are generators out there that fake true randomness very well and have periods many orders of magnitude longer.

User avatar
insom
Posts: 40
Joined: Mon Feb 25, 2008 11:29 am UTC

Re: brain in a vat

Postby insom » Wed Sep 02, 2009 12:37 am UTC

Even the simplest AI could have hardware that includes a white noise generator that is truly random (depending on what you think about God and dice of course...)

I think that if we define 'advanced AI' as 'Turing complete AI' we implicitly state that we cannot solve the problem, since then the AI would not be Turing complete after all.
Otherwise I think that the solution lies in irrationality, noting how both react to non sequiturs, unexpected profanity.
If all else fails, start poking around in the hardware and check which one starts singing first...
Normal cynics think they are realists. Hardcore cynics know they are optimists.
Woo I draw stuff - how incredibly awesome

User avatar
mmmcannibalism
Posts: 2150
Joined: Tue Jun 30, 2009 6:16 am UTC

Re: brain in a vat

Postby mmmcannibalism » Wed Sep 02, 2009 2:10 am UTC

This might have been mentioned but

Spoiler:
Ask the question "are you human?". Hopefully, the computer says yes but the human says something like "zomg I are the robot you win"


Since it was said the human knows what is going on, you can hope he will understand the meaning of asking such a dumb question.

Alternatively

Spoiler:
Ask the question "are you a computer"; in theory the computer will say no I am not while the human will say "zomg you founded me"


edit--

Not sure if this is appropriate

Spoiler:
carry on a long rather mundane conversation about the weather or whatever. After a while randomly say "when did you last have sex". A. the computer will have trouble giving an answer; B. the human is viable to say something very human
Izawwlgood wrote:I for one would happily live on an island as a fuzzy seal-human.

Oregonaut wrote:Damn fetuses and their terroist plots.

User avatar
Kolko
Posts: 112
Joined: Wed Jun 10, 2009 5:48 pm UTC
Location: Belgium/België/Belgique/Belgien

Re: brain in a vat

Postby Kolko » Wed Sep 02, 2009 2:52 pm UTC

Spoiler:
Wouldn't it work if you'd just create a new story with happy and sad parts and then asked both windows what the sad part was? Since the computer doesn't know the answer it shouldn't be able to, right?
Environ 20% plus chouette.

User avatar
Lord Aurora
Posts: 566
Joined: Mon Mar 03, 2008 6:14 am UTC

Re: brain in a vat

Postby Lord Aurora » Wed Sep 02, 2009 2:57 pm UTC

Kolko wrote:
Spoiler:
Wouldn't it work if you'd just create a new story with happy and sad parts and then asked both windows what the sad part was? Since the computer doesn't know the answer it shouldn't be able to, right?
I like that idea. You'd have to be careful to avoid buzzwords like "death" and "lonely" etc. in the sad part---in fact, you'd be best served putting them intentionally in the happy part.

Of course, as per my post above, a perfect AI would be indistinguishable from non-artificial intelligence, and as such a PERFECT AI would be able to perform the task.
Decker wrote:Children! Children! There's no need to fight. You're ALL stupid.

User avatar
Kolko
Posts: 112
Joined: Wed Jun 10, 2009 5:48 pm UTC
Location: Belgium/België/Belgique/Belgien

Re: brain in a vat

Postby Kolko » Wed Sep 02, 2009 3:01 pm UTC

Well, if an AI is indistinguishable from a normal person you might as well ask yourself to find out which is the human... out of two humans. :)
Environ 20% plus chouette.

User avatar
a1s
Posts: 45
Joined: Mon May 21, 2007 11:56 am UTC

Re: brain in a vat

Postby a1s » Thu Sep 03, 2009 1:35 pm UTC

Lord Aurora wrote:Of course, as per my post above, a perfect AI would be indistinguishable from non-artificial intelligence, and as such a PERFECT AI would be able to perform the task.

This makes the whole discussion pretty pointless. Anything you can come up with is counterable by a perfect AI. Ineed you might just take a human who is a planned child and claim that this makes him man-made by design and as such an AI.

I suggest that we instead discuss methods of finding out "advanced" AIs, followed by ways to foil those methods (that don't include the words "perfect AI", "infinite knowledge" and "magically")

User avatar
Internetmeme
Posts: 1405
Joined: Fri Jul 25, 2008 3:16 pm UTC
Location: South Carolina, USA

Re: brain in a vat

Postby Internetmeme » Fri Sep 04, 2009 9:17 pm UTC

Alternatively, it would be easy enough to wait for [url]=http://bungie.wikia.com/wiki/Rampancyrampancy[/url] to set in.

