The Turing Test: Affirming the Consequent?

[polytropos]

Affirming the consequent is a formal fallacy of making an argument of the form:
(P1) If P, then Q.
(P2) Q.
(C) P.

On the Turing test, Wikipedia says:

The test was introduced by Alan Turing in his 1950 paper “Computing Machinery and Intelligence,” which opens with the words: “I propose to consider the question, ‘Can machines think?’” Because “thinking” is difficult to define, Turing chooses to “replace the question by another, which is closely related to it and is expressed in relatively unambiguous words.”[3] Turing’s new question is: “Are there imaginable digital computers which would do well in the imitation game?”[4] This question, Turing believed, is one that can actually be answered. In the remainder of the paper, he argued against all the major objections to the proposition that “machines can think”.[5]

In other terms, thinking is a subjective process that cannot be observed. The Turing test aims to avoid this difficulty by determining not whether machines do think (for that cannot be observed directly) but rather whether machines can act as though they think.

I allow that the Turing test seems useful to define or determine a criteria for what it would mean for a machine to “think.” If a machine passes the Turing test, then we can call it “artificially intelligent.”

The fallacy crops up when one tries to use the Turing test to determine that machine’s behaving like us implies that the machine thinks like us. We have an argument of the form:

(P1) If X thinks, then X can behave as though it thinks.
(P2) X behaves as though it thinks.
(C) X thinks.

(P1) seems reasonable given our experience. We think, and our behavior determines what it would mean for something to behave as though it thinks. Likewise, we can grant (P2), that a machine capable of behaving as though it thinks is conceivable. But the issue is that this line of reasoning is the straightforward fallacy of affirming the consequent.

Notice that this is a fallacy regardless of the definition of thinking - so we need not be hung up by the definition. If we define thinking in terms of function and behavior, then we trivially get:

(P1’) If X behaves as though it thinks, then X can behave as though it thinks.
(P2) X behaves as though it thinks.
(C’) X behaves as though it thinks.

(P1’) has been transformed into a tautology by this move, and (C’) is now trivial.

 

[Bagheera]

polytropos:

The fallacy crops up when one tries to use the Turing test to determine that machine’s behaving like us implies that the machine thinks like us.

What do you mean by “thinks like us”? Uses the same process? No one says that.

We don’t say that the internal process is the same. You can reach the same result from the same premises by using different methodology. So your title is incorrect; there is no fallacy.

 

[tonyrey]

What do you mean by “thinks like us”? Uses the same process? No one says that.

We don’t say that the internal process is the same. You can reach the same result from the same premises by using different methodology. So your title is incorrect; there is no fallacy.

The expression “different methodology” is too nebulous to constitute a valid objection. Precisely what does it entail? For a start, can mechanistic answers entirely replace conclusions based on insight?

 

[polytropos]

What do you mean by “thinks like us”? Uses the same process? No one says that.

By “thinks like us,” I mean has what we (most non-eliminativists) would generally consider genuine thoughts: my reading your post, considering a response, doing a sum in my head, remembering my mother, etc. The precise definition, as I’ve said, does not matter (cf. the caveat about trivially defining “thinking” at the end of the OP); the argument, for any P, Q, is either fallacious or trivial.

We don’t say that the internal process is the same.

I did not say that the internal process is the same, or that the Turing test purports to conclude that the internal process is the same. The point is just that the inference from consequent to antecedent is a fallacy.

One could argue that the first premise is meant to be “X thinks if and only if X behaves as though it thinks,” but there is not really any support for this, whether empirical or otherwise, and since it is indeed what the Turing test is trying to show, it seems to beg the question.

You can reach the same result from the same premises by using different methodology.

What do you mean here?

 

[Linusthe2nd]

Affirming the consequent is a formal fallacy of making an argument of the form:
(P1) If P, then Q.
(P2) Q.
(C) P.

On the Turing test, Wikipedia says:

In other terms, thinking is a subjective process that cannot be observed. The Turing test aims to avoid this difficulty by determining not whether machines do think (for that cannot be observed directly) but rather whether machines can act as though they think.

