[List All Topics] [Start Topic] [Search] [Show Thread Index] [Help]
[Not Logged In / Login / Register]
Sort: [Threaded | Date]
 |
|
|||
 |
Turing Test = stupidest thing I've ever heard |
[Top] [Show Index] [Reply to this post] |
 |
|
Machines will always fail the Turing Test unless they have a personal history equivalent to a biological humanoid. All the human participant would have to do is ask, "where did you go to school?" or "what do you do for a living - how did you get that job?" The answers would have to be one of two varieties, both absurd.
[Reply] |
||||
|
 |
|
 |
|
|
|
|||
|
AI = ALIEN Intelligence |
[Top] [Show Index] [Reply to this post] |
|
|
I don't agree that an AI will never be able to mimic human responses enough to pass the Turing test -- I'm sure it will. In fact, I bet I could program an ANSWERING MACHINE to fool you and most humans 99% of the time.
|
||||
|
|
|
|
|
|
|
|||
|
Re: AI = ALIEN Intelligence |
[Top] [Show Index] [Reply to this post] |
|
|
An interactive answering machine.. never thought of it like that! It really uncovers the essence of the Turing Test as one to test the effectiveness of a deception.
Martin, a Pulitzer Prize nominee, has written one hundred textbooks—more than any other living person. That's TEXTBOOKS. Not self-help books, or novellas. |
||||
|
|
|
|
|
|
|
|||
|
Re: AI = ALIEN Intelligence |
[Top] [Show Index] [Reply to this post] |
|
|
Yeah. I don't understand why I was able to pick up a PILE of James Martin's book at MICRO CENTER for pennies on the dollar.
|
||||
|
|
|
|
|
|
|
|||
|
Re: Turing Test = stupidest thing I've ever heard |
[Top] [Show Index] [Reply to this post] |
|
|
The answer to your problem is very simple.
|
||||
|
|
|
|
|
|
|
|||
|
Re: Turing Test = stupidest thing I've ever heard |
[Top] [Show Index] [Reply to this post] |
|
|
It will be stupid to program a computer to have such limitations. I think we're actually very close to a agreement. I think it's very well absurd to waste time trying to fool a human into thinking that a machine is another human - we shouldn't program them with human "limitations," as you refer to them. However, the Turing Test is by definition unrestricted. This is from the definition of "Turing Test" from Dictionary.com: A human holds a written conversation on any topic with an unseen correspondent (nowadays it might be by electronic mail or chat). If the human believes he is talking to another human when he is really talking to a computer then the computer has passed the Turing test and is deemed to be intelligent.
The key to the stupidity of the Turing Test is the part where it says "any topic," it's lack of restriction. If certrain topics were out, such as personal topics, then the test would no longer be viable. The Loebner Prize (http://www.loebner.net/Prizef/loebner-prize.html) is an unrestricted Turing Test held annually. The prize is a solid gold medal. However, people like myself have criticized it, associating it with the likes of a "parlor game." Here's a very good (albeit somewhat scathing) paper about why an unrestricted Turing Test like the Loebner Prize is stupid: http://xxx.lanl.gov/abs/cmp-lg/9404002 And here's the response from Loebner: http://loebner.net/Prizef/In-response.html (I should probably stop throwing around the word "stupid," for I have much respect for aspiring AI programmers. I meant to use the word to expose the restrictions required to pass an UNrestricted test :/ ) |
||||
|
|
|
|
|
|
|
|||
|
Re: Turing Test = stupidest thing I've ever heard |
[Top] [Show Index] [Reply to this post] |
|
|
Machines will always fail the Turing Test unless they have a personal history equivalent to a biological humanoid. I agree and have given similar arguments in the past. your scenario displays the issue clearly though. the obvious conclusion would be that an AI must gain experience over time. to learn to be intellegent. for some reason AI coders never seem to agree with this and believe and keep believing they can hard code the intellegence into the system. its all just smoke and mirrors and striving for a short cut. well, there isn't one, and we should be more patient and develop the AI accordingly. we should follow the perspective that a newly generated AI system should be regaurded, respected, and treated like a new born child. human education should be a map for what to focus on and what sensory inputs to develop in what order. granted, human education techniques can be just as endlessly debated. but not even fundamentals are currently considered in most AI developments. this is of course if you are looking to pass the turing test(and not cheat). there are other directions for AI to take (and have taken) as many on this forum seem to agree. Don't forget that personal questions involve feelings too, and feelings are the effects of chemical reactions, not pure logic. Humans have specific chemical reactions that reward certain actions, and punish for others. It will be stupid to program a computer to have such limitations. this is a very narrow perspective. ignorant in a non-insulting use of the word. chemical reactions are at all times logical. Its the complexity and quantity of the reactions that surpases our ability to calucate them. therefore they appear chaotic. but they are not. it is still deterministic. and the positive/negative reinforcement system is a fundamental process of such highly distributed, complex systems. its a limitation of a computer's linear processing to not use such a function, not a failure of nature's design. it would help reduce buffer overruns and the like. coders must step away from the mindset of smallest number of operations posible per process that was needed for early machines when confronted with such concepts as massive parallelism and distributed computing. griffman |
