The Turing Test: Affirming the Consequent?

[MPat]

In an episode of the Simpsons everyone in the town, including the TV stations, police and law court, sting Homer to teach him a lesson.

Probably anyone can be conned about anything if you throw enough money and resource at the deception. All a posteriori knowledge would fall victim though, not just this, we wouldn’t be able to trust anything, it’s the road to paranoia.

(But in this case, a machine clever enough to con us would, kind of by definition, have to be intelligent enough to con us.)

But is it a machine that “cons us”? Or is it its maker?

If the machine had decided to pass the test all by itself, then we might discuss, if it is “intelligent” or “rational”. Actually, we could do this even if it would fail the test afterwards (as you can see, the test doesn’t really add much).

But in reality the first machine to pass the Turing test is going to be made to pass that test. Its creators will look for ways in which the “experts” could distinguish it from a human and try to make them fail. So, the makers of the machine will fool the experts, not the machine itself. And yes, they will be intelligent enough to be considered humans.

You seem to be suggesting that we use looser rules of evidence to judge humans than machines, that a human is innocent until proven guilty while a machine is guilty until proven innocent. I think the ethics are very dangerous, since it hinges on how you decide who is human and who is not, and we know all too well where that has led in the past.

How did you reach such conclusions?! I didn’t speak about any “guilt” or “innocence” nor about any “ethics”! And the “rules of evidence” are the same: if we know that something has been specifically made to fool us, we take that into account.

And if you think that defining “human” as “someone who passes the Turing test” can prevent the problems with deciding “who is human and who is not” - I think you are wrong. Just think - could it prevent Nazis from concluding that Jews or (perhaps especially) the ones with psychiatric diseases are “not humans”…? Can all humans pass the “Turing test” all the time? Not to mention that the Turing test is simply impossible unless we start with at least a good idea who is a human and who is not.

 

[polytropos]

The question again is: “is the difference significant?”. There can be two different scenarios about a proposition: 1) we do not know right now if the proposition is true or not, but we can investigate and find out – eventually; and 2) it cannot be known in principle if the proposition is true or not. The first kind merits attention, and while the investigation is underway the label we “stick” on that proposition is neither true, nor false rather “undecided”. The second kind does not merit attention, even though it may not be syntactically “meaningless”.

The difference, indeed, is significant. When I speak, my words mean something (necessarily, I think it is reasonable to grant). If I speak gibberish, my words don’t mean anything. When a machine speaks gibberish, its words don’t mean anything. The question is: when a machine speaks human language, does its words mean anything (necessarily, as would be necessary for the Turing test to produce usable results - if it can’t distinguish, then it fails)? If they don’t, then there is a clear (and substantial) disanalogy between our speech and robot speech, even if the sounds are the same.

But is the saying of the words sufficient for the words having meaning? That is doubtful, for then any gibberish of sufficient complexity would have meaning in some constructable language system. In the case of human language, the connection remains to be shown that for (name removed by moderator)uts to be mapped to outputs, the outputs must have syntactical meaning.

(Another consideration, kind of a twist on the Chinese rooom: say a man remotely operates a machine. So the machine passes the Turing test. What, in this case, is intelligent? Is it the man + machine, or is it just the man? I feel like one would be hard pressed to say that it is the system of the man + the machine which is intelligent, for in such a case, the machine’s computational power could be pretty slim.)

We need

1. a rigorous definition of “thinking” and
2. an epistemological method to decide if entity “X” is thinking or not, and
3. a definition of the “meaning”.

The demand for a definition of thinking seems like a clear double standard to me. Turing does not offer a definition of thinking. The point of his experiment is, as has been insisted both by me and by his supporters in this thread, to avoid defining thinking. So then why are those disputing him required to offer what he does not?

As I’ve discussed in the past, good philosophy is not done in defining terms. Disputing definitions, unless they are truly and uncontroversially incoherent, gets no one anywhere, since two parties making incompatible assertions (ie. asserting two incompatible definitions) does not lead to resolution. What one can do is what I have done consistently throughout this thread (and previous threads on similar subjects): offer what I regard as necessary conditions for any substantive definition of thinking, and demonstrate why the Turing test does not allow us to conclude that a machine meets (or likely meets) such conditions. If one does not meet my necessary conditions, then that is fine, that accomplishes my goal: one would have weakened his definition of thinking too much for it to warrant comparison to humans.

I don’t think we need to get into technical aspects of semantics here. I think the notion of a proposition having semantic content is clear enough for our purposes, and it is certainly plausible to say that a sentence’s being uttered does not entail its being a meaningful proposition, however one defines “meaning.” (This is clear because a tape recorder’s utterances have no semantic content, except as heard by a language community. As such, we need to decide why the machine that can engage in lengthier conversations utters propositions with semantic content, if it does.)

 

[polytropos]

Ah, but the semantic content is not something that exists in an objective manner. Between the two of us, we can create an artificial language, which is perfectly meaningful for us, and sounds like a meaningless (or even misleading) gobbledygook for everyone else. There is no “abstract, objective” meaning to words and propositions. The meaning gets established in a communication channel between two entities, who “agree” on the meaning. I cannot emphasize enough that the “meaning” is the result of agreement.

I agree that there is not meaning intrinsic to vocalized or written sentences qua sound waves or qua ink on paper (or qua binary digits). But I’m not saying that there is. What is relevant is that our artificial language is meaningful to us, while the human language need not be meaningful to a machine that passes the Turing test (by need not, I mean that some computers which pass would not need to understand the meaning of the words that they are uttering, so even if some computers could utter propositions with meaning, the Turing test is an inadequate epistemological method for distinguishing them).

