Jneale
February 2, 2005, 05:48 PM
John Searle and the Chinese Room: A Defence of Turing and Behaviourism
We are entering or indeed are already upon one of the most exciting periods of epistemic history. Epistemology concerns issues that were always considered to be suited to the realm of philosophy, essentially that the concepts it discussed were too abstract in their nature to be considered science. Today these issues are no longer involved in purely philosophical debate but in cognitive science and its derivatives also.
I shall refer to the mind as if I was advocating substance dualism, normally associated as part of Descartes meditations; however I don’t wish to enter into a debate over monism, dualism, as it really has no effect on the issues I want to deal with. Every psychological attribute has now, due to considerable recent advances in neuro-anatomical science, been associated with a specific location in the brain. The ability to assimilate visual input with the occipital lobe, the process of sorting objects into categories with the temporal lobe, Intentionality with the parietal lobe and reasoning with the frontal lobe. Even semantics have been associated with activity in parts of the brain.
Semantics is a frequently used term in cognitive science and it is important for me to clarify what I intend by the term. Language has two levels, syntax and semantics. Syntax is the sound, series of letters the symbol I use to represent the semantics behind language. When I read this essay you hear the symbols, syntax, and interpret its semantic meaning, when you ask questions at the end you will do so through syntax and I shall be interpreting with semantics. Despite this relationship being made slightly more complex by the presence of an internal monologue the concept remains the same.
One of the most interesting parts of cognitive and computational sciences is the artificial intelligence or A.I. It forms a large part of discussion in modern epistemology but also in ethics and computer programming.
Where do these ideas of Artificial intelligence originate from? It was in fact a British mathematician who first discussed the possibilities of computer science who also suggested A.I., his name was Alan Turning and he published his first paper in 1936 even before the creation of ASCII binary or any of today’s governing rules of computing. This single work given the rather un-catchy title, ‘On computable numbers, with an application to the entscheidungsproblem’ was a founding part of computer science and made possible the advent of computers as we know them today. Although the implications of this paper are very exciting to read it is inordinately complex and tedious, so I have paraphrased and slightly embellished this extract. It is important to note here that when he mentions a computer he means a human who undertakes a computation.
‘The behaviour of a computer is determined by the symbols which he is observing and his ‘state of mind’ at that moment. We may suppose that there is a bound to the number of symbols which the computer can observe at one moment. If he wishes to observe more he must use successive observations.’ If we are to imagine the operations performed by computer to be split up into the most simple base operations imaginable and each of these is denoted by a symbol on a tape ‘if we know the state of the system of we would also know the sequence of symbols on the tape’
Turing then goes on to suppose we made a machine that did exactly this then we could create a computing machine. This is the basis of today’s stored program computers. Indeed anything that ‘computes’ in the manner described above counts as a Turing machine. The Turing machine hypothesis was then used to solve a problem surrounding a concept known as mechanical method this is a mathematical notion of a class of problems and results (functions) that can be solved or attained by following fixed rules.
Around the same time Alonzo Church, an American logician, working independently from Turing solved the same problem through a concept of ‘lambda-definability’, which unfortunately being of beyond degree standard mathematics I cannot explain to you. However essentially he then again linked this to recursive functions, incidentally this is the basis for a technique used by modern day computer programming.
In an appendix of a later paper Turing then shows both of these solutions to be essentially the same. Resulting in the Church-Turing THESIS which says that every function whose values can be attained mechanically, by a fixed method are computable by a Turing Machine.
This thesis which is currently thought of as a conjecture as opposed to a definition is the starting point for the entire conception of Artificial Intelligence. If it is true electronic computers MUST be able to possess at least human level intelligence. This is because cognitive science tells us that any cognitive process no matter how complex is eventually split into primitive computations or routines that are mechanical in their nature. Thus anything a brain can do a Turing machine or computer could in theory do at least as effectively.
This thesis was the reason behind large amounts of financial investment in Artificial intelligence. Computer science truly believed that such a target was attainable. Machines slowly surfaced who, it was claimed, could understand stories, and over time they did so with increasing ability.