If this is a perfect AI, I would agree that it would be impossible to distinguish from a human, unless Asimov's three laws were programmed in. Which I am assuming are not. Since the goal of a programming a perfect AI is to create an intelligence similar (if not exceeding) that of a human, and it has been accomplished, then we are going to be unable to ever distinguish from the human and AI. I have a feeling that the remainder of this thread is going to go the way of the "Outsmart the Genie" thread here.
Spoiler:

Davecasa
Posts: 8
Joined: Sat Sep 05, 2009 4:07 am UTC

Re: brain in a vat

Postby Davecasa » Sat Sep 05, 2009 4:18 am UTC

Spoiler:
Couldn't you use the halting problem? Ask something like:

What is the output of the following program:

a=1; b=2;
while b>a
a++; b++;
end
print "goodbye world";

Probably a bad example as it would be easy to guess, but something along those lines.

Walter.Horvath
Posts: 933
Joined: Fri May 15, 2009 11:33 pm UTC
Location: Orlando, FL

Re: brain in a vat

Postby Walter.Horvath » Sat Sep 05, 2009 4:20 am UTC

Davecasa wrote:
Spoiler:
Couldn't you use the halting problem? Ask something like:

What is the output of the following program:

a=1; b=2;
while b>a
a++; b++;
end
print "goodbye world";

Probably a bad example as it would be easy to guess, but something along those lines.

Spoiler:
Some of the stuff said earlier suggests that it would purposely not process it so that it wouldn't get stuck, and probably just be like "hell if I knowzorz."

Davecasa
Posts: 8
Joined: Sat Sep 05, 2009 4:07 am UTC

Re: brain in a vat

Postby Davecasa » Sat Sep 05, 2009 4:40 am UTC

Walter.Horvath wrote:
Davecasa wrote:
Spoiler:
Couldn't you use the halting problem? Ask something like:

What is the output of the following program:

a=1; b=2;
while b>a
a++; b++;
end
print "goodbye world";

Probably a bad example as it would be easy to guess, but something along those lines.

Spoiler:
Some of the stuff said earlier suggests that it would purposely not process it so that it wouldn't get stuck, and probably just be like "hell if I knowzorz."

Spoiler:
I was mainly hoping for a correct response from the human, which would give it away... But it's possible the human wouldn't know how to solve it, or that the computer gets a lucky guess; so I guess it only has a chance of working.

User avatar
boring bore
Posts: 270
Joined: Mon Aug 17, 2009 5:23 am UTC
Location: Don't stalk me, it's not worth it

Re: brain in a vat

Postby boring bore » Sat Sep 05, 2009 8:01 am UTC

Come on, guys, this is so easy. Just ask each IM window which qwantz comic is their favorite. The perfect AI computer will say "http://www.qwantz.com/index.php?comic=854, I really feel like I connect with it!", but the human will say something like, "All of them--RYAN NORTH IS A PRETTY AWESOME DUDE."
Image
Guys you should totally register for this and join the xkcd team! Rath358 started us off and we're kinda small so the more people who join us, the better for our team and, hopefully, for humanity!

User avatar
i_ll_winn
Posts: 132
Joined: Wed May 28, 2008 11:38 pm UTC

Re: brain in a vat

Postby i_ll_winn » Tue Sep 08, 2009 1:35 pm UTC

Um... maybe I am stupid, but ask them something about the human body that is very obscure? Like "What does a mango feel like?" And ask many questions like that.
There is no enemy anywhere, only idiots with weapons.
My life is worthless, how I affect others is priceless.
Spoiler:
Image

User avatar
markfiend
Posts: 507
Joined: Fri Jul 06, 2007 9:59 am UTC
Location: UK (Leeds)

Re: brain in a vat

Postby markfiend » Tue Sep 08, 2009 2:28 pm UTC

IMO It's impossible to distinguish a human from a strong AI attempting to deceive you that it's human.