I allow that the Turing test seems useful to define or determine a criteria for what it would mean for a machine to “think.” If a machine passes the Turing test, then we can call it “artificially intelligent.”

The fallacy crops up when one tries to use the Turing test to determine that machine’s behaving like us implies that the machine thinks like us. We have an argument of the form:

(P1) If X thinks, then X can behave as though it thinks.
(P2) X behaves as though it thinks.
(C) X thinks.

(P1) seems reasonable given our experience. We think, and our behavior determines what it would mean for something to behave as though it thinks. Likewise, we can grant (P2), that a machine capable of behaving as though it thinks is conceivable. But the issue is that this line of reasoning is the straightforward fallacy of affirming the consequent.

Notice that this is a fallacy regardless of the definition of thinking - so we need not be hung up by the definition. If we define thinking in terms of function and behavior, then we trivially get:

(P1’) If X behaves as though it thinks, then X can behave as though it thinks.
(P2) X behaves as though it thinks.
(C’) X behaves as though it thinks.

(P1’) has been transformed into a tautology by this move, and (C’) is now trivial.

Sounds like the twin " Bs " to me.

Linus2nd

 

[Gorgias]

The fallacy crops up when one tries to use the Turing test to determine that machine’s behaving like us implies that the machine thinks like us. We have an argument of the form:

(P1) If X thinks, then X can behave as though it thinks.
(P2) X behaves as though it thinks.
(C) X thinks.

But, was this what Turing was really trying to do? Was he trying to answer the question “can a machine think?”…? It would seem that this is not the case; in fact, it would seem that Turing wanted to provide an alternative question: ‘can a machine do what thinking beings do?’. (Alternately, we could consider that he might have wanted to provide a different goal for AI than ‘machines that think’.)

With this in mind, I would have to answer that this is not a case of ‘affirming the consequent’:

(1) If Z exhibits certain behaviors which typically are indicators of thought, then Z can be considered to have been thinking.
(2) W exhibits certain behaviors which typically are indicators of thought.
(c) W can be considered to have been thinking.

Searle’s Chinese Room experiment blows a hole through this notion, but on the level of strict behavioral observation, Turing’s suggestion is intriguing…!

 

[polytropos]

But, was this what Turing was really trying to do? Was he trying to answer the question “can a machine think?”…? It would seem that this is not the case; in fact, it would seem that Turing wanted to provide an alternative question: ‘can a machine do what thinking beings do?’. (Alternately, we could consider that he might have wanted to provide a different goal for AI than ‘machines that think’.)

Yes, you’re right. I should have been clearer that this is not an error which is being attributed to Turing. As I said, the test is fine if it is taken to provide a definition for artificial intelligence.

With this in mind, I would have to answer that this is not a case of ‘affirming the consequent’:

(1) If Z exhibits certain behaviors which typically are indicators of thought, then Z can be considered to have been thinking.
(2) W exhibits certain behaviors which typically are indicators of thought.
(c) W can be considered to have been thinking.

Searle’s Chinese Room experiment blows a hole through this notion, but on the level of strict behavioral observation, Turing’s suggestion is intriguing…!

Right, the argument given here, since (1) is swapped around, does not affirm the consequent. I agree that Searle’s Chinese room calls (1) into question.

 

[Bagheera]

polytropos:

By “thinks like us,” I mean has what we (most non-eliminativists) would generally consider genuine thoughts: my reading your post, considering a response, doing a sum in my head, remembering my mother, etc.

In other words: accepting (name removed by moderator)ut, processing/manipulating data, creating an output, performing calculations, recalling previous (name removed by moderator)uts, etc… In simple words: “processing information”. Which one of these activities is the one that the “machine” is unable to perform?

polytropos:

The point is just that the inference from consequent to antecedent is a fallacy.

Yes, it would be – if anyone would do that.

polytropos:

What do you mean here?

I mean that the activities you described above can be performed several different ways. The human mind and the computer’s chips may be processing the information differently, but as long as they reach the same result the ways and means do not matter. You can climb a mountain using you’re your arms and legs, or hop on the cable car and take the ride. In both cases you will get to the top of the mountain.

polytropos:

As I said, the test is fine if it is taken to provide a definition for artificial intelligence.

