Intelligence and Machines as well as the Turing Test
_1. Is the Turing Test a sufficient test? That is, if a machine passes the test, would you agree it is intelligent?
It is not about giving the correct answer it is about how the machine`s answers resemble those of a human being. As applied in the manner described in Turing`s paper, I think the test is quite witty in that it seems to make intelligence more than an aptitude measurable through IQ tests: it seems to be able to instigate rational thinking on the part of the interrogator driven by not only truthful remarks of the human involved but also the “fake” answers of the computer or machine. Even though the test seems to be practically helpful in determining whether a machine can imitate a human being or not I have some drawbacks in my mind that also makes me suspicious of whether any machine that passes the test is intelligent:
1) Does the interrogator know that one of the other players/interlocutors is a machine? If not, would doing this would put the machine in a more challenging position?
2) What if the machine is assigned the role of interrogator? In this sense, if we assume that it is actually the interrogator who enrolls in a real intelligent behavior (reaching logical conclusions based on some truth values gained through interaction with the interlocutors), can a machine imitated a human in the role of an interrogator? Closely related, other “control” conditions may also be applied such as all parties would be machines, the machine can take the role of the interlocutor whose purpose is to help the interrogator etc. Accordingly,
3) If we assume that human intelligent behavior can include mistakes, learning from them and applying these to reach the ultimate solution of the test, can a machine can mimic this too in different test roles assigned to it?
So, is a machine passes the Turing test, it would mean that this approximates human intelligence from one perspective: imitating a human behavior. However, if other possibilities above are reflected on this approximation may fall short of the whole human intelligence.
2. Is the Turing Test a neccesary test? That is, does a machine have to pass this test in order to be intelligent?
My answer is both yes and no. If we reason from another perspective and ask the question: If a person fails the test, is she or he not intelligent? My answer would be ‘no’ since failing such a test no matter how close it addresses intelligence, does not mean that a person would show some other intelligent behaviors in some other contexts. As Pinker puts it, intelligence refers to being able to make rational decisions even in the face of obstacles driven from beliefs and desires sometimes in an intuitive manner. Does the Turing test include all these aspects of intelligence that we are aware of other than information processing? So, in this sense, we may need the test to determine a machine is intelligent or not from certain perspective(s) (I think in this case it is information processing) thus imitating human intelligent behavior to a certain extent. As Pinker also says, information processing is of utmost importance for intelligence. Therefore, at least from this perspective I think the Turing Test may work to a certain extent.
3. Will a machine ever pass the Turing Test? Why or why not?
I think that a machine will be able to pass the Turing test eventually if and only if the sort of thought going on in a Turing Test and the way it is put into written human language are described with well-specified rules and the machine is programmed to work on the basis of these rules. Moreover, it seems to me to be important to follow both causal and inferential aspects of symbols (in the form of written words etc.) This is important since in the Test the machine should be able to examine not only its output but also the other interlocutor`s output as well as those of the interrogator in order to able to give human-like responses to the interrogator.
We may never be able to produce a machine that has exactly the same physiology of the human brain that can do the things specified above. However, as Pinker emphasized, it is not the matter or energy running through it, but what each part of it stands for and the pattern of changes designed to reflect truthful relationships that make something smart. In this sense, I am of the opinion that even human mistakes work on a truthful basis. Take language as an example. While learning their mother tongues, children do never make wild mistakes despite the poverty of language input they take in. More interestingly, they do not so if they are learning a second, or third language. So, even developmental language mistakes humans make seem to be rule-governed in a sense indicating that they have pattern. So, it is my belief that a machine or program or whatever you call it can be built that can imitate human mind no matter it is size, shape etc. and it can pass the Turing test.
4. Will a machine ever be intelligent? Why or why not? (This may or may not be the same answer to the previous question).
As I stated in my answers above, I think, in terms of information processing, a machine can be intelligent even though it may work on different symbols than the ones used by human beings. A example would be ‘ME’ writing this paper: It sounds like I am listening to myself or my inner-speech and this is exactly the same as what appears on the Word page on my computer screen. Obviously, my symbols “Phonemes” I keep in my head and my computer`s “0” and “1” symbols are not the same, the product is hopefully the same though. Interestingly though, I may not do the same thing if my keyboard has symbols in the form of what my computer can process (not in the form of letters that would equal the phonemes in my head). Still though, when I listen to myself while writing these, I can understand (hopefully) and process symbols relating to me, and my computer turns them into a written form based on its own symbols.
However, if we go beyond this information processing view of intelligence and assume that there is also a part of intelligence independent of information processing, causal and inferential learning, computation etc., my answer would be “I do not know”.
