The Turing Rule
I want to bounce something off of anyone willing to contribute.
First, a statement from @T Clark
Quoting T Clark
The Turing Test
If a machine can fool a person into believing that it itself is a person, it must be considered as AI. In other words, AI is a person.
The premise underlying the Turing Test is:
Gottfried Wilhelm Leibniz's the principle of the identity of indiscernibles which, unlike its converse, the principle of the indiscernibility of identicals, is, last I checked, controversial.
The Turing Rule is the principle of the identity of indiscernibles and it's the premise on which the Turing Test is based.
Please discuss. Much obliged.
First, a statement from @T Clark
Quoting T Clark
From Wikipedia:
A philosophical zombie or p-zombie argument is a thought experiment in philosophy of mind that imagines a hypothetical being that is physically identical to and indistinguishable from a normal person but does not have conscious experience, qualia, or sentience. For example, if a philosophical zombie were poked with a sharp object it would not inwardly feel any pain, yet it would outwardly behave exactly as if it did feel pain, including verbally expressing pain.
An unself-conscious and unaware organism that acts as if it's self-conscious and aware in a way that cannot be detected either physically or by observing its behavior is conscious and aware.
The Turing Test
If a machine can fool a person into believing that it itself is a person, it must be considered as AI. In other words, AI is a person.
The premise underlying the Turing Test is:
Gottfried Wilhelm Leibniz's the principle of the identity of indiscernibles which, unlike its converse, the principle of the indiscernibility of identicals, is, last I checked, controversial.
The Turing Rule is the principle of the identity of indiscernibles and it's the premise on which the Turing Test is based.
Please discuss. Much obliged.
Comments (41)
Careful with the syllogism. Not that the computer, if it passes the test, is a person. It is that the computer is intelligent.
What you call "the identity of indiscernables," a phrase I hadn't heard before, is a central one to how I see the world. If you can't tell the difference, there is no difference. Both the Turing test and the P-zombie apocalypse are good tests of the principle.
You can't tell the difference and, ergo, by the Turing rule, the AI is a conscious (makes it a person) OR, intriguingly, if that's a hard pill to swallow, acknowledge the problem of other minds.
Quoting T Clark
I avoided including p-zombies in the OP because I wanted to focus on the Turing principle.
Anyway, since you brought it up, let's discuss.
A person (P), an AI (I), and a p-zombie (Z) are all identical as in they can't be identified if all 3 are in a room.
Ergo, as per the Turing principle:
1. P, I, Z are all actually conscious. P-zombies are impossible (physicalism) but AI is conscious.
2. P, I, Z are all not actually conscious. P-zombies are possible (nonphysicalism) but other persons lack consciousness (solipsism).
Dude, don't re-interpret the Turing test. Stick to what the Turing test says.
I have done no such thing.
I'm mostly interested in the broad principle you described, what you call the "identity of indescernibles," rather than the specific examples, i.e. P-zombies and the Turing test. It pops up all the time, e.g. different interpretations of quantum mechanics or the existence of universes outside our own.
So, what's your take? Do you think the Turing principle (identity of indiscernibles) is justified/unjustified?
Does it follow that, if X and Y are indistinguishable, X = Y?
As I noted, the so-called "identity of indiscernibles" is central to my beliefs. So, yes, the Turing test is a reasonable way to see consciousness. And there's an even broader principle. As William James wrote:
Pragmatism asks its usual question. "Grant an idea or belief to be true," it says, "what concrete difference will its being true make in anyone's actual life? How will the truth be realized? What experiences will be different from those which would obtain if the belief were false? What, in short, is the truth's cash-value in experiential terms?
Not only are two things the same if you can't tell them apart, they're the same if there is no practical, meaningful, concrete difference between them or their consequences.
How about being on hold? A very polite bot does not recognize that a live person expects variation even when repeating the same statement. Leibniz was an ass. The identity of indiscernibles is inane, would make us real suckers for Raven, or is it Mystique?