That is a bit sloppy. Maybe the test shows or proves or substantiates that the machine has “artificial intelligence”. But first you need to define what “artificial intelligence” IS. You cannot use a circular definition that the machine exhibits artificial intelligence if you define artificial intelligence as something that the machines exhibit. What is the definition of artificial intelligence, and how does it differ from “real” intelligence?

The question is: “if a machine acts as if it were thinking (processing information), then on what grounds should we doubt this?”. The duck principle comes to mind: “If it looks like a duck, walks like a duck, quacks like a duck and tastes like a duck, then it is rational to assume that it IS a duck”. Observe, we do not say explicitly that it MUST be a duck and cannot be anything else, we just say that it is very likely that it is a duck.

Therefore your fallacy is inapplicable to this question.

 

[inocente]

The fallacy crops up when one tries to use the Turing test to determine that machine’s behaving like us implies that the machine thinks like us.

Turing’s original 1950 paper is online here (with some uncorrected OCR errors) - loebner.net/Prizef/TuringArticle.html

Following on from Gorgias’ post, his paper discusses a game (now called his test) which he invented to avoid defining what is meant by “think” and “machine”, as he knew he’d never get agreement - he says “The original question, “Can machines think?” I believe to be too meaningless to deserve discussion.”

His paper includes discussion of objections he identified, including what he calls the “theological objection”.

 

[polytropos]

In other words: accepting (name removed by moderator)ut, processing/manipulating data, creating an output, performing calculations, recalling previous (name removed by moderator)uts, etc… In simple words: “processing information”. Which one of these activities is the one that the “machine” is unable to perform?

No, I am referring to the internal state of representing something abstract to myself. Whether there was an (name removed by moderator)ut/output is irrelevant to the examples I gave (and the calculations need not be done algorithmically, as a computer must do them).

I mean that the activities you described above can be performed several different ways. The human mind and the computer’s chips may be processing the information differently, but as long as they reach the same result the ways and means do not matter. You can climb a mountain using you’re your arms and legs, or hop on the cable car and take the ride. In both cases you will get to the top of the mountain.

I agree that a man and a machine can take different routes to calculate different things. That’s fine and good. But it’s precisely my point in this topic that the machine does not thereby think. It’s not just that what’s going on inside my brain is different from what’s going on in the computer (I don’t care about that). It’s the fact that the computer does not have intentional/qualitative states, which are necessary for “thinking” - even if it gets the right outputs. My argument is that the inference from the outputs to the intentional and states/thinking is invalid.

Let’s say that thinking is the activity we’re talking about. We have no interest in any output. There is no question of functional equivalence, since there’s no function for the man or computer to perform. Surely you have just sat down and reflected sometimes, without any particular. But then we run into a wall if we try to determine thinking in terms of function.

That is a bit sloppy. Maybe the test shows or proves or substantiates that the machine has “artificial intelligence”. But first you need to define what “artificial intelligence” IS. You cannot use a circular definition that the machine exhibits artificial intelligence if you define artificial intelligence as something that the machines exhibit. What is the definition of artificial intelligence, and how does it differ from “real” intelligence?

Actually, no, this seems to be precisely the point, and the main sense in which Turing’s test can be construed as non-fallacious. Turing meant to redirect the field of artificial intelligence. Artificial means apparent, or simulated, so artificial intelligence is just that which looks like intelligence (what we have). The test, as I’ve said, is fine if it provides a standard for artificial intelligence, or if you say, “The goal of computer science is to create a machine that can pass the Turing test; we will call such a machine artificially intelligent.”

The difference simply is that artificial intelligence, because it is artificial (coming from artifice), is just meant to look like intelligence, so it is defined by its outputs. But as intelligent beings ourselves, we are painfully aware of the facts that a). intelligence is predominantly internal and b). it’s impossible to peek into another subject. Artificial intelligence, then, is as good as a machine can get.

The question is: “if a machine acts as if it were thinking (processing information), then on what grounds should we doubt this?”. The duck principle comes to mind: “If it looks like a duck, walks like a duck, quacks like a duck and tastes like a duck, then it is rational to assume that it IS a duck”. Observe, we do not say explicitly that it MUST be a duck and cannot be anything else, we just say that it is very likely that it is a duck.