You say that “meaning” is the result of agreement. That may be so, in some sense. But the issue is that this is too weak to vindicate the Turing test. My TI-84 and I agree on the rules of mathematical syntax. Microsoft Sam and I agree on phonics of the English language. Two computers communicating in binary over a network (say, sending encrypted data for Facebook users) have agreements in their communication systems. But my calculator and Microsoft Sam and most computers do not understand the meaning of their “utterances,” which only make sense if seen/heard/used by humans. “Agreement” seems to refer to too broad of a set of phenomenon for it to be sufficient for a machine that passes the Turing test to be said to be intelligent. If the meaning of “agreement” is restricted, then it has not been established that it applies uniquely to, say, humans and machines that pass the Turing test, or that it is necessary that a machine that passes the Turing test has it.

The “torturing” aspect was only a convenient example to highlight the central problem: namely the question of emulation vs. the real thing.

The emulation does not have to be identical to the real stuff in EVERY respect. Just like in the Turing test, we do not care if the tested entity is “made of” electronic or biological parts. We only care if the verbal and comprehension skills are comparable to humans. And the comprehension can only be gauged indirectly – by evaluating the responses of the other party.

Well, I’ve given my response that judgments may be defeasible. If it fools me, then I’m fooled. That doesn’t mean I’m right.

What do you mean by “we do not care if the tested entity is ‘made of’ electronic or biological parts. We only care if the verbal and comprehension skills are comparable to humans”? Who is “we”? What does “care” mean here? As I’ve reiterated constantly, if we just care about creating machines that can act like humans (say, for commercial reasons), then have at it: that’s what the Turing test gets you. I suspect you mean something else by care.

If comprehension can only be gauged indirectly, and the indirectness of the epistemological method imposes intractable limits on what it can tell us, then so much the worse for the epistemological method (or so much the worse for the philosophical conclusions one would like to draw).

I definitely deny that the “verification principle” is applicable to ALL propositions (which was the assertion of the logical positivists). This principle is not applicable to the propositions in an abstract, axiomatic system (one cannot “verify” the axioms of mathematics and all the non-axiomatic propositions are true if and only if they are logical corollary of the axioms). It is also not applicable to the subjective propositions like “I prefer chocolate ice cream over the vanilla flavored one”. However, it is applicable to the statements pertaining the external, objective reality. If a proposition reflects the external reality, it is “true” proposition, otherwise it is not.

Actually the logical positivists did not apply their verification principle to all propositions. They, like you, would have excused propositions in axiomatic systems (ie. analytic truths, like 2+2=4 or There are no married bachelors). They also excluded subjective propositions. “The logical positivists’ initial stance was that a statement is “cognitively meaningful” only if some finite procedure conclusively determines its truth.[18] By this verifiability principle, only statements verifiable either by their analyticity or by empiricism were cognitively meaningful.” (Wikipedia)

Incidentally, Quine, one of the major figures who demolished logical positivism, was open in principle to rethinking the axioms of logic and other axiomatic systems.

 

[polytropos]

You’ll see you said “intelligence” originally, not “thinking”, and I replied to say that intelligence is not subjective. There are IQ tests and so on, it can be observed and measured.

I did not deny that I said “intelligence.” I said that “I mean that thinking is fundamentally subjective.” The word “mean” implied that, yes, I said intelligence, and I am now saying that I should have said thinking, because that is what I meant.

So if intelligence can be measured, and if as you say thinking is required for intelligence, we have now in principle cracked how to know if machines think - just measure their intelligence. Get your guys to contact my guys, they can hammer out the details.

But that won’t do. As I said, thinking is required for intelligence. Thinking is subjective.

There are other issues faced by the crude syllogism intelligence can be measured, thinking is necessary for intelligence, so we have now in principle cracked how to know if machines think. Measurement of intelligence presupposes the existence of intelligence. But no one would test the intelligence of a calculator, or even a powerful program, even though they could perform better than we could on some tests.

You know as well as me that many philosophers reject the term, it’s simply an unnecessary and misleading hypothesis.

(Philosophers reject qualia for various reasons. My reason is I think it’s sloppy and misleading to lump disparate subjective experiences together on such a tenuous basis).

Incidentally, rejecting it on that basis does not really solve any of the difficulties it proposes. The reason qualia are considered as a group is that each individual subjective experiences proposes the same problem that materialism has no answer for in principle. But if you like, divide them up (as I have in this conversation). The experience of seeing red is as intractable for materialism as all of qualia; materialists only “handle” it be ignoring it.

 

[polytropos]

Now suppose you and an alien machine, part of a group which flew to Earth, come along asking me to judge which of you truly thinks. You ask me to judge using your criteria of fundamental subjective qualities, and the alien machine agrees with that basis. You then say you can think and it can’t, while it says it can think and you can’t. But neither of you have given me any objective way to choose, all I have are subjective claims, I can’t make a ruling.

I understand the point you’re making. The terrible problem for you is that it doesn’t vindicate the Turing test. If I build a computer that passes the Turing test, there is simply not a reason to believe that it thinks; my knowledge about what went into the computer gives me reason, on the contrary, to doubt that thinks (as I’ve explicated constantly… the fact that it has no semantic meaning, the fact that it could utter equally complex gibberish which would be equally meaningful in another possible language system, the fact that I know by virtue of making it that it maps (name removed by moderator)uts to outputs without a need for internal, subjective representation). Humanlike behavior is not a sufficient condition for thinking.

The alien scenarios are tough to answer because they strain epistemology. But they also don’t go one way. Suppose that there is another alien race that is extremely intelligent. But they communicate supersonically and don’t move, but are constituted by a valuable mineral that humans would like to harvest. The air on their planet is filled with beautiful music and discussion of wildly advanced mathematics. They tell intricate and meaningful stories to one another. (Perhaps they communicate by pheromones or using magnetic fields.) But because their intelligence is not manifested in visibly and audibly human-like behavior, they are unintelligent.

I’m not suggesting that a test ought to catch every intelligent being, or that I have any epistemology for figuring out if the above are intelligent. The problem is just that there’s no one-to-one correspondence between human behavior and intelligence, and the supposition that there is central to the Turing test’s failure.

Not sure why any sane person would deny the existence of their thoughts.

Because they’re subjective and unverifiable, like qualia?