It would be prudent of me to introduce to you a very important distinction, between two conceptions. Although these had always been present in discussion of AI, it was John Searle who was the first to define them adequately. These distinctions are not approaches to AI but conceptions of its aims.
The first, Strong AI sees AI as an attempt to design and build a machine that displays a range of psychological attributes, things such as reason or intentionality, even consciousness and eventually feeling and emotion. Someone who subscribes to strong AI must also agree to these propositions says Searle:
“a) An appropriately programmed computer really would have, or be, a mind in the same sense that you or I have.�
And,
“b) Its following the programs in question would explain its ability to do the psychological things it does.�
Something that is usually associated with, although not always defined as part of Strong AI, is the belief in the Turing test. This is a test Turing proposed that is essentially borne out of behaviourism, due to Descartes, cogito ergo sum it is only through empirical means that I can suggest, not know you are intelligent. It is merely because you exhibit similar behaviour to me that I conclude you are also intelligent. Turing says that if an invigilator cannot discriminate between the responses of a machine and those of a human then it should be attributed with intelligence.
The opposite of this is logically named Weak A.I. and it says simply that electronic digital computers are powerful instruments, for helping us to model, and thereby understand, the mind.
These distinctions are often used whilst committing a straw man fallacy, the term strong AI has become somewhat derogatory partly due to the effect of Searle’s arguments and also due to the claims made by early subscribers to Strong AI. In 1957 Simon and Newell two advocates of these ideas of Strong AI and founding fathers of AI research, expressed this view.
‘There are now in the world machines that think, that learn and that create. Moreover their ability to do these things is going to rapidly increase until… they can handle a range of problems coextensive to those to which the human mind has been applied. ‘
An example of such a program is one of the story understanding programs I previously described. These machines could take a simple story such as, to use Searle’s example:
‘A man went into a restaurant and ordered a hamburger. When the hamburger arrived it was burned to a crisp, and the man stormed out of the restaurant angrily, without paying for the hamburger or leaving a tip.’
When asked "Did the man eat the hamburger?" the machine would be able to answer, "No, he did not.� It could do this with a range of different although fairly similar stories and a range although again fairly similar questions and answers.
It was exactly the claim that this constituted understanding that Searle originally set out to refute, to do so he created this argument:
1. Programs a purely syntactical (formal)
2. Minds (human ones at least) have semantics, mental contents.
3. Syntax by itself is neither the same as or sufficient for semantic content.
Therefore:
4. Programs by themselves are neither constitutive of nor sufficient for minds.
Although he has rephrased this same argument in countless volumes the basic structure remains the same. The following thought experiment the ‘Chinese Room Argument’ was or at least Searle would have it that this argument was intended to illustrate the truth value of the third premise. Many people see it as a reformulation rather than the underlying argument of the thought experiment however in truth this is unimportant.
‘Suppose that I’m locked in a room and given a large batch of Chinese writing. Suppose furthermore (as is indeed the case) that I know no Chinese, either written or spoken, and that I’m not even confident that I could recognize Chinese writing as Chinese writing distinct from, say, Japanese writing or meaningless squiggles. To me, Chinese writing is just so many meaningless squiggles. Now suppose further that after this first batch of Chinese writing I am given a second batch of Chinese script together with a set of rules for correlating the second batch with the first batch. The rules are in English, and I understand these rules as well as any other native speaker of English. They enable me to correlate one set of formal symbols with another set of formal symbols, and all that "formal" means here is that I can identify the symbols entirely by their shapes. Now suppose also that I am given a third batch of Chinese symbols together with some instructions, again in English, that enable me to correlate elements of this third batch with the first two batches, and these rules instruct me how to give back certain Chinese symbols with certain sorts of shapes in response to certain sorts of shapes given me in the third batch. Unknown to me, the people who are giving me all of these symbols call the first batch a "script," they call the second batch a "story," and they call the third batch "questions." Furthermore, they call the symbols I give them back in response to the third batch "answers to the questions," and the set of rules in English that they gave me, they call the "program." Now just to complicate the story a little, imagine that these people also give me stories in English, which I understand, and they then ask me questions in English about these stories, and I give them back answers in English. Suppose also that after a while I get so good at following the instructions for manipulating the Chinese symbols and the programmers get so good at writing the programs that from the external point of view—that is, from tile point of view of somebody outside the room in which I am locked—my answers to the questions are absolutely indistinguishable from those of native Chinese speakers. Nobody just looking at my answers can tell that I don’t speak a word of Chinese. Let us also suppose that my answers to the English questions are, as they no doubt would be, indistinguishable from those of other native English speakers, for the simple reason that I am a native English speaker. From the external point of view—from the point of view of someone reading my "answers"—the answers to the Chinese questions and the English questions are equally good. But in the Chinese case, unlike the English case, I produce the answers by manipulating uninterpreted formal symbols. As far as the Chinese is concerned, I simply behave like a computer; I perform computational operations on formally specified elements. For the purposes of the Chinese, I am simply an instantiation of the computer program.’