||||
|
|
|
|
|
|
|
|||
|
Re: Turing Test = stupidest thing I've ever heard |
[Top] [Show Index] [Reply to this post] |
|
|
griffman,
|
||||
|
|
|
|
|
|
|
|||
|
Re: Turing Test = stupidest thing I've ever heard |
[Top] [Show Index] [Reply to this post] |
|
|
If this helps the discussion of brain tissue and rewards:
|
||||
|
|
|
|
|
|
|
|||
|
Re: Turing Test = stupidest thing I've ever heard |
[Top] [Show Index] [Reply to this post] |
|
|
The thing I didn't understand about this whole experiment is this: why the heck should this brain do anything at all ?, in the article it wasn't specified that this brain was given any rewards whatsoever cell growth/use is a reward in itself. neurons that go for long periods of time with no use tend to die off faster than cells that are healthy, in use, and circulating chemicals. this is why I called the positive/negative renforcement a fundamental property. death is a failure and a negative outcome. hence the outrage of kurzweil (though I don't think he sees it in that context) the neurons don't care what task they are given or even if they are given a task. they just want to grow. a functioning neuron is more likely to grow. all this means is we can't look for "intellegence" on the human level (for a turing test) from a system of less processing than would be expected in a human brain. the intention and consiousness is not some hard to find piece but emergent from the colective whole. as for the flight experiment.... I think it was sort of ironic more than anything. it proved how little brain power it takes to fly an f-16 8). what would be interesting is what this makes posible. is this a method for implanted abilities? it would rely on our ability to integrate "learned" neurons into an existing "conscious" brain network. the system of training a cluster of neurons is a process that needs to be perfected but I wouldn't rule it out as imposible (the implanting is a different story, that could fail.) So here's a question..... If we assume that to effectivly pass a turing test, an AI must have a learned experince similar to a human lifetime. If we also assume that a computer coulf be programed with an AI process that could evolve similar to an infant's mind. the question is. would that AI process take the same period of time to reach a grown human level of intellegence (i.e. 20+ years)? hence the predictions of a singularity must be given more time (i.e. 30-50 more years) here's two ways to look at this, both optimistic. 1. no, it wont, with proper configuration, the continual "learning" process of a machine would take less time to reach maturity since a minimum of 1/3 of a human's life is not spent learning. 2. it could be considered that much of the AI learning development has already been in the works for 20+ years. although not chronologically similar to a human's life year by year, the end result could be equal. griffman |
||||
|
|
|
|
|
|
|
|||
|
Re: Turing Test = stupidest thing I've ever heard |
[Top] [Show Index] [Reply to this post] |
|
|
I think that even though a human brain has great potential, it wires itself really slow. As an evidance, it takes at least 2-3 years of non-stop stimuli before it start to master our basic skills (talk, walk etc.).
|
||||
|
|
|
|
|
|
|
|||
|
Re: Turing Test = stupidest thing I've ever heard |
[Top] [Show Index] [Reply to this post] |
|
|
I think that one can try to write a simulator for human life (something like the Sims), and let the computer experience it in first person, and simulate something like 20 years in 1 week. |
||||
|
|
|
|
|
|
|
|||
|
Re: Turing Test = stupidest thing I've ever heard |
[Top] [Show Index] [Reply to this post] |
|
|
You know, modern human experiences are so plastic, that that could actually work. |
||||
|
|
|
|
|
|
|
|||
|
Re: Turing Test = stupidest thing I've ever heard |
[Top] [Show Index] [Reply to this post] |
|
|
The answer to your problem is very simple.
Actually most of the things humans come up with -- say and ask -- are actually quite repetitive. Ask any social counselor or extremely good looking woman. 1 trillion questions?! Ah, quite preposterous. If you took 25 people and gave them each a microphone and asked them to spit out as many different answers to the question: Do birds fly? as possible, you would quickly see a very finite, quasi-repetitive, database of answers develop. An audio database that could be randomly played back in a Turing test that would fool many humans. Of course you would have to repeat this exercise with numerous additional questions. As I have said before, I am working on developing an answering machine patter that I bet will be able to fool most people that call my office. This is a sort of Turing test in and of itself -- even though my answering machine will still be as dumb as Ramona(1) James Jaeger ----------------- (1) Hey Ramona, no offence. I know that someday a butterfly will emerge from the caterpillar: that someday Ray's going to develop you into the sexiest, most intelligent AI chick on the planet and make a fool out of all of us non-believers. |
||||
|
|
|
|
|
|
|
|||
|
Re: Turing Test = stupidest thing I've ever heard |
[Top] [Show Index] [Reply to this post] |
|
|
You think its stupid, because YOU are a machine.