It’s interesting that those philosophers who believe in qualia can’t explain them yet expect neuroscience to do so. Do they complain that religion is limited because reading the NT doesn’t tell them what it feels like to know Christ? Why do they expect others to describe in words what they can’t put into words?

There seem to be a few assumptions at work here… a lot of philosophers who believe in qualia don’t expect neuroscience to do so (if by “neuroscience” one means neuroscience under its current methodological assumptions). Nagel, Chalmers, and Searle are examples. But generally they think that materialism is simply incomplete and that neuroscience would have to be overhauled to explain qualia.

The other assumption is that people who believe in qualia are religious. (I’m assuming “they” refers to “those philosophers” - otherwise, I apologize for reading into the vague pronoun.) Nagel and Searle are naturalists and atheists. Chalmers is a dualist, but still an atheist. The problem of qualia is bad enough for materialism that one need not be a credulous, unscientific old theist to endorse it! Imagine that.

Science rejects unprovable beliefs, but does not treat subjective aspects of reality as second class. Every day, peoples’ feelings, tastes and opinions are used as objective data. For example, it is perfectly legitimate science to study why some people think subjective feelings are second class.

The distinction between “primary qualities” and “secondary qualities” was not, I don’t think, invoked by anyone to imply that subjectivity is “second class.”

That said, some comments on that claim that “Science rejects unprovable beliefs”:

• As I’ve said, something like qualia aren’t “unprovable” to anyone. They can’t be quantified. You can’t write about them and get them into a peer-reviewed science journal. You can’t take another person and confirm that he has them. But it still could very well follow that it would be incoherent to deny them, since one easily proves them to oneself. (Experiment: Do I have subjective experiences? Yes.)
• Such a statement might depend on what one means by “rejects.” If it means, “science can’t investigate unprovable beliefs,” then that’s quite true; but if we have independent reasons to believe in qualia (ie. the fact that we are conscious subjects), then it just follows that science is insufficient to populate our ontology If “rejects” is some emotional casting-away, then we need not be concerned with it. If “rejects” means that science pronounces them false, then the statement itself is unfalsifiable, and so false.

 

[polytropos]

When I speak, my words mean something (necessarily, I think it is reasonable to grant). If I speak gibberish, my words don’t mean anything. When a machine speaks gibberish, its words don’t mean anything. The question is: when a machine speaks human language, does its words mean anything (necessarily, as would be necessary for the Turing test to produce usable results - if it can’t distinguish, then it fails)?

To clarify my meaning of “necessarily” here, I don’t mean that all of my utterances have meaning. I could say something in my sleep, or even just absentmindedly, that does not have meaning. I mean that, necessarily some of my utterances are meaningful. While it is possible that all of the utterances of a given machine that passes the Turing test are meaningless.

I will clarify because this is my main point regarding semantics.
(∃x)(Uxp & Mx): there is an utterance by me (polytropos, or any human speaker) which is meaningful.
(∃x)(Tx & (∀y)(Uyx ⊃ ~My)): for some machine which passes the Turing test, each of the machine’s utterances are meaningless.

This would show that we cannot conclude, on the basis of the Turing test, that a machine thinks. Its passing the Turing test is not sufficient for saying that its utterances have semantic content (for reasons given throughout my previous posts).

 

[Jewel34]

polytropos:

When I speak, my words mean something (necessarily, I think it is reasonable to grant).

No, not necessarily. You might be drunk. If you say this:

`Twas brillig, and the slithy toves
Did gyre and gimble in the wabe:
All mimsy were the borogoves,
And the mome raths outgrabe.

…does it mean something to someone? To you? But here we talk about a conversation between two different entities (not a monologue). Any conversation presupposes two entities who exchange information and a communication channel, where the information is transmitted. What entity “A” says may be meaningful to him, and meaningless to “B”.

polytropos:

If I speak gibberish, my words don’t mean anything.

What is gibberish? If I would switch to different language, which is unknown to you, my words would be gibberish to you. There is no “objective” gibberish, only unresolved informational content.

The meaning of a word or a proposition is resolved by the receiver of the information. The recipient reads or listens – in other words: “receives” the information, and it modifies its internal structure. If the modified structure corresponds to the structure of the sender, we have mutual understanding.

The question is how do we find out if the internal structures correspond to each other. We do not have any direct way to compare them. (If I would say “zwei” and you would understand German, some part of your internal structure would change to represent “two”. The neural states would be completely different, yet there would be mutual understanding. How to find out if that is the case? You got it: “the Turing test”. Keep on conversing and eventually you can ascertain that there was mutual understanding.

The sheer beauty of the Turing test is that it can be used to gauge human understanding, too.

polytropos:

The demand for a definition of thinking seems like a clear double standard to me.

It was you who introduced the proposition: “I was thinking about you last night”, and asked me about its “truth-value”. You can’t have it both ways.

polytropos:

As I’ve discussed in the past, good philosophy is not done in defining terms.

Well, you said that before, and I reject it. Discussing anything, not just philosophy demands that the basic terms are defined clearly, in a mutually agreeable manner, otherwise the discussion cannot even start. The preliminary part of the conversation may very well be about clarifying the terms – that is fine. But if that part does not lead to mutual understanding, any continuation is futile.

polytropos:

I don’t think we need to get into technical aspects of semantics here.

On the very contrary, it is the central question. The point of the test to find out if the other party is only “parroting” some syntactically correct answers, or actually understands what it says. And this latter is a much more difficult problem to tackle.

“Agreement” seems to refer to too broad of a set of phenomenon for it to be sufficient for a machine that passes the Turing test to be said to be intelligent.

Agreement is not a formal hand-shaking. It is revealed by the coherence of the conversation.

polytropos:

If comprehension can only be gauged indirectly, and the indirectness of the epistemological method imposes intractable limits on what it can tell us, then so much the worse for the epistemological method (or so much the worse for the philosophical conclusions one would like to draw).