So what has Searle achieved in this experiment? Well he has illustrated a machine that possesses no understanding whatsoever. However it appears it could pass the Turing test as described earlier. In effect it could fake being a Chinese speaker. Secondly that contrary to the belief of Strong A.I. a formal program operating as Turing originally conceived, one that simply manipulates symbols could never constitute a mind. This argument forms the most damning critique of Strong A.I. Calling into question one of its most founding parts the Church-Turing thesis described earlier. Mathematical criticisms of this idea have already been put forward establishing the possibility that there exists within the brain non computable neural processes.
There are many oppositions to this second part of the Chinese Room, a typical argument that although one program as illustrated above does not understand Chinese, a compliment or group of them – an entire system would, essentially he has committed a part whole fallacy. However clearly if this program was able to give out purely syntactic responses without ever engaging in semantics, then any number of further similar programs grouped with the first could take these syntactical responses and further operate them, none of them gleaning any kind of semantic meaning. It would be just a more complex yet equally ineffective program. I must concede that I have not yet seen an effective denial of this second half of the Chinese room argument.
I do however have questions relating to the denial of the Turing test. I feel there is one underlying assumption in Searle’s argument which no one has ever truly exploited. The reason for this is perhaps because if supporters of Strong AI were to use it they would most likely be pointing out fatal flaws in their own thinking. This assumption is that the room in its current form would be able to pass a Turing test. If we were agreed upon the previously mentioned theory of non computable neural processes it would make this even less likely. As a conversation is not a series of unrelated questions answers and statements, but there is a correlation between the end of a conversation and the beginning. To illustrate this further,
Imagine overhearing this conversation between the tester T and the Intelligent Person P:
T: Hi there my name is James I would like to apply for a job.
C: Ok James just go through that blue door and take an application form.
T: Thanks, oh and also is there a toilet near here?
C: Yes its signposted from the same room you get an application form.
T; Ok thanks a lot for your help.
C: No problem. Oh and good luck with the job application.
This conversation flows from one question to the next and each part bears some relevance to earlier parts of the conversation. A program operating like the Chinese Room suggests would not exhibit such qualities although it may hold the conversation it would not assimilate earlier information it would sound more disjointed, perhaps like this.
T: Hi there my name is James id like to apply for a job.
C: Ok through the blue door and take an application form.
T: Thanks, oh and also is there a toilet near here?
C: Yes, through the blue door and its signposted.
T: Ok thanks a lot for your help.
C: No problem.
As you can see although this may hint at understanding it does not flow in the same manner, the computer has no recollection of describing the blue door because it was merely an automatic response manipulating the words to give an appropriate answer. Although this machine would be very good at mimicking intelligent behaviour it would, and this would become increasing clear over the extended test period, be unable to make references to information a human would take from the conversation because it would not understand it. It would not behave adequately to pass the Turing test. In effect it would treat a conversation as a series of statements rather than a conversation.
This is how I always understood a Chinese room program to operate, however, there is another way that such a device could be envisaged, which Ned Block first raised he called this machine a ‘blockhead’. Block said that there is a finite amount of conversations a computer could be involved in. if the test lasted 2 weeks there is an enormous, but not infinite number of possible conversations. If the program could automatically respond not only to each individual statement, but to the conversation as an entire entity it would be able to pass the test perfectly. In a Turing test this machine would appear to be intelligent. Thus despite the near impossibility that a block program could be built I must concede that the Turing test in its current form is not an adequate test. However I suggest it would be very obvious to anyone who looked internally at the machine and saw its program that they would know that it was not an intelligent device but merely droid.