It is not about giving the correct answer it is about how the machine`s answers resemble those of a human being. As applied in the manner described in Turing`s paper, I think the test is quite witty in that it seems to make intelligence more than an aptitude measurable through IQ tests: it seems to be able to instigate rational thinking on the part of the interrogator driven by not only truthful remarks of the human involved but also the “fake” answers of the computer or machine. Even though the test seems to be practically helpful in determining whether a machine can imitate a human being or not I have some drawbacks in my mind that also makes me suspicious of whether any machine that passes the test is intelligent:
1) Does the interrogator know that one of the other players/interlocutors is a machine? If not, would doing this would put the machine in a more challenging position?
2) What if the machine is assigned the role of interrogator? In this sense, if we assume that it is actually the interrogator who enrolls in a real intelligent behavior (reaching logical conclusions based on some truth values gained through interaction with the interlocutors), can a machine imitated a human in the role of an interrogator? Closely related, other “control” conditions may also be applied such as all parties would be machines, the machine can take the role of the interlocutor whose purpose is to help the interrogator etc. Accordingly,
3) If we assume that human intelligent behavior can include mistakes, learning from them and applying these to reach the ultimate solution of the test, can a machine can mimic this too in different test roles assigned to it?
So, is a machine passes the Turing test, it would mean that this approximates human intelligence from one perspective: imitating a human behavior. However, if other possibilities above are reflected on this approximation may fall short of the whole human intelligence.
2. Is the Turing Test a neccesary test? That is, does a machine have to pass this test in order to be intelligent?
My answer is both yes and no. If we reason from another perspective and ask the question: If a person fails the test, is she or he not intelligent? My answer would be ‘no’ since failing such a test no matter how close it addresses intelligence, does not mean that a person would show some other intelligent behaviors in some other contexts. As Pinker puts it, intelligence refers to being able to make rational decisions even in the face of obstacles driven from beliefs and desires sometimes in an intuitive manner. Does the Turing test include all these aspects of intelligence that we are aware of other than information processing? So, in this sense, we may need the test to determine a machine is intelligent or not from certain perspective(s) (I think in this case it is information processing) thus imitating human intelligent behavior to a certain extent. As Pinker also says, information processing is of utmost importance for intelligence. Therefore, at least from this perspective I think the Turing Test may work to a certain extent.
3. Will a machine ever pass the Turing Test? Why or why not?
I think that a machine will be able to pass the Turing test eventually if and only if the sort of thought going on in a Turing Test and the way it is put into written human language are described with well-specified rules and the machine is programmed to work on the basis of these rules. Moreover, it seems to me to be important to follow both causal and inferential aspects of symbols (in the form of written words etc.) This is important since in the Test the machine should be able to examine not only its output but also the other interlocutor`s output as well as those of the interrogator in order to able to give human-like responses to the interrogator.
We may never be able to produce a machine that has exactly the same physiology of the human brain that can do the things specified above. However, as Pinker emphasized, it is not the matter or energy running through it, but what each part of it stands for and the pattern of changes designed to reflect truthful relationships that make something smart. In this sense, I am of the opinion that even human mistakes work on a truthful basis. Take language as an example. While learning their mother tongues, children do never make wild mistakes despite the poverty of language input they take in. More interestingly, they do not so if they are learning a second, or third language. So, even developmental language mistakes humans make seem to be rule-governed in a sense indicating that they have pattern. So, it is my belief that a machine or program or whatever you call it can be built that can imitate human mind no matter it is size, shape etc. and it can pass the Turing test.
4. Will a machine ever be intelligent? Why or why not? (This may or may not be the same answer to the previous question).
As I stated in my answers above, I think, in terms of information processing, a machine can be intelligent even though it may work on different symbols than the ones used by human beings. A example would be ‘ME’ writing this paper: It sounds like I am listening to myself or my inner-speech and this is exactly the same as what appears on the Word page on my computer screen. Obviously, my symbols “Phonemes” I keep in my head and my computer`s “0” and “1” symbols are not the same, the product is hopefully the same though. Interestingly though, I may not do the same thing if my keyboard has symbols in the form of what my computer can process (not in the form of letters that would equal the phonemes in my head). Still though, when I listen to myself while writing these, I can understand (hopefully) and process symbols relating to me, and my computer turns them into a written form based on its own symbols.
However, if we go beyond this information processing view of intelligence and assume that there is also a part of intelligence independent of information processing, causal and inferential learning, computation etc., my answer would be “I do not know”.