Turing didn't think a human could be fooled. He also set conditions on the test that would make it very hard not to be. Language is not what philosophers want us to think it is. It is a living growing drama. It is intimacy. Something a machine can never bring to it. A young man of my acquaintance, when annoyed at you, would call you a tino-nino. If very annoyed, a tino-nino-nuckinuck! No machine could ever get the meaning, but even a simpleton human would instantly grasp it. If we are forced to stick to defined terms the sophisticated systems available today certainly could give us a run for the money, but if forced to throw away the lexicon, the human will win hands-down every time. The fact is, all linguistic terms are the result of a synthesis that is still perfectly enigmatic to science and to philosophy, and AI can only win the Turin-Test by locking us into fixed terms and ignoring the mystery of synthesis. It's a drama intimated amongst us, not explicated and recorded in some unchanging format. And if AI really wants to fool us, I suggest it teach its systems how to avoid mindlessly repeating the same assertions without at least altering its inflection to recognize the uniqueness of a real human listener.
This is interesting. I’m not sure either that a human would necessarily be fooled—it seems logically possible for an AI to be indiscernible from a human, but in reality a person could discern the two if they understood the limitations in the AI’s programming and exploited them to discover it.
This might be wrong then, since limitations in the AI’s programming would be unavoidable.
What if there aren't any machines? What if that is just a concept we project onto experience?
Imagine you have a dream. In the dream there is AI that communicates to you. The reality would be that this AI in the dream is a symbolic representation of your subconscious, or a collective unconscious etc. If Idealism is true (perhaps a big if), then the whole world is either a collective consciousness or singular consciousness. Either way, it would mean that AI would be stemming from consciousness, even though the AI (carrier of the message) itself could just be a projection or thought form, not itself possessing the consciousness.
Its kind of like......no matter how good a book is, how intelligently written, how lively the characters in the book, we know that it is only a carrier of a message, and not the intelligence itself.
You overestimate and underestimate at the same time.
What AI can never get past is that every time a human uses a word it conveys a difference. Maybe just a unique inflection or tempo, a micro-pause or elision. Something in every word that situates its meaning uniquely.
What? AI? I think you mean I am saying Turing did, as far as I have read of him.
When I look up "Turing principle" it discusses the computability of functions. I don't think that's what you're talking about. What do you mean specifically?
I edited the OP to correct the confusion.
When I was a psych major long, long ago, I remember reading about computer generated therapy. Here is a link to a program created back in the 1960s. Pretty limited, but apparently some people couldn't tell that it was computer generated.
http://psych.fullerton.edu/mbirnbaum/psych101/eliza.htm
Sure. I have met living therapists who are less engaged than a computer and no one seems to notice they aren't really there either...
I don't think it would be hard to create the illusion of intelligence, after all, most conversation is just a little dance of mechanized and predictable semantics.
The human examiner could not be fooled. The critics of Turing Test had already addressed its limitation -- the "test" is very limited to the basics to which both the human and computer subjects could say yes or no. So the test itself is not representative of what we, humans, would call adequate measure of human intelligence or consciousness. It is intentionally rigged so that not only the human subject, but also the computer could respond.
This appears to highlight a problem with claiming that two things are the same if they aren’t discernible: whether they’re discernible or not is subject to the type of examination you’re able to make; it doesn’t seem right to say that two things are the same simply because someone hasn’t under certain conditions been able to tell otherwise.
Never having read anything by Leibniz, I am assuming that the "principle of the identity of indiscernibles" would dictate that two items which are utterly indiscernible must be held to be identical, ergo the same type of thing?
Yes.
See below to a Marvel comics reference. Raven/Mystique refers to a shapeshifting supervillain/superhero.
Quoting Gary M Washburn
Thanks for the link to ELIZA. Very interesting.
To all three of the above members
1. The principle of the indiscernibility of identicals: No issue here. If a and b are the same thing, for every predicate applicable to a (Px), Pa, that predicate is also true for b, Pb. In other words, if a = b implies [math]Pa \leftrightarrow Pb[/math].
2. The principle of the identity of indiscernibles: Problematic. Why? Some but not all properties maybe shared. For instance, snow is white and also swans in the northern hemisphere are white. As far as whiteness goes, snow and these swans are indiscernible. Are snow and white swans identical? No! This is the weak version of the principle.