To resolve this issue I would like to return briefly to the idea which arose earlier, that of Descartes, cogito ergo sum, and the simple fact I can never know that any of you are intelligent in the same way I am. I assume you are because the behaviour you exhibit is so similar to mine and when you react to situations you appear to do it in the same way as I do. Suppose for a moment however I had a serious doubt that you were intelligent, I truly believed that you merely appeared to be operating in a similar fashion to myself but you were not thinking at all. You were simply an enormous ‘Blockhead’ program. It would not be a considered a bad idea for me to compare myself to you, on a neurological level. I mentioned earlier how every function of the mind has been attributed to activity in some part of the brain. With an MRI scan I could see that not only did you externally behave in a similar manner to myself, but internally as well.
Such lengths would perhaps be a little bit over the top, it is more likely I would just be placed in some kind of institution, however if computer science was claiming to have created an artificial intelligence I don’t believe examining its behaviour to such an extent would be considered over the top in the least. I would suggest therefore that the Chinese Room argument forms part of a two pronged attack. First it suggests that syntax is not sufficient for semantics, this part of the argument I believe to be sound, and I would not dispute it. I would therefore deny Strong AI under perhaps its original definition, and I would agree that true intelligence could not be borne out of pure syntax, or symbol manipulation. The second attack that the Chinese room argument makes is a refutation of the Turing test. If we were to take the original definition of behaviourism I would also agree with this first part, however I do not feel this is the only way in which we could define such an idea of behaviourism.
All mental phenomena however we choose to refer to them have some impact on the physical world. I wish to suggest that this impact should also be referred to as behaviour. Not simply the obvious behaviours perceivable solely through our senses, but those which we can perceive with the aid of tools, the way in which the program works is just as important and just as much a component of its behaviour as the way it outwardly appears to work. Therefore I would suggest that the Chinese room argument does not completely deny a Turing test and this part of Strong AI if no other remains intact.
We are entering or indeed are already upon one of the most exciting periods of epistemic history. Epistemology concerns issues that were always considered to be suited to the realm of philosophy, essentially that the concepts it discussed were too abstract in their nature to be considered science. Today these issues are no longer involved in purely philosophical debate but in cognitive science and its derivatives also.
I shall refer to the mind as if I was advocating substance dualism, normally associated as part of Descartes meditations; however I don’t wish to enter into a debate over monism, dualism, as it really has no effect on the issues I want to deal with. Every psychological attribute has now, due to considerable recent advances in neuro-anatomical science, been associated with a specific location in the brain. The ability to assimilate visual input with the occipital lobe, the process of sorting objects into categories with the temporal lobe, Intentionality with the parietal lobe and reasoning with the frontal lobe. Even semantics have been associated with activity in parts of the brain.
Semantics is a frequently used term in cognitive science and it is important for me to clarify what I intend by the term. Language has two levels, syntax and semantics. Syntax is the sound, series of letters the symbol I use to represent the semantics behind language. When I read this essay you hear the symbols, syntax, and interpret its semantic meaning, when you ask questions at the end you will do so through syntax and I shall be interpreting with semantics. Despite this relationship being made slightly more complex by the presence of an internal monologue the concept remains the same.
One of the most interesting parts of cognitive and computational sciences is the artificial intelligence or A.I. It forms a large part of discussion in modern epistemology but also in ethics and computer programming.
Where do these ideas of Artificial intelligence originate from? It was in fact a British mathematician who first discussed the possibilities of computer science who also suggested A.I., his name was Alan Turning and he published his first paper in 1936 even before the creation of ASCII binary or any of today’s governing rules of computing. This single work given the rather un-catchy title, ‘On computable numbers, with an application to the entscheidungsproblem’ was a founding part of computer science and made possible the advent of computers as we know them today. Although the implications of this paper are very exciting to read it is inordinately complex and tedious, so I have paraphrased and slightly embellished this extract. It is important to note here that when he mentions a computer he means a human who undertakes a computation.