I think you are missing my point. What is in question is precisely whether it does “look like a duck, walk like a duck, quack like a duck, and taste like a duck.” If “thinking” for us requires intentional and qualitative states, then until we know that machine has those intentional and qualitative states, it doesn’t meet the criteria of “looks like a duck.” Only by stipulating that thinking doesn’t require those states can you get a machine to pass the test - but then, the stipulation has weakened and trivialized the notion of thinking to preclude genuine activities that we ourselves perform.

 

[polytropos]

Following on from Gorgias’ post, his paper discusses a game (now called his test) which he invented to avoid defining what is meant by “think” and “machine”, as he knew he’d never get agreement - he says “The original question, “Can machines think?” I believe to be too meaningless to deserve discussion.”

Yes, and I agree with Gorgias. If the test is taken to define what it means for a machine to “think,” then its application commits no fallacy. But defining something in one way and applying it to something else (ie. what we mean by saying that we think, referring to an intentional, internal, representational process) would still be a fallacy, with a bit of sleight of hand.

His paper includes discussion of objections he identified, including what he calls the “theological objection”.

He does not really seem to answer any real objection in that section. He talks about how theists would seem to be restricting God’s omnipotence by saying that a machine could not think. But that is, of course, irrelevant, since if God were to confer intelligence upon a machine (or an elephant, to take his example), then it would still not be the physical facts about the machine that make it intelligent.

 

[Bagheera]

Hold it, please! I asked you what do you consider “thinking” and your reply was:

By “thinks like us,” I mean has what we (most non-eliminativists) would generally consider genuine thoughts: my reading your post, considering a response, doing a sum in my head, remembering my mother, etc.

Which I translated into more general concepts:

In other words: accepting (name removed by moderator)ut, processing/manipulating data, creating an output, performing calculations, recalling previous (name removed by moderator)uts, etc… In simple words: “processing information”.

So, now you say:

polytropos:

No, I am referring to the internal state of representing something abstract to myself.

In other words a state of affairs within your brain which corresponds to some abstraction. Is this something that a “machine” cannot do?

polytropos:

Artificial means apparent, or simulated, so artificial intelligence is just that which looks like intelligence (what we have).

That is nice, but vague to the point of meaningless – unless you can define what “real” intelligence is and point out the difference between the “real” and the “simulation”. A difference – mind you – that can be detected by some epistemological method.

You see we are hindered by the lack of two proper definitions:

1. What is thinking?
2. What is intelligence?

Without them we are only “peeling balloons”. And you know what we get when the skin of a balloon is peeled away… we get nothing. By the way, we already agreed that the Turin test does not involve the fallacy of “affirming the antecedent”, so a proper definition of “thinking” and “intelligence” is a must.

 

[Al_Moritz]

Recently I heard an interesting story on NPR in their science hour. There is this annual meeting on artificial intelligence and the Turing test, where they have those judges in a blind test who have to decide if they are having a conversation with a human person or a computer. In 2008 or 2009 for the first time 25 % of the judges where fooled by a computer, dangerously close to the threshold of 30 % that Turing had proposed (the discussion on NPR then also was, why not 50 or 51 % ?).

Yet then they told the story of that one computer that had been learning for many years; with the arrival of the internet it then could have conversations with millions of people instead of just a few. During that process it continuously learned, and became better and better, until one day, after that 25 % threshold had been crossed in that competition, one judge found out how to unmask it as just a machine instead of a human person. He said something outrageous and absurd, like: “Last night I did not sleep well because a meteor crashed into my bedroom.” The computer than replied: “I woke up at 1 in the afternoon.”

The computer obviously was thrown off by the absurd and unexpected remark – it lay outside of what it had encountered and learned during those millions of other conversations – and it could not reply like a human being would. Why? The incident exposed that while a computer can learn and learn how to react – after all, a computer is really good at computing – it does not have the mental judgment needed to reply to an absurd remark in an appropriate manner. In other words, it cannot think.