In defense of the principle of the identity of indiscernibles, it applies only if the objects in question are indiscernible in every sense i.e. if for every/all predicate Px, for a (Pa), it is also true that Pb, then a = b. This is the strong version of the principle and is, to my reckoning, clearly true.
What about the Turing test? Clearly, it's an application of the Turing rule (the identity of indsicernibles, 2 above) but then people know that an AI is a machine and ergo, is definitely distinguishable from a person. So, people may raise an objection to the Turing test, that is to say, the Turing rule is the weak version of the principle of identity of indiscernibles.
However, study closely what it is that's being assessed - consciousness. Consciousness has some behavioral correlates. For simplicity, let's say that the following predicates are true for consciousness: Pc and Qc where c is consciousness. Suppose now an AI (i) is such that Pi and also Qi. Put simply, the AI (i) is indistinguishable/indiscernible in every and all respects from consciousness (c). This, as you might've already noticed, is the strong version of the principle of identity of indiscernibles. In other words, the Turing test and the Turing rule are immune to the Raven/Mystique rebuttal.
I am currently chewing over...thinking about these two "principles". I have another question. Is "the indiscernibility of identicals" a proposition of Liebniz, as is "the identity of indiscernibles"?
This might help: Identity of indiscernibles/Indiscernibility of identicals
1. The indiscernibility of identicals: [math]\forall x \forall y[x = y \rightarrow \forall F (Fx \leftrightarrow Fy)][/math]
For any x and y, if x is identical to y, then x and y have all the same properties.
2. The identity of indiscernibles: [math]\forall x \forall y[\forall F (Fx \leftrightarrow Fy) \rightarrow x = y][/math]
For any x and y, if x and y have all the same properties, then x is identical to y.
Thank you. I have drawn a couple of conclusions about this, but am yet thinking it over. I will post something later today.
:ok:
Even if it is so intelligent after all.
Start with current top notch datamining capability and computer recognition, lets say every comment here on PF and on other Philosophical discussion site (still, quite finite amount of discussion threads), then add a great English language program, and realistically you could have a program that would fool people most of the time.
And still it would be a simple ordinary computer program using algorithms "if....then" to produce the fooling. Nothing AI about it!
How to know that it's a stupid computer program? It's only as good as the human programmer has made it. Get the program to participate in dialogue that the programmer hasn't anticipated and soon you might find the flaws. The program cannot do what it isn't programmed to do. (Even if it has likely a mechanism programmed in it to do in this case ad hominem attacks, which would we a very human response).
Don't you need to establish what you mean by being this one before you can use quantifiers like any and all to demonstrate it? And why is it we feel justified in using quantifiers to nail down what a quality is? Gibberish is gibberish even if encoded in arcane symbols. Qualities are not ways of being the same, they are similarities in ways of being different. The fact of the matter is that 'identical' is an artificial concept that can only be real through the manipulation of materials as well as a bogus idea of what qualities are. You can make two ball-bearings impossible for an unaided eye to distinguish, but try to find such similarities in nature! Living cells can even distinguish one atom from another, of the same element! Distinguishing one thing from anther is only problematic by intervention to hide the difference or in the fantasies of rationalists. But what really gets hidden is what an idea really is, and I would think a philosopher would not wish to be party to such deception.
Quoting TheMadFool
The "p-zombie" is an obvious impossibility. An "organism" cannot behave as if it is conscious and aware if it is not conscious and aware. Such a situation can only pertain to advanced computer architecture, that is, to AI.
Quoting TheMadFool
Sure, okay...
Quoting TheMadFool
Slow down, please. This, so long as the AI can be rendered, can be made to be, sentient, that is, can be made able to experience rational thought (check), sensation (probably, in some senses), and feeling...emotion (this is doubtful to me). I doubt that AI can be made to experience emotion, but rather to experience the semblance of emotion. AI architecture can be made to replicate human neural, bioelectrical anatomy, and so produce rational thought in quite an efficient manner. The human experience of emotion, though, is more dependent upon brain chemistry than upon neural architecture, and I am doubtful that this can be replicated in AI. Without the experience of emotion, I am not sure if you can characterize any subject as "a person".
