Talk:Blockhead (thought experiment)

Sorry if I'm not editing this correctly, but I have a burning question. What would happen if you simply sent blockhead this message: "Wow what do you think of this list? [Here, the finite bank of all sentences]" Wouldn't the program simply crash because it is logically impossible for this input to be on the list of possible inputs? —Preceding unsigned comment added by 128.220.29.101 (talk) 23:01, 9 March 2011 (UTC)

Isn't it true that a sentence could theoretically be infinite in length?

- ProgressivePantheist


 * The article talks about "correct responses". There is no question which could have an infinitly long sentence as a "correct response". The computer would then be programmed to have enough correct responses to response to a limited number of sentences for a limited amount of time, making such a computer logically possible, since there is nothing infinite in all that. --SuperBleda 18:34, 10 August 2005 (UTC)

This is assuming that the conversation follows a logical progression, and that the entire conversation remain strictly grammatically correct, neither of which are certain to occur. Actually, there are a lot of logical falacies with this argument - I'll see if I can find records of other scientists who disagreed with Block, to try to make the article more balanced. Does anyone know anything about this?


 * No, it does not assume any rules for the scentences. You're just wronly assuming it would because you forgot that you also could add hash keys for all variations of all scentences, including all bad structured or even completely sensless ones. The argument still holds. Unfortunately you neither say what those logical falacies are that you think there are, nor do i find any one of them. And as you can see i'm good at logic. ;) (Ok, doesn't matter because it does not prove that my arguments are flawless. ;) -- 212.100.48.53 16:34, 21 October 2006 (UTC)
 * Some links would be really appreciated. And a list of arguments proving the falacy would be even nicer... if there actually are ones. At least i find it perfectly clear and actually pretty simple, because correcly seen it only states: "There is a finite list of stuff you can say. And to pass the test there's a finite list of stuff you can answer. Thus limiting the whole conversation to a finite set of ways." So a simple hash of "question => {answers}" with a random generator choosing one of the answers out of the set for the matchin "question" would actually pass every turing test by definition. -- 212.100.48.53 16:34, 21 October 2006 (UTC)

I also strongly disagree that there are more possibilities for scentences than there are particles in the universe, when you exclude scentences from outside the universe. This is a simple following from the fact that you can't form a scentence more complex than the universe, as long as the universe includes the scentence. -- 212.100.48.53 16:34, 21 October 2006 (UTC)

There are an infinite number of possible sentences, just imagine all the sentences consisting of one word repeated as a list, you can always make the sentence longer and so the set of such sentences is infinite (though boring). The blockhead requires that sentences not be infinitely long and that the conversation length also be limited, else the table would be infinitely large and thus not possible. However it would contain all possible inputs regardless if they even have the appearance of sentences so that if you feed the blockhead a garbled string of words and meaningless characters it would be able to locate a suitable response in the table, likely a question as to what you meant or whether the interface hardware is malfunctioning. This table would already be far to large for the universe to contain if you allowed sentences to be only a few words long and the conversation to last only a few minutes. -- MasterTriangle12 06:57, 20 May 2015 (UTC)


 * I can't believe nobody added a complete section regarding the Blockheads in Gumby. This omission is a glaring black hole on the face of Wiki-knowledge.--Blockhead 23:02, 26 September 2005 (UTC)


 * What prevents dou from doing it yourself, when nobody else seems to know them. Besides i would appreciate to read what you wrote. :) -- 212.100.48.53 16:43, 21 October 2006 (UTC)


 * There is also an amazing hip hop producer on the Ninja Tune label named Blockhead. Check him out! www.ninjatune.net


 * Yup. He's found under Blockhead (disambiguation) or from there under Blockhead (music).
 * Off topic, but i recommend you listen to "You've got maelstroem" if you don't know him. Check out the www.bigbeats.ru radio station to hear the track very now and then. :) [just klick the link that looks like "listen to mp3". I'm geman so i don't understand a word of the page too. Got it from shoutcast. The music rules! ;] -- 212.100.48.53 16:43, 21 October 2006 (UTC)


 * Ummm Hello?! Stephanie J. Block anyone?! Check HER out! She's so amazing and talented on and off the stage and has the most amazing fans in the world! They call themselves "blockheads," and that word sure does describe them perfectly!
 * http://www.stephaniejblock.com


 * If she matters in terms of the wikipadia rules (look 'em up to see), go add her yourself under "Blockheads" redirecting to "Stephanie J. Block". Thank you for your contribution. :)

Knowledge size vs. size of the universe

 * Although the number of sentences required for a 30 minute conversation is said to be greater than the number of particles in the universe, it is clear that such a machine could at least logically exist.