How can you directly gauge my comprehension? As usual, I am open to suggestions. What other method are you going to offer how to gauge comprehension? Simple criticism should only be a first step. How do you know if I comprehended (understood) what you say? Obviously, by comparing your original meaning (subjective to you) and my expressed meaning, which you derive from my answers to your propositions. It is you who can gauge my understanding. I may assert that I understood what you say, but if my answers do not support it, then you can reveal my lack of comprehension.

polytropos:

Actually the logical positivists did not apply their verification principle to all propositions.

Yes, they did. Their basic stance was that one needs to verify empirically every proposition to be considered true. But this is not relevant here.

 

[Jewel34]

polytropos:

To clarify my meaning of “necessarily” here, I don’t mean that all of my utterances have meaning. I could say something in my sleep, or even just absentmindedly, that does not have meaning. I mean that, necessarily some of my utterances are meaningful. While it is possible that all of the utterances of a given machine that passes the Turing test are meaningless.

I will clarify because this is my main point regarding semantics.
(∃x)(Uxp & Mx): there is an utterance by me (polytropos, or any human speaker) which is meaningful.
(∃x)(Tx & (∀y)(Uyx ⊃ ~My)): for some machine which passes the Turing test, each of the machine’s utterances are meaningless.

This would show that we cannot conclude, on the basis of the Turing test, that a machine thinks. Its passing the Turing test is not sufficient for saying that its utterances have semantic content (for reasons given throughout my previous posts).

And here you keep talking about “thinking” and “meaning”. Yet you reject the necessity of rigorously define these terms. Are you serious?

 

[inocente]

The essence need not be “beyond the stars,” but in reality itself. How would you know there is no objective essence without a presumption?

If “duck” is an approximation, what is it an approximation of or about? You assume the “reality” of an objective essence in that statement.

There are two assumptions possible

1. A “duck” is only what human determination makes it.
2. A human determination of “duck” is an attempt to grasp the reality of what a “duck” really is as an aspect of all reality.

The first is human centered as if our knowledge is prime and determines reality.

The second views the enterprise from a reality focus rather than as knowledge focused.

The difference might be summed up as the difference between “our knowledge tells us” and “reality tells us.”

The first assumes “duck” is only what we decide it to mean.

The second assumes the “duck” is real and exists in itself and apart from our determinations about it as an aspect of a larger world view that reality exists “in itself” and apart from our approximations that, hopefully, draw us closer to it.

Truth exists, truth is not just our current “opinion” of things, though our current opinions may be distant or close approximations of reality.

---

Added after edit

Unfortunately, a great deal of “science” that occurs today is of the Type 1) variety. “We” will decide what reality is, based upon the means “we” have decided to use to determine reality.

A more fundamental approach would be to recognize that our methods are “our methods” and limited, but there is room for allowing reality to “reveal itself” to us under its terms, not our predetermined ones.

The Turing Test is another “predetermination” that fails to take account of what it tries to explain in the first place. Thinking is reducible to or “nothing but” behaviour, therefore certain behaviours entail thinking. Thinking is denied in the redefinition.

I can’t really make head or tail of that, except apparently you believe that duck-ness, along with crushed-beer-can-ness, pile-of-rubble-I-made-yesterday-ness and next-years-version-of-the-iPad-ness have and always will exist as eternal truths.

Have a look at a scientist explaining how scientists do it, which is way more simple (the clip is two minutes):

youtube.com/watch?v=05WS0WN7zMQ

 

[inocente]

That might be your “foundation stone” of ethics, but it isn’t a sufficient one, precisely because “behaving” like a human leaves undefined exactly what “like a human” means. How can the behaviour of a human (trivially defined as the behaviour humans demonstrate) suffice as an ethical foundation? Humans kill, rape, steal, etc. These are all entailed under “behaves like a human” which means, according to your stipulation, that these are all ethically acceptable because they are human behaviours.

Ethical cannot mean or equate to “human” because it would then not make sense to characterize some behaviours as inhuman or unethical since humans do, in fact, engage in them.

I have no qualms characterizing rape or child molestation as “sub-human” even though some humans do engage in these behaviours. Ethical value (worth) does not hinge upon the current psychological state or set of behaviours exhibited by an individual at any moment in time precisely because psychology and behaviours are malleable in a way that “human” is not. Humans may act in sub-human, inhuman and unethical ways without completely foregoing their humanity in any final sense. Although, that, too, may be possible.

Again you seem prone to complicating things. It sounds as if you’re willing to characterize rapists as sub-human. That, when formalized, is surely how white supremacists’ or Nazis’ do things. They replace “if it looks like a human and behaves like a human” with “if it looks like me and behaves like me”.

Have a look at the UDHR. Ethically, it assumes “if it looks like a human and behaves like a human” and then, in article 2, ensures that has the widest possible meaning.

 

[inocente]

If “debating games” does not refer to Philosophy (as far, as I remember, the names of “subjects” tend to be capitalised in English - aren’t they?), what do those words refer to…?

How on earth did you imagine I was posting to all philosophers all over the world? I was posting to you, MPat. I wrote down a number of explicit questions which can be answered by research and said they will lead to others and that’s how science progresses, to which you made sarcastic remarks, and I called that playing debating games.

Pleeeeese can we move on now?

But is it a machine that “cons us”? Or is it its maker?

If the machine had decided to pass the test all by itself, then we might discuss, if it is “intelligent” or “rational”. Actually, we could do this even if it would fail the test afterwards (as you can see, the test doesn’t really add much).

But in reality the first machine to pass the Turing test is going to be made to pass that test. Its creators will look for ways in which the “experts” could distinguish it from a human and try to make them fail. So, the makers of the machine will fool the experts, not the machine itself. And yes, they will be intelligent enough to be considered humans.

No. Obviously the first thing to do is isolate the machine, as otherwise it could just contain a mobile phone with humans at the other end answering the questions. Then you pose novel problems requiring learning which it could not be programmed for. You do this all day every day for years if you want, until reaching the required statistical significance level over the null hypothesis. Experimenters design protocols all the time.