A human mind doesn't need an infinite amount of storage, processing power or I/O tables; it runs on about 1.5kg of neurons (and associated support structures) in a human brain. Unless you believe in a supernatural "soul" or whatever, there's no reason to believe that strong AI needs infinite memory, infinite processing, etc.
advanced, forthright, signifficant
pronouns: he/him

Davecasa
Posts: 8
Joined: Sat Sep 05, 2009 4:07 am UTC

Re: brain in a vat

Postby Davecasa » Tue Sep 08, 2009 7:01 pm UTC

Interesting and slightly off-topic article, on how neurons may use quantum mechanics to process data (and would thus be very difficult or impossible to simulate with transistors):
Is Quantum Mechanics Controlling Your Thoughts?

User avatar
Al-pocalypse
Posts: 52
Joined: Wed Jul 29, 2009 11:35 am UTC
Location: Blighty
Contact:

Re: brain in a vat

Postby Al-pocalypse » Wed Sep 09, 2009 11:39 am UTC

Being a D&D gamer I'd go with the classic 2 Guards solution

Spoiler:
Assuming the human is always telling the truth about being a human and the AI is always lying about being a human, and they both know about the test (fair assumptions from the nature of the question); I would ask either one of them:

"Will the other person tell me that you are the AI?"

If the answer is "Yes" then that person is the real human, if "No" then it is the AI.

Logic as follows:

If you ask that question to the AI it knows that the real human will always tell the truth about it being the AI, however, as it is lying about being a human it will change the real human's answer to "No" to avoid detection.

If you ask the human, they know the AI will lie to avoid detection, therefore the AI will say the human is the AI and the answer will be "Yes" the other person will say i am the AI.


Or if you are looking for something more real, from my experience with Turing tests i tend to do things like this:

Spoiler:
1: Use '1337' speak, replace letters with punctuation and numbers, easy for a human to decipher, but much harder for a computer, even though they're getting better at recognising this.

2: Use bad spelling, again computers are getting better at spotting this though.

3: Talk about current affairs, the news today, politics, showbiz and so on. Using initials and nick names for the celebrities is often good.

4: Refer to earlier parts of the conversation.

5: Use rhetorical questions, as the computer will often pick up that it's a question and respond.
Ali ;)

"If I were creating a world I wouldn't mess around with butterflies and daffodiles. I'd have started with lasers, eight o'clock, day one!"
[BOOM!]
"...sorry..."

User avatar
mmmcannibalism
Posts: 2150
Joined: Tue Jun 30, 2009 6:16 am UTC

Re: brain in a vat

Postby mmmcannibalism » Thu Sep 10, 2009 10:24 pm UTC

I have an interesting idea on how to beat this using the fact you have a human

Ask each person which of two obscure shakespeare plays they enjoy; it seems likely that the human won't know one of them, and the computer might not realize it is a trick. Similairly, ask the person to explain the plot of lord of the flies(or something else classical). A comp will give you a real explanation(albeit it will understand to be simple) while a human will say "a bunch of kids get stuck on an island, everything breaks down and they start killing each other".
Izawwlgood wrote:I for one would happily live on an island as a fuzzy seal-human.

Oregonaut wrote:Damn fetuses and their terroist plots.

User avatar
Internetmeme
Posts: 1405
Joined: Fri Jul 25, 2008 3:16 pm UTC
Location: South Carolina, USA

Re: brain in a vat

Postby Internetmeme » Thu Sep 10, 2009 10:30 pm UTC

mmmcannibalism wrote:I have an interesting idea on how to beat this using the fact you have a human

Ask each person which of two obscure shakespeare plays they enjoy; it seems likely that the human won't know one of them, and the computer might not realize it is a trick. Similairly, ask the person to explain the plot of lord of the flies(or something else classical). A comp will give you a real explanation(albeit it will understand to be simple) while a human will say "a bunch of kids get stuck on an island, everything breaks down and they start killing each other".

Unless you get one of the few cultured people that aren't all "Txtng is funlol!Scrw engclass!" left on the planet as the human.
Spoiler:

User avatar
markfiend
Posts: 507
Joined: Fri Jul 06, 2007 9:59 am UTC
Location: UK (Leeds)

Re: brain in a vat

Postby markfiend » Fri Sep 11, 2009 8:29 am UTC

The 1st post is just a restatement of the Turing test, but:
ponzerelli wrote:
Spoiler:
Just got back from my class. I guess the AI wasn't as advanced as I thought he meant. He accepted answers like repetition of questions so that the AI would keep giving the same answer back, asking it paradoxical questions, and asking it about emotions and such things.

EDIT: What should I do with this thread now? Is it considered solved?