Well, if there are more sentences than particles in the universe, and since I don't know of any information storage system, or compression algorithm, which doesn't require at least one particle of the universe per sentence, then in my opinion it is most definitely not clear that such a machine can exist. To say the least. – Tintazul msg 21:42, 26 September 2006 (UTC)


 * Regardless of the number of sentences in any given language, it's quite possible for it to exceed the number of particles in the universe because a sentence is only a *possible* arrangement of particles. If you required all the sentences to be written down or uttered at the same time, it would be a different story. Consider that the number of possible arrangements of particles in the universe far exceeds the total number of those particles, or that the number of arrangements of pieces on a chess board is greater than 32.
 * On the other hand the above quote is uncited, uncitable and weaselly. If there's a better way of saying the same thing... Leushenko 12:56, 25 September 2007 (UTC)


 * But the point that Tintazul was making (although in a more circumspect way) is that the Blockhead machine can't be implemented because it would have to be larger than our universe. Hence the argument fails for our universe. A thought experiment using the Blockhead machine to prove that it is possible for a non-intelligent machine to pass the Turing test in some universe where Blockhead can be built is basically similar to a thought experiment using a Perpetual Motion Machine to prove that it is possible for a non-intelligent machine to pass the Turing test in some universe where a PMM can be built: the argument may be sound but it is irrelevant to our universe where neither Blockhead nor a PMM can be built. -- Derek Ross | Talk 05:36, 27 October 2008 (UTC)


 * Particles in the universe aside, Block's argument (at least as described) fails merely on the assumption that there is a finite number of possible correct responses. New words and new meanings for old words are created every day. The moment Blockhead comes upon a word it doesn't recognize but should understand (so "I don't understand" is not a correct response), it fails. It can't be assumed to pass a Turing test if it has to ask what you mean by "sheeple" or "VPILF"—and, by definition, Blockhead can't figure anything out from context; it only has canned responses. Someone somewhere must have pointed that out already. --Tysto (talk) 17:42, 29 April 2009 (UTC)

I have not read the original paper but it appears this argument fails to consider that even if there was a computer able to store every possible sentence, then this does not make a machine that can suddenly pass a Turing test. The computer must also store every possible question and have some mechanism for connecting every possible question to (at least) one appropriate response. So now we're talking relational database. Remember too that the computation has to be performed quickly enough to mimic a human. It doesn't sound likely to me. In any event the point Block is trying to make is that a non-intelligent machine could pass the Turing Test. Isn't that stating the obvious? The whole point of the Turing test is that a machine does not have to be intelligent - it just has to create the illusion that it is intelligent. —Preceding unsigned comment added by Salome777 (talk • contribs) 11:51, 16 May 2010 (UTC)


 * I thought that the idea behind the Turing Test was that if a machine could pass it, then it would be reasonable to conclude that it was intelligent. In short, intelligence is hard to define, but the test is a concrete and simple means of verifying it; consider the analogy that there is no quick means to algorithmically describe all solutions to an NP-Complete problem, but there is a quick means to verify any given one; I've always looked at the turing test the same way, we can't describe all intelligent machines, though we can easily verify one. At any rate, there is a finite number of lists of sentences of length K for any K; thus, we can just imagine a computer that stores all lists of conversations containing K sentences, to get its next response it randomly selects a list that is prefixed by the conversation thus far, then returns the next sentence. Now, suppose you use N sentences in the turing test, there is no way to determine if the passing machine was of the form above with K >= N; this holds no matter how large N is since there is always a larger K and always a logically possible machine for each K. Thus, without looking at the internals, you cannot verify intelligence. That the machine is not physically possible is not the point, intelligence as a concept should not depend upon the number of particles in the universe; thus, since this argument is against the Turing Test being able to verify intelligence, any argument based upon physical impossiblity is not a counterargument:) Phoenix1177 (talk) 05:28, 4 June 2010 (UTC)

Cleverbot
Cleverbot seems to be an attempt to pass the Turing test similarly to this idea (though not taking it to the logical extreme of being pre-programmed with ALL possible sentences) - the aim is to record a LOT of possible sentences, and the contexts in which they make sense. Combined with sufficient processing power, this apparently works quite well. They claim that in an official test (with more processing power than is given to their online version), 59% of people thought it was human, which was less than the rate for real humans, but still quite impressive. — Preceding unsigned comment added by 124.149.177.170 (talk) 12:51, 6 April 2012 (UTC)