After this judge had learned to ‘crack the code’ and informed his fellow judges, the next year none of the judges were fooled on the competition, not a single one. So much for the Turing test and ‘thinking machines’.

 

[Bagheera]

Al Moritz:

Recently I heard an interesting story on NPR in their science hour. There is this annual meeting on artificial intelligence and the Turing test, where they have those judges in a blind test who have to decide if they are having a conversation with a human person or a computer. In 2008 or 2009 for the first time 25 % of the judges where fooled by a computer, dangerously close to the threshold of 30 % that Turing had proposed (the discussion on NPR then also was, why not 50 or 51 % ?).

Yet then they told the story of that one computer that had been learning for many years; with the arrival of the internet it then could have conversations with millions of people instead of just a few. During that process it continuously learned, and became better and better, until one day, after that 25 % threshold had been crossed in that competition, one judge found out how to unmask it as just a machine instead of a human person. He said something outrageous and absurd, like: “Last night I did not sleep well because a meteor crashed into my bedroom.” The computer than replied: “I woke up at 1 in the afternoon.”

The computer obviously was thrown off by the absurd and unexpected remark – it lay outside of what it had encountered and learned during those millions of other conversations – and it could not reply like a human being would. Why? The incident exposed that while a computer can learn and learn how to react – after all, a computer is really good at computing – it does not have the mental judgment needed to reply to an absurd remark in an appropriate manner. In other words, it cannot think.

After this judge had learned to ‘crack the code’ and informed his fellow judges, the next year none of the judges were fooled on the competition, not a single one. So much for the Turing test and ‘thinking machines’.

Very good story. But I would amend the last sentence of yours: “That particular” machine did not think. It stands to reason that non-sequitur statements are probably the best way to “unmask” a machine. Especially jokes! And among them the “puns”! The trouble is that many people have no sense of humor. And of those who do, many (or most) cannot explain why they find a particular joke funny.

Looking for “hidden” meanings and especially introducing a fresh, unexpected meaning, like in puns will be a very hard nut to crack – as long as the machine will not really “understand”. And that presents the question: “what is understanding”?

On a lighter note, here comes a joke, which many people might not even understand (if their English is poor) and of those who do many are outraged. The funny thing is that the words are perfectly “clean” and “wholesome”, and if one finds the joke in very bad taste… it is his “unclean” mind that produces the ill effect. So the joke from the Clinton presidency era:

Be warned! Do not continue, if you are squeamish. Highlight the text if you are interested.

At the beginning of the baseball season, Clinton and Hillary wait for the first game of the season to start. One of the organizers walks up to Clinton and whispers something in his ear. Clinton jumps up, grabs Hillary and throws her way out into the field. The organizer turns white and say: “No, Mr. President! The first pitch!”

Did you find it funny?

 

[tonyrey]

Recently I heard an interesting story on NPR in their science hour. There is this annual meeting on artificial intelligence and the Turing test, where they have those judges in a blind test who have to decide if they are having a conversation with a human person or a computer. In 2008 or 2009 for the first time 25 % of the judges where fooled by a computer, dangerously close to the threshold of 30 % that Turing had proposed (the discussion on NPR then also was, why not 50 or 51 % ?).

Yet then they told the story of that one computer that had been learning for many years; with the arrival of the internet it then could have conversations with millions of people instead of just a few. During that process it continuously learned, and became better and better, until one day, after that 25 % threshold had been crossed in that competition, one judge found out how to unmask it as just a machine instead of a human person. He said something outrageous and absurd, like: “Last night I did not sleep well because a meteor crashed into my bedroom.” The computer than replied: “I woke up at 1 in the afternoon.”

The computer obviously was thrown off by the absurd and unexpected remark – it lay outside of what it had encountered and learned during those millions of other conversations – and it could not reply like a human being would. Why? The incident exposed that while a computer can learn and learn how to react – after all, a computer is really good at computing – it does not have the mental judgment needed to reply to an absurd remark in an appropriate manner. In other words, it cannot think.

After this judge had learned to ‘crack the code’ and informed his fellow judges, the next year none of the judges were fooled on the competition, not a single one. So much for the Turing test and ‘thinking machines’.