|
||||
|
|
|
|
|
|
|
|||
|
Re: Turing Test = stupidest thing I've ever heard |
[Top] [Show Index] [Reply to this post] |
|
|
I'm flattered ;) |
||||
|
|
|
|
|
|
|
|||
|
Re: Turing Test = stupidest thing I've ever heard |
[Top] [Show Index] [Reply to this post] |
|
|
Your assertion that machines will never pass the Turing Test is weakly supported. You have divided responses to personal history queries into two categories: deceitful and honest.
|
||||
|
|
|
|
|
|
|
|||
|
Re: Turing Test = stupidest thing I've ever heard |
[Top] [Show Index] [Reply to this post] |
|
|
You have divided responses to personal history queries into two categories: deceitful and honest. Is that inappropriate or incomplete? If so, how would you categorize answers to personal history questions? In the first category, you assume that a machine must be pre-programmed with false answers to personal questions. Not exactly. I used the umbrella term "preplanned deception," which means that the programmer could use any myriad of tactics to deceive, only one of which would be to directly program false answers. As stated in one of the above sub-threads, a programmer could run the AI through a "school program," complete with virtual teachers, classmates, and scenarios. However, when the human subject asks the AI "Where did you go to school?" the human is speaking about a human school, not a virtual one. If the AI in question answers "Boston College," presumably because that was the name the programmer gave to the virtual school, then the human has been deceived. The AI would pass because of a preplanned deception on part of the programmer. The programmer could also create a series of "avoidance algorhythms," designed to skillfully change the topic. Whereas a human would change the topic to keep from talking about something intimate or inappropriate, the AI would do it because there really is nothing to talk about. This is misdirection on part of the programmer, not proof of intelligence. The Turing Test does not ask, "who can program the most intelligent AI?" it asks "who can program the best deception?" In the second category, if a machine had a true personal history, it could potentially answer this type of question without even being intelligent. Well of course it could. That's a pretty weak counterargument :P Intelligence can be measured by way better means than a Turing Test. This may sound trite, but think of the intelligence tests you've taken. You were probably asked to read a paragraph and summarize it, solve equations, complete analogies, and rotate geometrical figures in your head. Some IQ tests also measure knowledge of history and important facts. Your intelligence is measured by your ability to independently absorb information and associate it with your continued experience - not just on whether you can carry on a conversation (although that's part of it). This is why I think the Turing Test is silly. It doesn't measure human intelligence - it doesn't measure anything, save what it takes to distract a person long enough from discovering they're not talking to a human. we must consider what the Turing Test says about a machine that was successful. Is it determined to be intelligent, or does it simply say that the machine has deceived the human judge into thinking so? Let's leave that for another thread. Actually, that's what this thread is all about. :/ |
||||
|
|
|
|
|
|
|
|||
|
Re: Turing Test = stupidest thing I've ever heard |
[Top] [Show Index] [Reply to this post] |
|
|
On redundancy: One of the rights in the Universal Declaration of Human Rights is the right to education. If an AI were given human rights, there would be no need for it take a Turing Test - it would be human. But what if we were to ask the question, "does this AI have a right to education?" The spirit in which the Turing Test was conceived seems as though it should verify the "humanity" of a person and whether or not it deserves human rights. Appearances aren't allowed in the test because it could lead to prejudice. But the test doesn't help because of catch-22.
|
||||
|
|
|
|
|
|
|
|||
|
AI: The Simplest Thing Imaginable |
[Top] [Show Index] [Reply to this post] |
|
|
The Turing Test does not ask, "who can program the most intelligent AI?" it asks "who can program the best deception?" Exactly. Thus, at the end of the day, the Turing Test is really of cosmetic value. When humans can't (or won't) answer a (sensitive) question, because they don't know (or won't give out) the answer, they will often simply say: "I don't know" or "none of your business." This is pretty human, no? An AI computer, attempting to pass the Turing Test, could do the same thing and thus be convincingly human too. So the Turing Test is really all about deception NOT intelligence. As an AI gained (recorded) more and more possible responses to a given patter (or set of questions), it would appear to be more "conscious," if not "intelligent" at some point. After all, intelligent OR dumb people equally appear to be "conscious." More often than not, their answers are nonsequitor, illogical and irrational. AI programmers attempting to get machines to make any sense may be the essence of the problem. So what's all the whoopla about consciousness these past 3,000 years?! I say programming an AI machine to HEAR and RECORD responses, and then MIMIC a human patter IS SO SIMPLE, NO ONE HAS BEEN ABLE TO DO IT YET BECAUSE THEY ARE LOST IN THE IDEA THAT IT MUST BE COMPLEX. 95% of the things that come out of a human are PURE reactions to a given stimuli. There's no "thought" involved. So why are AI programmers attempting to make machines THINK!? Let the machines EVOLVE with their own "thinking" as De Garis is doing. Just because WE humans don't necessarily recognize it as thinking or intelligence, doesn't mean it's not. James Jaeger |
||||
|
|
|
|
|