‘The behaviour of a computer is determined by the symbols which he is observing and his ‘state of mind’ at that moment. We may suppose that there is a bound to the number of symbols which the computer can observe at one moment. If he wishes to observe more he must use successive observations.’ If we are to imagine the operations performed by computer to be split up into the most simple base operations imaginable and each of these is denoted by a symbol on a tape ‘if we know the state of the system of we would also know the sequence of symbols on the tape’
Turing then goes on to suppose we made a machine that did exactly this then we could create a computing machine. This is the basis of today’s stored program computers. Indeed anything that ‘computes’ in the manner described above counts as a Turing machine. The Turing machine hypothesis was then used to solve a problem surrounding a concept known as mechanical method this is a mathematical notion of a class of problems and results (functions) that can be solved or attained by following fixed rules.
Around the same time Alonzo Church, an American logician, working independently from Turing solved the same problem through a concept of ‘lambda-definability’, which unfortunately being of beyond degree standard mathematics I cannot explain to you. However essentially he then again linked this to recursive functions, incidentally this is the basis for a technique used by modern day computer programming.
In an appendix of a later paper Turing then shows both of these solutions to be essentially the same. Resulting in the Church-Turing THESIS which says that every function whose values can be attained mechanically, by a fixed method are computable by a Turing Machine.
This thesis which is currently thought of as a conjecture as opposed to a definition is the starting point for the entire conception of Artificial Intelligence. If it is true electronic computers MUST be able to possess at least human level intelligence. This is because cognitive science tells us that any cognitive process no matter how complex is eventually split into primitive computations or routines that are mechanical in their nature. Thus anything a brain can do a Turing machine or computer could in theory do at least as effectively.
This thesis was the reason behind large amounts of financial investment in Artificial intelligence. Computer science truly believed that such a target was attainable. Machines slowly surfaced who, it was claimed, could understand stories, and over time they did so with increasing ability.
It would be prudent of me to introduce to you a very important distinction, between two conceptions. Although these had always been present in discussion of AI, it was John Searle who was the first to define them adequately. These distinctions are not approaches to AI but conceptions of its aims.
The first, Strong AI sees AI as an attempt to design and build a machine that displays a range of psychological attributes, things such as reason or intentionality, even consciousness and eventually feeling and emotion. Someone who subscribes to strong AI must also agree to these propositions says Searle:
“a) An appropriately programmed computer really would have, or be, a mind in the same sense that you or I have.�
And,
“b) Its following the programs in question would explain its ability to do the psychological things it does.�
Something that is usually associated with, although not always defined as part of Strong AI, is the belief in the Turing test. This is a test Turing proposed that is essentially borne out of behaviourism, due to Descartes, cogito ergo sum it is only through empirical means that I can suggest, not know you are intelligent. It is merely because you exhibit similar behaviour to me that I conclude you are also intelligent. Turing says that if an invigilator cannot discriminate between the responses of a machine and those of a human then it should be attributed with intelligence.
The opposite of this is logically named Weak A.I. and it says simply that electronic digital computers are powerful instruments, for helping us to model, and thereby understand, the mind.
These distinctions are often used whilst committing a straw man fallacy, the term strong AI has become somewhat derogatory partly due to the effect of Searle’s arguments and also due to the claims made by early subscribers to Strong AI. In 1957 Simon and Newell two advocates of these ideas of Strong AI and founding fathers of AI research, expressed this view.
‘There are now in the world machines that think, that learn and that create. Moreover their ability to do these things is going to rapidly increase until… they can handle a range of problems coextensive to those to which the human mind has been applied. ‘
An example of such a program is one of the story understanding programs I previously described. These machines could take a simple story such as, to use Searle’s example:
‘A man went into a restaurant and ordered a hamburger. When the hamburger arrived it was burned to a crisp, and the man stormed out of the restaurant angrily, without paying for the hamburger or leaving a tip.’
When asked "Did the man eat the hamburger?" the machine would be able to answer, "No, he did not.� It could do this with a range of different although fairly similar stories and a range although again fairly similar questions and answers.