*How did you reach such conclusions?! I didn’t speak about any “guilt” or “innocence” nor about any “ethics”! And the “rules of evidence” are the same: if we know that something has been specifically made to fool us, we take that into account.

And if you think that defining “human” as “someone who passes the Turing test” can prevent the problems with deciding “who is human and who is not” - I think you are wrong. Just think - could it prevent Nazis from concluding that Jews or (perhaps especially) the ones with psychiatric diseases are “not humans”…? Can all humans pass the “Turing test” all the time? Not to mention that the Turing test is simply impossible unless we start with at least a good idea who is a human and who is not.*

The innocent/guilty analogy was an analogy.

I don’t understand the bit about humans having to take the Turing test to be considered human. Turing proposed his test to as a way of to see if machines can think. Not as a way to declare machines human or as something every child has to go through as some kind of rite of passage.

I am saying that you (that would be MPat ) are going to town on finding every reason under the sun for never allowing that a machine can think. But if you applied that same extreme skepticism to who is and isn’t a human being, you would be going down a very dangerous ethical path.

 

[inocente]

There are other issues faced by the crude syllogism intelligence can be measured, thinking is necessary for intelligence, so we have now in principle cracked how to know if machines think. Measurement of intelligence presupposes the existence of intelligence. But no one would test the intelligence of a calculator, or even a powerful program, even though they could perform better than we could on some tests.

I got lost. Come back and rescue me.

I can tip a bucket containing 7 apples into a barrel containing 5 apples to do addition. Intelligence isn’t counting. Take one definition of intelligence: “A very general mental capability that, among other things, involves the ability to reason, plan, solve problems, think abstractly, comprehend complex ideas, learn quickly and learn from experience. It is not merely book learning, a narrow academic skill, or test-taking smarts. Rather, it reflects a broader and deeper capability for comprehending our surroundings—“catching on,” “making sense” of things, or “figuring out” what to do”.

To score above zero, a machine would need to display all of that capability, not just one limited area. But that is essentially what Turing is doing, he is using our capability across that entire range to test the machine’s capability.

Incidentally, rejecting it on that basis does not really solve any of the difficulties it proposes. The reason qualia are considered as a group is that each individual subjective experiences proposes the same problem that materialism has no answer for in principle. But if you like, divide them up (as I have in this conversation). The experience of seeing red is as intractable for materialism as all of qualia; materialists only “handle” it be ignoring it.

We might as well cut to the chase and call qualia a god of the gaps. What’s the difference? Just give something we don’t understand a name so we’ll feel better.

But if I knew about more experiments I think it may already be possible to reject qualia on evidence alone.

For example, Russians apparently divide dark and light blue in the same way that English speakers divide blue and green. This study found that they distinguish dark from light blue more quickly as a result. This supports the claim that color sensations are learned responses which are culturally dependent.

A number of languages do not distinguish blue and green, again meaning differences between cultures.

Perhaps dissonance and equal temperament in music could also be included, since different peoples’ have different sensations of what they find musical.

Whatever, there seems to be no evidence for anything mystical going on, unless you can name any.

 

[inocente]

I understand the point you’re making. The terrible problem for you is that it doesn’t vindicate the Turing test. If I build a computer that passes the Turing test, there is simply not a reason to believe that it thinks; my knowledge about what went into the computer gives me reason, on the contrary, to doubt that thinks (as I’ve explicated constantly… the fact that it has no semantic meaning, the fact that it could utter equally complex gibberish which would be equally meaningful in another possible language system, the fact that I know by virtue of making it that it maps (name removed by moderator)uts to outputs without a need for internal, subjective representation). Humanlike behavior is not a sufficient condition for thinking.

See my post #143, point 3. You need to explain why, come the day when we know as much about ourselves as we do about our machines, your argument won’t evaporate. So far you’ve only said there’s something mystic in the air which will deny us a full explanation. But my point is that even without whatever it is you think will still be missing, once we know the language of thought then, as I’ve explicated constantly… the fact is that it has no semantic meaning.

*The alien scenarios are tough to answer because they strain epistemology. But they also don’t go one way. Suppose that there is another alien race that is extremely intelligent. But they communicate supersonically and don’t move, but are constituted by a valuable mineral that humans would like to harvest. The air on their planet is filled with beautiful music and discussion of wildly advanced mathematics. They tell intricate and meaningful stories to one another. (Perhaps they communicate by pheromones or using magnetic fields.) But because their intelligence is not manifested in visibly and audibly human-like behavior, they are unintelligent.

I’m not suggesting that a test ought to catch every intelligent being, or that I* have any epistemology for figuring out if the above are intelligent. The problem is just that there’s no one-to-one correspondence between human behavior and intelligence, and the supposition that there is central to the Turing test’s failure.

That would make a fine short story. Unfortunately, you don’t answer my point. If you cannot come up with a test, you either have to accept Turing’s or accept that you are saying God cannot ever, under any circumstances, make a machine think. Either way, Turing has got you.

Because they’re subjective and unverifiable, like qualia?

There seem to be a few assumptions at work here… a lot of philosophers who believe in qualia don’t expect neuroscience to do so (if by “neuroscience” one means neuroscience under its current methodological assumptions). Nagel, Chalmers, and Searle are examples. But generally they think that materialism is simply incomplete and that neuroscience would have to be overhauled to explain qualia.

They should get together with ID fans, who also want to castrate science to let in their beliefs along with astrology and whatever else. Interesting that so many people idolize science to the extent that they want their superstitions to be let in.

The other assumption is that people who believe in qualia are religious. (I’m assuming “they” refers to “those philosophers” - otherwise, I apologize for reading into the vague pronoun.) Nagel and Searle are naturalists and atheists. Chalmers is a dualist, but still an atheist. The problem of qualia is bad enough for materialism that one need not be a credulous, unscientific old theist to endorse it! Imagine that.

See above. Also, the penultimate sentence would look just as good if you replace “qualia” with “evolution”.

*The distinction between “primary qualities” and “secondary qualities” was not, I don’t think, invoked by anyone to imply that subjectivity is “second class.”