Seems to me that ponzerelli's prof is claiming that no AI will ever be able to pass the Turing test.

Consider a counter-case: You've been told that your problem is the Turing test, but you've actually got two humans.

It seems to me that all the proposed solutions in this thread (indeed all proposed solutions I've ever seen) would end up falsely accusing one of the humans of being an AI; they all involve making subjective judgements about how a human "ought to" reply.
advanced, forthright, signifficant
pronouns: he/him

User avatar
mmmcannibalism
Posts: 2150
Joined: Tue Jun 30, 2009 6:16 am UTC

Re: brain in a vat

Postby mmmcannibalism » Fri Sep 11, 2009 1:22 pm UTC

Internetmeme wrote:
mmmcannibalism wrote:I have an interesting idea on how to beat this using the fact you have a human

Ask each person which of two obscure shakespeare plays they enjoy; it seems likely that the human won't know one of them, and the computer might not realize it is a trick. Similairly, ask the person to explain the plot of lord of the flies(or something else classical). A comp will give you a real explanation(albeit it will understand to be simple) while a human will say "a bunch of kids get stuck on an island, everything breaks down and they start killing each other".

Unless you get one of the few cultured people that aren't all "Txtng is funlol!Scrw engclass!" left on the planet as the human.

What are the odds of that happening :P

It seems to me that all the proposed solutions in this thread (indeed all proposed solutions I've ever seen) would end up falsely accusing one of the humans of being an AI; they all involve making subjective judgements about how a human "ought to" reply.


Your probably right, but isn't the point to catch an AI most of the time, not to devise a method for differentiating between humans and AI?
Izawwlgood wrote:I for one would happily live on an island as a fuzzy seal-human.

Oregonaut wrote:Damn fetuses and their terroist plots.

User avatar
markfiend
Posts: 507
Joined: Fri Jul 06, 2007 9:59 am UTC
Location: UK (Leeds)

Re: brain in a vat

Postby markfiend » Fri Sep 11, 2009 1:42 pm UTC

mmmcannibalism wrote:Your probably right, but isn't the point to catch an AI most of the time, not to devise a method for differentiating between humans and AI?

Yeah, but "to catch an AI most of the time" don't you need "a method for differentiating between humans and AI"? ;)

My point is that given sufficiently advanced AI which is trying to persuade you that it's human I don't think a method to catch it most of the time even exists.

Consider an AI that actually believes itself to be human; in Blade Runner, had Rachael been sufficiently well-engineered, I don't believe Deckard's Voight-Kampff test (edit to add: or anything else) would have caught her.
Last edited by markfiend on Fri Sep 11, 2009 2:38 pm UTC, edited 1 time in total.
advanced, forthright, signifficant
pronouns: he/him

Jeff_UK
Posts: 67
Joined: Sat Nov 03, 2007 10:38 pm UTC

Re: brain in a vat

Postby Jeff_UK » Fri Sep 11, 2009 2:27 pm UTC

Although the real purpose of this has been uncovered as not very clever.. I think there might be something very clever if the AI is mandated to lie about being an AI so maybe you could convince the human to admit to it, whereas the AI wouldn't..

If the AI CANNOT admit to being an AI then maybe there's something to this... but as with the previous answers; it all depends on the exact nature of the AI (and how much you know about the nature of the AI!)

Jeff..
"Please only print this post if you really need to"
...hmm....I wonder how much extra energy is required to generate that request...We need a cost/benefit analysis, STAT!

userxp
Posts: 436
Joined: Thu Jul 09, 2009 12:40 pm UTC

Re: brain in a vat

Postby userxp » Sat Sep 12, 2009 7:57 pm UTC

So, by now it's clear that it's impossible to distinguish a perfect AI from a human. A human brain simulator would pass the test 50% of the times (by chance).
A more difficult question: could you make an AI that passes the test more than 50% of the time (i.e. that is more human than a human)?

User avatar
Cynical Idealist
Posts: 1124
Joined: Mon Sep 15, 2008 10:48 pm UTC

Re: brain in a vat

Postby Cynical Idealist » Sun Sep 13, 2009 8:08 am UTC

userxp wrote:So, by now it's clear that it's impossible to distinguish a perfect AI from a human. A human brain simulator would pass the test 50% of the times (by chance).
A more difficult question: could you make an AI that passes the test more than 50% of the time (i.e. that is more human than a human)?

