“The idea behind digital computers may be explained by saying that these machines are intended to carry out any operations which could be done by a human computer. The human computer is supposed to be following fixed rules; he has no authority to deviate from them in any detail. We may suppose that these rules are supplied in a book, which is altered whenever he is put on to a new job. He has also an unlimited supply of paper on which he does his calculations.” (Turing 1950: 436).So a Turing machine is intended to do the work of a human being engaged in computation.
“The reader must accept it as a fact that digital computers can be constructed, and indeed have been constructed, according to the principles we have described, and that they can in fact mimic the actions of a human computer very closely.
The book of rules which we have described our human computer as using is of course a convenient fiction. Actual human computers really remember what they have got to do. If one wants to make a machine mimic the behaviour of the human computer in some complex operation one has to ask him how it is done, and then translate the answer into the form of an instruction table.” (Turing 1950: 438).
But a computer does not need to be conscious to do this, nor does it need to have any conscious perception, nor sensation, nor understanding of what it manipulates.
John Searle has therefore made a fundamental point on this issue: the computation of Turing machines, as these are designed, is purely syntactic—there is no semantics.
If a person manipulating symbols does not understand their meaning, as in Searle’s Chinese room argument, then no Turing machine can either if it only ever engages in automatic rule-governed symbol manipulation (Searle 1980: 82).
BIBLIOGRAPHY
Searle, J. 1980. “Minds, Brains, and Programs,” Behavioral and Brain Sciences 3: 417–458.
Turing, Alan M. 1950. “Computing Machinery and Intelligence,” Mind 59.236: 433–460.
Thank you for this series of posts on the Turing machine and human thinking. As so often, I have learned a lot from your competent, intelligible and illuminating writing. Another important issue clarified that I would not have had the time to delve into in sufficient depth.
ReplyDeleteHerzlichen Dank und beste Grüße aus Deutschland
Ihr
Georg Thomas
Many of LK's posts on many topics are good. But I did grad work in this area and taught this theory at a university level. And I will tell you he is deeply confused.
Delete"So a Turing machine is intended to do the work of a human being engaged in computation."
ReplyDeleteNo.
In this passage Turing is explaining to people who have never seen a digitial computer the essence of how it works (and that essence is indeed simple symbolic manipulation.) There is not about "intent".
In modern parlance the "human computer" is the CPU and the "instructions" are software.
Searle's first mistake is to think that because the *person* doesn't understand Chinese nothing here does. This is a simple -- and frankly stupid -- confusion. When I lose at chess to a program no-one claims the *CPU* beat me. The claim is that the *program* beat me. It's stupid to contend that "the CPU just does symbolic manipulation, it doesn't play chess better than Ken, hence nothing here plays chess better than Ken."
If you are going to continue to post on this you really do need to learn some of the basic computability theory. Martin Gardiner has good book which has the theory up to the mid 50s, which is all that's required for this.
(1) the issue isn't whether chess programs can defeat you, Ken B. Nor is it whether we can design programs that cab best humans at certain tasks.
ReplyDeleteThe issue is whether they can attain human-level consciousness.
Unless you are going to tell me chess programs have consciousness, you don't refute anything here.
(2) also I am fully aware Turing is here trying to explain the Turing machine by showing what it does in *a functional sense* by his remarks on the human computer, and that computers of course have a different, given architecture (software versus hardware).
This post affords a good way to (again) expose the circular prejudice in your argument.
DeleteTuring: Look, we make digital computing machines. They manipulate symbols. Just like a human computer can do. Nothing magic here, just computation.
SearLK: Then it cannot think. All it does is syntax. Cannot be conscious, cannot.
Turing: Well you seem to have made up your mind before we even try. (And you deny that there's a prejudice here?) So let's look. There are many ways a machine *might* be conscious and able to think but let's see if they can mimic people well. So I propose the imitation game.
Time passes. Turing builds machines that fool SearLK who after 7 years of deep discussion with the computer announces it is a person.