That said, some comments on that claim that “Science rejects unprovable beliefs”:

• As I’ve said, something like qualia aren’t “unprovable” to anyone. They can’t be quantified. You can’t write about them and get them into a peer-reviewed science journal. You can’t take another person and confirm that he has them. But it still could very well follow that it would be incoherent to deny them, since one easily proves them to oneself. (Experiment: Do I have subjective experiences? Yes.)
• Such a statement might depend on what one means by “rejects.” If it means, “science can’t investigate unprovable beliefs,” then that’s quite true; but if we have independent reasons to believe in qualia (ie. the fact that we are conscious subjects), then it just follows that science is insufficient to populate our ontology If “rejects” is some emotional casting-away, then we need not be concerned with it. If “rejects” means that science pronounces them false, then the statement itself is unfalsifiable, and so false.*

It means if there is no evidence for a hypothesis, and no logical requirement either, and no reason under the sun to support it, then bin it, why carry around useless baggage which is neither use nor ornament?

 

[MPat]

How on earth did you imagine I was posting to all philosophers all over the world? I was posting to you, MPat. I wrote down a number of explicit questions which can be answered by research and said they will lead to others and that’s how science progresses, to which you made sarcastic remarks, and I called that playing debating games.

Pleeeeese can we move on now?

Ah, that’s what you meant… And I thought that you consider all philosophical discussion (or this specific discussion) worthless - and in such case it would have been fair to remind you that you are also participating in it and not doing research (and that there is no good reason to spoil the fun for everyone else)…

Thus, sorry about this misunderstanding…

No. Obviously the first thing to do is isolate the machine, as otherwise it could just contain a mobile phone with humans at the other end answering the questions. Then you pose novel problems requiring learning which it could not be programmed for. You do this all day every day for years if you want, until reaching the required statistical significance level over the null hypothesis. Experimenters design protocols all the time.

And what if the maker of the machine managed to steal those “novel problems requiring learning which it could not be programmed for” and prepared for them in advance…? Or (more likely) guessed them…?

Some guesses are clear. For example, the computer that passes the Turing test will have to add some “latency” (and maybe even mistakes) answering any question that includes computation (like adding two large numbers).

A second problem is that humans also have problems solving “novel problems requiring learning which it could not be programmed for”…

The innocent/guilty analogy was an analogy.

Then I have to admit that I don’t get it…

I don’t understand the bit about humans having to take the Turing test to be considered human. Turing proposed his test to as a way of to see if machines can think. Not as a way to declare machines human or as something every child has to go through as some kind of rite of passage.

Well, I was trying to respond to this:

I think the ethics are very dangerous, since it hinges on how you decide who is human and who is not, and we know all too well where that has led in the past.

Presumably, your point was that if we declared any machine that passed the Turing test a “human”, we would avoid “crimes against humanity against machines”. Thus I indicated that such a standard wouldn’t have prevented the “crimes against humanity against humans” either… Thus the Turing test is not very useful here.

I am saying that you (that would be MPat ) are going to town on finding every reason under the sun for never allowing that a machine can think. But if you applied that same extreme skepticism to who is and isn’t a human being, you would be going down a very dangerous ethical path.

And where is the fun in having an opponent who gives up easily…? By the way, you do not seem close to giving up either…

And I hope you will spell out the argument (if there was an argument) more clearly. At the moment it is only clear that it will have some problems with “slippery slope”…

Anyway, one difference between humans and machines in this case is clear: humans haven’t been made with a clear intent to pass the Turing test.

 

[inocente]

Ah, that’s what you meant… And I thought that you consider all philosophical discussion (or this specific discussion) worthless - and in such case it would have been fair to remind you that you are also participating in it and not doing research (and that there is no good reason to spoil the fun for everyone else)…

Thus, sorry about this misunderstanding…

No problem. My turn to apologize - sorry it’s taken so long to rrply.

And what if the maker of the machine managed to steal those “novel problems requiring learning which it could not be programmed for” and prepared for them in advance…? Or (more likely) guessed them…?

You wouldn’t devise the problems until you’d got the machine isolated under lock and key with no possibility of tampering.

*Some guesses are clear. For example, the computer that passes the Turing test will have to add some “latency” (and maybe even mistakes) answering any question that includes computation (like adding two large numbers).

A second problem is that humans also have problems solving “novel problems requiring learning which it could not be programmed for”…*

It’s easy to slow machines down, although if the machine has human intelligence, it might not be any quicker at sums than we are. The human brain is a masterpiece of engineering is terms of low energy consumption, it’s orders of magnitude more efficient than today’s computers, so the problem is more likely to be that the machine is too slow rather than too fast.

*Presumably, your point was that if we declared any machine that passed the Turing test a “human”, we would avoid “crimes against humanity against machines”. Thus I indicated that such a standard wouldn’t have prevented the “crimes against humanity against humans” either… Thus the Turing test is not very useful here.
*

I think we both got lost now.

*And where is the fun in having an opponent who gives up easily…? By the way, you do not seem close to giving up either…

And I hope you will spell out the argument (if there was an argument) more clearly. At the moment it is only clear that it will have some problems with “slippery slope”…

Anyway, one difference between humans and machines in this case is clear: humans haven’t been made with a clear intent to pass the Turing test.*

I am saying that any argument in the form “I am X and can think, therefore I will only accept that others who are X can think” is dangerous. It might seem innocuous where X=human, but not when X=white skinned or X=male, for instance.

 

[Peter_Plato]

I am saying that any argument in the form “I am X and can think, therefore I will only accept that others who are X can think” is dangerous. It might seem innocuous where X=human, but not when X=white skinned or X=male, for instance.

Since when is “white skinned” or “male,” apart from being human, a reason for according equal treatment? Birch trees are “white skinned” and warthogs can be “male.” Your “argument” actually supports the fact that it is dangerous to look only at externals such as behaviour, white skin or “maleness” as determiners of qualities or characteristics that ought not so easily be determined.