The proof of the pudding is how the materialist behaves. Like a machine?

Mechanical conclusions lead to conviction but it is criminal rather than intellectual…

 

[inocente]

He does not really seem to answer any real objection in that section. He talks about how theists would seem to be restricting God’s omnipotence by saying that a machine could not think. But that is, of course, irrelevant, since if God were to confer intelligence upon a machine (or an elephant, to take his example), then it would still not be the physical facts about the machine that make it intelligent.

I think that’s his whole point.

A dualist would argue that intelligence/thinking is a function of an immaterial soul. A baby is produced by its parents but her soul can only be created and conferred by God.

OK, Turing says to the dualist, a machine is produced by engineers but its soul can only be created and conferred by God. And if God is omnipotent, there’s no logical reason why God cannot do exactly that.

Turing uses the dualist’s own belief in the immaterial to prove that it is no barrier to the possibility of thinking machines.

 

[polytropos]

I think that’s his whole point.

A dualist would argue that intelligence/thinking is a function of an immaterial soul. A baby is produced by its parents but her soul can only be created and conferred by God.

OK, Turing says to the dualist, a machine is produced by engineers but its soul can only be created and conferred by God. And if God is omnipotent, there’s no logical reason why God cannot do exactly that.

Turing uses the dualist’s own belief in the immaterial to prove that it is no barrier to the possibility of thinking machines.

There are still issues with this line of argument.

1. You don’t need to be a dualist or a theist to doubt that the Turing test can deliver actually intelligent machines (cf. Searle, Nagel, Tallis, Dreyfus). This is because the inference from behavior to intelligence is a fallacy regardless of your philosophical commitments.
2. This is why it was irrelevant to bring up the “theological objection” in the first place - I’m not appealing to theology, nor arguing for dualism, ITT.
3. I don’t think theists would mind conceding that God alone could make machines intelligent. That would kinda vindicate them.
4. I do think that functionalists would mind conceding that God alone could make machines intelligent.
5. The fact that it might be possible for God to make intelligent machines does not demonstrate that God actually does cooperate with computer scientists in the way he does with biological parents.
6. We have thoughts, and we infer that other things of the same kind also have thoughts. But machines aren’t the same kind, so there does not seem to be any warrant to the inference that machines are ensouled like we are.

 

[polytropos]

In other words a state of affairs within your brain which corresponds to some abstraction. Is this something that a “machine” cannot do?

If we were to define it as a “state of affairs within your brain,” then we would be begging the question in favor of materialism. But yes, thinking can be abstract, and that is part of what a machine cannot capture.

Suppose, for example, that you tell someone, “I was thinking about you last night.” What were you doing? This is obviously a case where external behavior is irrelevant. Of what does “thinking” here consist? I was representing the person in question to my own subject, apart from her presence in the actual world. (Perhaps I’m thinking about a particular quality - kindness - of hers.) There seems to be a gulf between what we do in such a case and what a machine that passes the Turing test could do.

For example, can the machine represent the person to itself and consider some property of her in the abstract? The Turing test doesn’t seem to tell us; the Turing test tells us if the machine can parse questions and formulate semantically correct responses. Whether it can “think” in the way we mean when we say “I was thinking about you last night” is extraneous to the Turing test.

That is nice, but vague to the point of meaningless – unless you can define what “real” intelligence is and point out the difference between the “real” and the “simulation”. A difference – mind you – that can be detected by some epistemological method.

I think you’ve got the burden of proof backwards here. I think intentionally and qualitatively. These are well-defined terms in philosophy of mind that refer to qualities of our first-person experience that have thus far been incommensurate with materialism. What I’m pointing out is that the results of the Turing test tell us nothing about whether a machine does these things. I am not the one who needs a better epistemological method.

I have a first-person experience is a necessary condition of what I call intelligence (it seems pointless to squabble over the whole definition). If you take a machine that, to our knowledge, does not have a first-person experience, then the burden of proof is on you to show that it does or can in principle. The Turing test doesn’t do this.