It was exactly the claim that this constituted understanding that Searle originally set out to refute, to do so he created this argument:
1. Programs a purely syntactical (formal)
2. Minds (human ones at least) have semantics, mental contents.
3. Syntax by itself is neither the same as or sufficient for semantic content.
Therefore:
4. Programs by themselves are neither constitutive of nor sufficient for minds.
Although he has rephrased this same argument in countless volumes the basic structure remains the same. The following thought experiment the ‘Chinese Room Argument’ was or at least Searle would have it that this argument was intended to illustrate the truth value of the third premise. Many people see it as a reformulation rather than the underlying argument of the thought experiment however in truth this is unimportant.
‘Suppose that I’m locked in a room and given a large batch of Chinese writing. Suppose furthermore (as is indeed the case) that I know no Chinese, either written or spoken, and that I’m not even confident that I could recognize Chinese writing as Chinese writing distinct from, say, Japanese writing or meaningless squiggles. To me, Chinese writing is just so many meaningless squiggles. Now suppose further that after this first batch of Chinese writing I am given a second batch of Chinese script together with a set of rules for correlating the second batch with the first batch. The rules are in English, and I understand these rules as well as any other native speaker of English. They enable me to correlate one set of formal symbols with another set of formal symbols, and all that "formal" means here is that I can identify the symbols entirely by their shapes. Now suppose also that I am given a third batch of Chinese symbols together with some instructions, again in English, that enable me to correlate elements of this third batch with the first two batches, and these rules instruct me how to give back certain Chinese symbols with certain sorts of shapes in response to certain sorts of shapes given me in the third batch. Unknown to me, the people who are giving me all of these symbols call the first batch a "script," they call the second batch a "story," and they call the third batch "questions." Furthermore, they call the symbols I give them back in response to the third batch "answers to the questions," and the set of rules in English that they gave me, they call the "program." Now just to complicate the story a little, imagine that these people also give me stories in English, which I understand, and they then ask me questions in English about these stories, and I give them back answers in English. Suppose also that after a while I get so good at following the instructions for manipulating the Chinese symbols and the programmers get so good at writing the programs that from the external point of view—that is, from tile point of view of somebody outside the room in which I am locked—my answers to the questions are absolutely indistinguishable from those of native Chinese speakers. Nobody just looking at my answers can tell that I don’t speak a word of Chinese. Let us also suppose that my answers to the English questions are, as they no doubt would be, indistinguishable from those of other native English speakers, for the simple reason that I am a native English speaker. From the external point of view—from the point of view of someone reading my "answers"—the answers to the Chinese questions and the English questions are equally good. But in the Chinese case, unlike the English case, I produce the answers by manipulating uninterpreted formal symbols. As far as the Chinese is concerned, I simply behave like a computer; I perform computational operations on formally specified elements. For the purposes of the Chinese, I am simply an instantiation of the computer program.’
So what has Searle achieved in this experiment? Well he has illustrated a machine that possesses no understanding whatsoever. However it appears it could pass the Turing test as described earlier. In effect it could fake being a Chinese speaker. Secondly that contrary to the belief of Strong A.I. a formal program operating as Turing originally conceived, one that simply manipulates symbols could never constitute a mind. This argument forms the most damning critique of Strong A.I. Calling into question one of its most founding parts the Church-Turing thesis described earlier. Mathematical criticisms of this idea have already been put forward establishing the possibility that there exists within the brain non computable neural processes.
There are many oppositions to this second part of the Chinese Room, a typical argument that although one program as illustrated above does not understand Chinese, a compliment or group of them – an entire system would, essentially he has committed a part whole fallacy. However clearly if this program was able to give out purely syntactic responses without ever engaging in semantics, then any number of further similar programs grouped with the first could take these syntactical responses and further operate them, none of them gleaning any kind of semantic meaning. It would be just a more complex yet equally ineffective program. I must concede that I have not yet seen an effective denial of this second half of the Chinese room argument.