In short, your answer here seems to imply that behaviour, like white skin are “transcendent” properties that can be equally applied because they are important in themselves. That is questionable logic. And your own examples demonstrate the deficiency.

Interesting news story about the supposed “deaf signer” at Nelson Mandela’s funeral - a story that seems appropriate in this discussion. The deaf man obviously displayed behaviour that appeared to be sign language and fooled not a few people, until one person recognized that the “signs” did not correspond with the meaning of the audibles coming from the speakers.
A lesson, perhaps, in not trusting behaviours which only appear to correspond with meaning. If it looks like a signer and moves like a signer are NOT sufficient reasons to call it a signer.

 

[MPat]

No problem. My turn to apologize - sorry it’s taken so long to rrply.

Not a problem: it is understood that discussions here are not our main occupation in life.

You wouldn’t devise the problems until you’d got the machine isolated under lock and key with no possibility of tampering.

Well, are we still dealing with the same Turing test after all those precautions…?

Another point is that if it is possible to “cheat” in the Turing test, then that test does not appear to be as useful, as it was supposed to be. That is, a machine that is not able to think can also pass it.

It’s easy to slow machines down, although if the machine has human intelligence, it might not be any quicker at sums than we are. The human brain is a masterpiece of engineering is terms of low energy consumption, it’s orders of magnitude more efficient than today’s computers, so the problem is more likely to be that the machine is too slow rather than too fast.

The point is that if the computer can find the answers to other problems reasonably fast, it will have to be slower while answering something like “5643646346384634 + 6387617368786387 = ?”. And that is the type of question that can be foreseen and will be prepared for.

I am saying that any argument in the form “I am X and can think, therefore I will only accept that others who are X can think” is dangerous. It might seem innocuous where X=human, but not when X=white skinned or X=male, for instance.

Unfortunately, it doesn’t get any better with “X=‘someone who can pass the Turing test’”… It is at least not certain that one needs to be able to think in order to pass the Turing test and it is almost certain that one can fail the Turing test while being able to think.

Also, I don’t think anyone is arguing in favour of that proposition with “X=human”. For example, we can probably agree that angels exist and that they are able to think, but angels are not humans. Thus they are a counterexample. By the way, I suppose that some books of the Bible do indicate that angels are perfectly capable of passing the “informal” Turing test…

Also we could happily agree that aliens capable of thinking are possible (maybe they exist, maybe not). They would not be “conventional” humans, thus they can also be a counterexample. By the way, it is not certain that they would pass the Turing test.

What is doubted here is something different: that the computer that passes the Turing test is to be considered a thinking being for that reason alone. By the way, if we got some special revelation from God, saying that a specific machine has been given the ability to think, that would be a different piece of evidence…

 

[polytropos]

No, not necessarily. You might be drunk.

The quantifications given in the #164 clarify this point. I do not mean that every statement uttered by a human is meaningful, but that humans can utter meaningful things. (But the evidence for this is not external, so the Turing test could now show us that a machine makes meaningful utterances, even if the machine does somehow make meaningful utterances. That is the Turing test’s epistemic issue: even if what it were looking for were present, which is dubious, it wouldn’t be able to find it.)

`Twas brillig, and the slithy toves
Did gyre and gimble in the wabe:
All mimsy were the borogoves,
And the mome raths outgrabe.

…does it mean something to someone? To you? But here we talk about a conversation between two different entities (not a monologue). Any conversation presupposes two entities who exchange information and a communication channel, where the information is transmitted. What entity “A” says may be meaningful to him, and meaningless to “B”.

What is gibberish? If I would switch to different language, which is unknown to you, my words would be gibberish to you. There is no “objective” gibberish, only unresolved informational content.

Your example of a meaningless statement makes my point. It is meaningful to us. The issue for a Turing test is that for any utterance (such as the one you provided), there exists a conceivable language system in which it has semantic meaning (we need only generate a set of rules with which it is consistent). (Cf. indeterminacy of translation.) What this shows is not that speaking gibberish is meaningful, but that the coincidence of outputs with a language system (even English) is not sufficient for semantic content. But then all the Turing test delivers is the coincidence of outputs with a language system.

The meaning of a word or a proposition is resolved by the receiver of the information.

This criterion doesn’t work. You are stranded on an island. No one is in the vicinity. You shout, “Get me out of here!” There is no recipient. But your word had meaning.

A machine is on an island. It has an infrared sensor; it is programmed to emit a supersonic signal when there is no live entity within 100 yards of it. It emits such a signal, but again there is no recipient. Does the signal have meaning?

The issue is that intelligence and thinking are not social phenomena (even if we use them socially in many cases). Words with no recipients are still meaningful, since the meaning depends not on the recipient but on the person speaking. Intelligence is not a mutable quality that changes depending on who’s listening.

If a machine passes the Turing test, but then all English language speakers die, we can’t change our evaluation of its intelligence. But then we face another issue: if a machine is given a random set of language rules that do not conform to any extant language, it could garble gibberish (relative to every extant language) and it would have to be as meaningful as if it constructed an English sentence from a set of English rules. That is the issue with the Turing test; following rules to generate outputs is not sufficient for meaning.
Jewel34;11475531I:

t was you who introduced the proposition: “I was thinking about you last night”, and asked me about its “truth-value”. You can’t have it both ways.

That’s not the issue. The issue is that Turing (or at least his proponents) takes his test to show that machines can think. But he does not define thinking either; he explicitly avoids doing so. If my argument needs a definition of thinking for it to have any meaning, then Turing’s needs a definition of thinking for it to have any meaning.

This is purely ad hominem, of course, since I don’t think all philosophy should be still-born until there is agreement in terms. That just seems like a way to halt conversation, since two sides in a debate like this are never going to agree on definitions.

 

[polytropos]

Well, you said that before, and I reject it.

Did I say it to you, or to Bagheera?