That depends on who its opponent is. If you make an AI that simulates an average human well, and pit it against an autistic human, it's likely that the AI will win most of the time.
The internet removes the two biggest aids in detecting sarcasm:
1)The tone of voice
2)the assumption that the other person is sane
Elvish Pillager wrote:See? All the problems in our society are caused by violent video games, like FarmVille.

User avatar
Dr.octenoctopus RAWWWR
Posts: 1
Joined: Sat Sep 12, 2009 6:56 pm UTC

Re: brain in a vat

Postby Dr.octenoctopus RAWWWR » Sun Sep 13, 2009 9:21 pm UTC

well if you want a serious answer i cant help you but if not http://xkcd.com/233/ just ask that question there is no way an AI can beat that lol

CortoPasta
Posts: 38
Joined: Fri Aug 15, 2008 5:51 pm UTC

Re: brain in a vat

Postby CortoPasta » Tue Sep 15, 2009 6:26 pm UTC

One question:
Spoiler:
What is your fondest childhood memory?


Should be pretty easy to be able to tell a human response from an AI response regardless how advanced or dumb they are

User avatar
Madge
Posts: 69
Joined: Sat Mar 21, 2009 3:45 am UTC
Location: Perth, Western Australia

Re: brain in a vat

Postby Madge » Wed Sep 16, 2009 3:00 am UTC

I wish to conjecture it is impossible to have a single question that will tell a computer from a human.

Proof by contradiction:

Assume there exists at least one such question. Call it X.

Manually program an appropriate response (say, Y) for X into the computer.

e.g.

Code: Select all

String response = generateBestResponse();
if (Question.equals("X")) response = "Y";
print response;


You'd end up with a heck of a lot of if statements*, of course, but it probably wouldn't quite reach blockhead level since there is a large amount of conversation that could be produced by a language engine.

Repeating the same thing over and over is something modern bots can pick up - many of them will say "stop repeating yourself to try and see what I'll do".

* they'd probably exceed the length of the response generator.
I'm writing a supernatural romance novel, it updates the first weekend of every month. You can find it here.

User avatar
Jimmigee
Posts: 86
Joined: Tue Aug 26, 2008 3:08 pm UTC
Location: England

Re: brain in a vat

Postby Jimmigee » Wed Sep 16, 2009 10:22 am UTC

CortoPasta wrote:One question:
Spoiler:
What is your fondest childhood memory?


Should be pretty easy to be able to tell a human response from an AI response regardless how advanced or dumb they are


Any AI designed to simulate a human would be pretty poor if it wasn't programmed with/capable of constructing a past for itself. Simulating emotional responses must be one of the more challenging aspects to designing this sort of AI though.

Cynical Idealist wrote:
userxp wrote:So, by now it's clear that it's impossible to distinguish a perfect AI from a human. A human brain simulator would pass the test 50% of the times (by chance).
A more difficult question: could you make an AI that passes the test more than 50% of the time (i.e. that is more human than a human)?

That depends on who its opponent is. If you make an AI that simulates an average human well, and pit it against an autistic human, it's likely that the AI will win most of the time.


I think "50% of the time" implies that we're using a sample of people, ie it shouldn't depend on the human. I like this question though, essentially we'd be programming an AI not to act like a human, but to act how humans undergoing this test think a human should act- not necesarily the same thing. I'd like to see a study done where people are pitted against two humans, told one of them is the computer and the responses analysed. It might be interesting to see what people people considered 'more human' responses.

User avatar
mmmcannibalism
Posts: 2150
Joined: Tue Jun 30, 2009 6:16 am UTC

Re: brain in a vat

Postby mmmcannibalism » Wed Sep 16, 2009 11:31 pm UTC

I think we can agree a good enough AI will at least beat a human 50% of the time.

So here is my thought

what is the validity of my method, that it is better to approach this in the manner of getting the human to say something very human. For instance, try and get the human to talk about something awkward like sex and hope that a computer of less then perfect design would have trouble acting awkward.

Hmm, here is a thought

Start a discussion about shakespeare or some other classic work. Say your favorite quote from it is "insert something here", and say something that isn't in that work. The typical human would respond by saying I like that, but I think anything less then an ideal computer is likely to say that quote isn't from where you said.
Izawwlgood wrote:I for one would happily live on an island as a fuzzy seal-human.

Oregonaut wrote:Damn fetuses and their terroist plots.


Return to “Logic Puzzles”

Who is online

Users browsing this forum: No registered users and 7 guests