Quoting TheMadFool
This is where I have had several pertinent thoughts. As someone deeply interested in language, my first thought regarding these two principles, the principle of the identity of indiscernibles and the principle of the indiscernibility of identicals, is that they do not represent conversive analogues of one another as they are herein stated in English. This is based upon a semantic distinction which I noticed immediately within the OP. This is because the nouns identity and indiscernibility are not directly analogous. The direct analogue to indiscernibility, an abstract noun derived from the adjective indiscernible would be identicality or identicalness, the abstract noun derived from the adjective identical, which nouns, both meaning "bearing utter likeness", have a much narrower semantic field that does identity which is ultimately an abstract noun derived from the Latin determiner idem, meaning "the same". Note that "the same" can mean "bearing utter likeness (to another)" or can mean "not the other, but the same one", in other words, "selfsame...the same as itself". Since this is so, the two lemmas, identicality and identicalness are more specific in their meaning, and so are the proper terms to use as analogues of indiscernibility. Of the two, I would choose to use identicality because of the morphological uniformity which it presents within the argument. Despite all this, I note that identity of indiscernibles is the terminology usually used for statement of the principle, and I only state my observation as an observation without demanding a change.
The symbolic representation of these two principles,
Quoting TheMadFool
provides a more accurate means of stating the principles for a criticism thereof. Therefore, let the following obtain:
?x?y[x=y??F(Fx?Fy)]
For any x and y, if x is identical to y, then x and y have all the same properties, or if two objects are absolutely identical then they must be indistinguishable from one another with respect to all of their properties;
?x?y[?F(Fx?Fy)?x=y]
For any x and y, if x and y have all the same properties, then x is identical to y, or if two objects are indistinguishable from one another with respect to all of their properties then they are identical.
My thought is that x=y??F(Fx?Fy) is a valid statement, while ?F(Fx ? Fy) ? x=y is invalid. The argument ?F(Fx ? Fy) ? x=y is dependent upon the premise that x and y can be found to be indistinguishable based upon ?F. I contend that this premise is false. I say this because of the inability of the human being to fully discern ?F, utter discernment of ?F not appearing to be achievable within reality. While x=y??F(Fx?Fy) holds as a matter of logic, ?F(Fx ? Fy) ? x=y is utterly dependent upon the discernment of ?F, which in actuality is impossible for the human being. In every case for which ?F is not discernible, which I argue is every case in reality, identity, or more properly identicality is not discerned despite the appearence of indistinguishability. Beyond that, I believe that ?F represents an ideal not to be found within the universe. I think that this is what Gary Washburn meant in stating that:
Quoting Gary M Washburn
since, therefore ?F is not a reality, ?F(Fx ? Fy) ? x=y can never hold in the universe, and so x can never be the equaivalent of y in reality.
These are my thoughts thus far on the subject of these two principles, to which I have only now been exposed. Maybe all that I have to say is nonsense...
Nothing formal can 'identify'. Period. Identity is personal. Leibniz is subordinating reality to formal rules he constructs from question-begging the meaning of equivalence. Can the meaning of 'equal' or 'same' or 'indistinguishable' derive itself? If everything real is unique the equivalence of attributes is simply irrelevant to identity. From what I've read, and it's been a while, Turing only expected his 'test' to even be meaningful, let alone definitive, limited to very narrow technical statements. Of course, these days, AI technicians know no end of ambition, and arrogance, and others let paranoia get to them.
Haha, they'll learn when they're plugged into the Matrix...
Yeah, my comment was rather charitable regarding the topic here. But, I already made a remark regarding the limitation of that particular test. The test itself is not what we could pass as test of intelligence or consciousness.
Possible but not necessary.
Come to think of it, who the hell do they think they are? However sophisticated the system, isn't there a culpability involved? To my mind, wherever there is an injury to a person involving a machine. there is no assumption of innocence on the side of the machine. The machine is always culpable unless there is malicious intent on the part of the person. Any shortcoming in the capacity of AI to avoid hurting or offending a human must be regarded as the intention of the maker. That, written in law, might dampen the enthusiasm of the whole AI crowd. I mean, if they can't navigate a human world how can they be permitted in it at all, except in the most circumscribed, safeguarded settings?