Discussing anything, not just philosophy demands that the basic terms are defined clearly, in a mutually agreeable manner, otherwise the discussion cannot even start. The preliminary part of the conversation may very well be about clarifying the terms – that is fine. But if that part does not lead to mutual understanding, any continuation is futile.

I disagree. Professional philosophers do not begin discussions by agreeing in terms first, because there will never be agreement in terms. (If you recall, we have both provided definitions of what we mean by thinking in this discussion. A while back I said that thinking is “judgmental understanding” and listed a few necessary characteristics like intentionality. I don’t think we commented on each other’s definitions. Why? Because that is what doesn’t get us anywhere.) Philosophers advocate for their own views in their own terms. When critiquing another view, they adopt the terms of the opponent and show why they think they don’t work (ie. why they lead to an inconsistency, or entail some highly counterintuitive result). That is my goal; I’m not advocating a positive position, so I do not have terms to define. I am saying X is a necessary component of any substantive definition of thinking (I have defined it several times in this discussion), and clarifying why the Turing test does not work on that basis.

If my opponent’s definition of thinking is weaker than my necessary condition, then that’s fine. My argument won’t touch him, but it also won’t need to touch him, because his definition will be too weak to draw any robust conclusions from the Turing test.

The point of the test to find out if the other party is only “parroting” some syntactically correct answers, or actually understands what it says. And this latter is a much more difficult problem to tackle.

I agree, that is the question. But the problem is that “parroting” is not limited to searching an index of known questions and giving an answer. Pumping a conversant’s utterance through a series of equations and applying several sets of linguistic rules is just as much “parroting.” A computer doesn’t care which it is doing.

That said, the first option doesn’t seem conceptually impossible at all. I don’t see why one could not in theory compile a full list of questions and answers, and somehow decrease the runtime with the proper hardware, such that a computer could “brute force” the Turing test.

Agreement is not a formal hand-shaking. It is revealed by the coherence of the conversation.

I didn’t say it’s a “formal hand-shaking” - and the examples I gave certainly should not have given that impression. I gave examples of systems in which there are sets of rules that are followed - leading to coherent “conversations” (or sets of phenomena that could conceivably be isomorphic with a human conversation). So this point doesn’t meet my argument.

How can you directly gauge my comprehension? As usual, I am open to suggestions. What other method are you going to offer how to gauge comprehension?

Hm? I’m not the one making positive claims. I’m not saying that’s a way to watch someone and definitively say that it comprehends without the additional premises that 1). he’s a human, 2). I’m a human, and 3). I can comprehend things. I don’t need an alternative to the Turing test to point out that it can’t show what it claims to show. The fact that there might not exist an alternative is not my problem.

Yes, they did. Their basic stance was that one needs to verify empirically every proposition to be considered true. But this is not relevant here.

Logical positivists did not see analytic truths as needing empirical justification. That’s a historical fact. But you’re right, we don’t need to debate this here.

 

[polytropos]

I can tip a bucket containing 7 apples into a barrel containing 5 apples to do addition.

An aside: Is it? Certainly we could use apples in a barrel as a rudimentary calculator (like children count on their fingers). But how do you know you are not just adding billions of atoms to billions of atoms? How do you know that you are not multiplying the quantity of apples in the first bucket by 2.4? There is a natural interpretation, of course, but the “translation” into mathematical language is indeterminate. But the same difficulties in principle face a computer. Its outputs and (name removed by moderator)uts have meaning because I attribute meaning to them and develop a computer to accomplish certain ends of my own. But the computer isn’t objectively telling me anything - its operations are always indeterminate.

An example: for the computer, is 0 + 1 = 1 addition (“0 plus 1 equals 1”), or formal disjunction (“p v q is true where p is false and p is true”)? The answer is neither: it depends on what a human is using it for.

Take one definition of intelligence: “A very general mental capability that, among other things, involves the ability to reason, plan, solve problems, think abstractly, comprehend complex ideas, learn quickly and learn from experience. It is not merely book learning, a narrow academic skill, or test-taking smarts. Rather, it reflects a broader and deeper capability for comprehending our surroundings—“catching on,” “making sense” of things, or “figuring out” what to do”.

To score above zero, a machine would need to display all of that capability, not just one limited area. But that is essentially what Turing is doing, he is using our capability across that entire range to test the machine’s capability.

I have no glaring objections to the definition. But the issue is simply that the Turing test can’t be used to verify most of those, because the coincidence of outputs does not need to be accompanied by reasoning, abstract thought, comprehension, “catching on,” “making sense,” or “figuring out.” All of those terms, in any ordinary use of the English language, denote internal processes. As long as we believe that the subject in question has a mind, then it makes sense to suppose that his performance is an indicator of what he’s thinking. But the performance alone does not allow us to conclude that he has a mind (as I said to Jewel, we need additional premises: 1). he is a human, 2). I am a human, 3).I know that I think).

(Note, one could soften the definitions for reasoning, abstract thought, comprehension, mind, etc., but then the conclusion would be trivial.)

But if I knew about more experiments I think it may already be possible to reject qualia on evidence alone.

For example, Russians apparently divide dark and light blue in the same way that English speakers divide blue and green. This study found that they distinguish dark from light blue more quickly as a result. This supports the claim that color sensations are learned responses which are culturally dependent.

A number of languages do not distinguish blue and green, again meaning differences between cultures.

Perhaps dissonance and equal temperament in music could also be included, since different peoples’ have different sensations of what they find musical.

Hm? No one said everyone experiences qualia in the same way. I could have an inverted spectrum compared to you (and no one would know it) but it still wouldn’t make qualia less coherent.

Not to mention: some people are colorblind. No one is denying that colorblindness has a physical basis depending on what information reaches the brain. The question of how what does reach the brain is experienced as a quale remains intractable because, of course, the data was irrelevant to it.

Whatever, there seems to be no evidence for anything mystical going on, unless you can name any.

I agree they’re not “mystical.” I’ve said that I think qualia and consciousness have a “material” basis, just that the conceptual tools required to explain them are unavailable to philosophical naturalism, based on the way it has defined itself historically.