By the way, we already agreed that the Turin test does not involve the fallacy of “affirming the antecedent”, so a proper definition of “thinking” and “intelligence” is a must.

Actually, I don’t think we did agree on that. If you want to employ the Turing test to show that machines “think like us” (whatever you take that to mean), then you have two options: affirm the consequent, or define thinking trivially and minimally in terms of behavior and function.

 

[inocente]

There are still issues with this line of argument.

1. You don’t need to be a dualist or a theist to doubt that the Turing test can deliver actually intelligent machines (cf. Searle, Nagel, Tallis, Dreyfus). This is because the inference from behavior to intelligence is a fallacy regardless of your philosophical commitments.
2. This is why it was irrelevant to bring up the “theological objection” in the first place - I’m not appealing to theology, nor arguing for dualism, ITT.
3. I don’t think theists would mind conceding that God alone could make machines intelligent. That would kinda vindicate them.
4. I do think that functionalists would mind conceding that God alone could make machines intelligent.
5. The fact that it might be possible for God to make intelligent machines does not demonstrate that God actually does cooperate with computer scientists in the way he does with biological parents.
6. We have thoughts, and we infer that other things of the same kind also have thoughts. But machines aren’t the same kind, so there does not seem to be any warrant to the inference that machines are ensouled like we are.

1a The “theological objection” is only one of nine objections which Turing considers.

1b His test isn’t intended to “deliver actually intelligent machines”, it’s intended to decide whether a machine can think without us getting bogged down trying to define “thinking” and “machine”.

1c Surely behavior is the only objective measure of intelligence (IQ etc.)?

2 Your point 6 is exactly why it was relevant.

3 A dualist might argue that only God can ensoul, but why would a theist argue that only God can make a machine intelligent?

4, 5 Turing is only dealing with dualists at that point.

6 I think you need to be careful here about the ethics. In what sense are machines not the same kind as us? They are physical, they are made of the same stuff, so at what level of detail do you draw the distinction of kind? There have been those who believed women are not the same kind as men, or black skins not the same kind as white. If space aliens land and have Star Trek matter transporters, mathematics way beyond ours, stunning literature and so on, doesn’t that indicate they’re intelligent, not whether they’re silicon-based seven-legged cyborgs?

 

[Bagheera]

polytropos:

Actually, I don’t think we did agree on that. If you want to employ the Turing test to show that machines “think like us” (whatever you take that to mean), then you have two options: affirm the consequent, or define thinking trivially and minimally in terms of behavior and function.

The fallacy would be stating that the “human like responses definitely proves that the machine thinks”. That is not what we say. We say that the “human like responses make the assumption that the machine thinks a likely one”. So the “fallacy” is out the window.

polytropos:

I think intentionally and qualitatively.

So you say. Do all people always think “intentionally” and “qualitatively”? Or are there exceptions?

polytropos:

These are well-defined terms in philosophy of mind…

A metaphysical statement is vacuous if there is no accompanying epistemology. Why is there no “well-defined” term for thinking? And intelligence?

Until you can give a definition about what “thinking” means, what is “intelligence”, there is nothing to talk about. If you cannot p(name removed by moderator)oint what the machine is not supposed to be able to do, how could be even start to consider if it can or cannot do it? Can toddlers think? The certainly cannot think in abstractions. Can autistic people think? They cannot think in abstractions either. When an autistic person hears the word “dog”, she will recall all the dogs she ever encountered in her life, and not some “abstract” dog. No wonder that Turing did not want to get “bogged down” in the problem of “thinking”.

polytropos:

I have a first-person experience is a necessary condition of what I call intelligence (it seems pointless to squabble over the whole definition).

Unfortunately I do not understand the first part… and I disagree that providing a full definition is “pointless squabbling”.

polytropos:

We have thoughts, and we infer that other things of the same kind also have thoughts. But machines aren’t the same kind, so there does not seem to be any warrant to the inference that machines are ensouled like we are.

Oh, the infamous “kind”. Would be nice to see a coherent definition of this elusive term, too. And “ensoulment”? The church is reluctant (read: unable) to tell when the “ensoulment” is supposed to take place… By the way, would a fully biological android be considered a different “kind”, too?