I do however have questions relating to the denial of the Turing test. I feel there is one underlying assumption in Searle’s argument which no one has ever truly exploited. The reason for this is perhaps because if supporters of Strong AI were to use it they would most likely be pointing out fatal flaws in their own thinking. This assumption is that the room in its current form would be able to pass a Turing test. If we were agreed upon the previously mentioned theory of non computable neural processes it would make this even less likely. As a conversation is not a series of unrelated questions answers and statements, but there is a correlation between the end of a conversation and the beginning. To illustrate this further,
Imagine overhearing this conversation between the tester T and the Intelligent Person P:
T: Hi there my name is James I would like to apply for a job.
C: Ok James just go through that blue door and take an application form.
T: Thanks, oh and also is there a toilet near here?
C: Yes its signposted from the same room you get an application form.
T; Ok thanks a lot for your help.
C: No problem. Oh and good luck with the job application.
This conversation flows from one question to the next and each part bears some relevance to earlier parts of the conversation. A program operating like the Chinese Room suggests would not exhibit such qualities although it may hold the conversation it would not assimilate earlier information it would sound more disjointed, perhaps like this.
T: Hi there my name is James id like to apply for a job.
C: Ok through the blue door and take an application form.
T: Thanks, oh and also is there a toilet near here?
C: Yes, through the blue door and its signposted.
T: Ok thanks a lot for your help.
C: No problem.
As you can see although this may hint at understanding it does not flow in the same manner, the computer has no recollection of describing the blue door because it was merely an automatic response manipulating the words to give an appropriate answer. Although this machine would be very good at mimicking intelligent behaviour it would, and this would become increasing clear over the extended test period, be unable to make references to information a human would take from the conversation because it would not understand it. It would not behave adequately to pass the Turing test. In effect it would treat a conversation as a series of statements rather than a conversation.
This is how I always understood a Chinese room program to operate, however, there is another way that such a device could be envisaged, which Ned Block first raised he called this machine a ‘blockhead’. Block said that there is a finite amount of conversations a computer could be involved in. if the test lasted 2 weeks there is an enormous, but not infinite number of possible conversations. If the program could automatically respond not only to each individual statement, but to the conversation as an entire entity it would be able to pass the test perfectly. In a Turing test this machine would appear to be intelligent. Thus despite the near impossibility that a block program could be built I must concede that the Turing test in its current form is not an adequate test. However I suggest it would be very obvious to anyone who looked internally at the machine and saw its program that they would know that it was not an intelligent device but merely droid.
To resolve this issue I would like to return briefly to the idea which arose earlier, that of Descartes, cogito ergo sum, and the simple fact I can never know that any of you are intelligent in the same way I am. I assume you are because the behaviour you exhibit is so similar to mine and when you react to situations you appear to do it in the same way as I do. Suppose for a moment however I had a serious doubt that you were intelligent, I truly believed that you merely appeared to be operating in a similar fashion to myself but you were not thinking at all. You were simply an enormous ‘Blockhead’ program. It would not be a considered a bad idea for me to compare myself to you, on a neurological level. I mentioned earlier how every function of the mind has been attributed to activity in some part of the brain. With an MRI scan I could see that not only did you externally behave in a similar manner to myself, but internally as well.
Such lengths would perhaps be a little bit over the top, it is more likely I would just be placed in some kind of institution, however if computer science was claiming to have created an artificial intelligence I don’t believe examining its behaviour to such an extent would be considered over the top in the least. I would suggest therefore that the Chinese Room argument forms part of a two pronged attack. First it suggests that syntax is not sufficient for semantics, this part of the argument I believe to be sound, and I would not dispute it. I would therefore deny Strong AI under perhaps its original definition, and I would agree that true intelligence could not be borne out of pure syntax, or symbol manipulation. The second attack that the Chinese room argument makes is a refutation of the Turing test. If we were to take the original definition of behaviourism I would also agree with this first part, however I do not feel this is the only way in which we could define such an idea of behaviourism.
All mental phenomena however we choose to refer to them have some impact on the physical world. I wish to suggest that this impact should also be referred to as behaviour. Not simply the obvious behaviours perceivable solely through our senses, but those which we can perceive with the aid of tools, the way in which the program works is just as important and just as much a component of its behaviour as the way it outwardly appears to work. Therefore I would suggest that the Chinese room argument does not completely deny a Turing test and this part of Strong AI if no other remains intact.