Tyring: Nope. it's the Callahan 3000.
SearLK: Then it's not conscious. Syntax only.
Turing: OK, how do you know *I* am conscious?
SearLK: I talk to you, I can see that mental constructs are needed to understand or predict your behavior.
Turing: So why not the C3000? What's different? You apparently have a standard for telling a conscious awake person from a brain dead or sleeping one. You just applied those criteria to me. Why not the C3000?
We are still awaiting an answer to that.
Pathetic, Ken B.
DeleteThis comment makes me think you never even read my original post on Turing, nor any of my other posts.
Let us go through the list of your points:
(1) "Turing: Look, we make digital computing machines. They manipulate symbols. Just like a human computer can do. Nothing magic here, just computation."
Correct. No magic. But I never claimed there was something supernatural to this.
(2) "SearLK: Then it cannot think."
How do you define "thinking"? One of Turing's major and laughable errors is to never even properly define "intelligence". Without proper definitions, any such arguments will collapse into fallacies of equivocation.
I already made it clear that if you define "intelligence" as:
(1) a quality in human digital computers in which information processing takes input and creates output that allows action or sentences, or external behaviour of the type that animals and human beings engage in.
then, yes, computers can be defined as "intelligent". But this is not the philosophically and scientifically interesting question.
For that we need to go to (3) below.
(3) the crucial scientific issue is: what is consciousness and what causes it?
You avoid this question, just as the lazy Alan Turing did.
You can't avoid it.
You will NEVER be able to declare that we can create consciousness artificially unless we have a complete scientific understanding of consciousness.
The evidence, for those who want to look, suggests more and more that consciousness is something that is causally dependent on certain biological/physical processes.
Turing machines do not duplicate these processes, any more than they can duplicate digestion.
(4) "SearLK: I talk to you, I can see that mental constructs are needed to understand or predict your behavior... You just applied those criteria to me. Why not the C3000? "
Where did you pull this straw man B.S. from?
I have very good reasons to think other people are conscious *NOT* just because of external behaviour, but fundamentally because of internal causally necessary processes that produce consciousness that we naturally both share because of biology and Darwinian evolution.
The Turing test is precisely a naïve behaviourist method. We cannot simply rely on external behaviour.
Incidentally, as I said on previous posts (but which you ignore), it may well turn out, once we have understood human consciousness properly that, yes, we might be able to create new and synthetic technologies that reproduce consciousness and that will be true and consciousness artificial intelligences as we are, because they will duplicate the causally necessary physical processes that produce consciousness.
DeleteEven some of the supporters of the Conscious Electromagnetic Information (Cemi) Field Theory of consciousness accept this possibility:
"The ... electromagnetic field theory of consciousness predicts that in principle, consciousness could be generated using hardware instead of wetware. This single prediction not only renders the electromagnetic field theory testable where the neural identity theory is not – it also opens wondrous vistas with respect to possible future technologies."
http://www.scholarpedia.org/article/Field_theories_of_consciousness
---------
Those artificial systems won't be simple Turing machines, however.
"Searle's first mistake is to think that because the *person* doesn't understand Chinese nothing here does.
ReplyDeleteIt is not a mistake. If a human being cannot understand Chinese symbols he manipulates by a rule book, nor can a computer. In fact, we have no rational reason to think computer have any sensation or perception or consciousness of any kind.
Rather, we have much better reasons to think consciousness is an emergent property from biological processes rather like photosynthesis or digestion.
I propose an additional thought experiment to John Searle's Chinese room.
ReplyDeleteAssume John Searle learns to fluently speak Chinese. Now he can enter another Chinese room and answer questions posted to him in Chinese with written answers. He does this now without following any kind of instruction book or using a file system.
We must conclude that if this room experiment is a valid test of intelligence or consciousness (that is its possible for a machine to display intelligence or consciousness via this test) that John Searle is fluent enough in Chinese to demonstrate his properties on intelligence or consciousness (or both).
But now both the original and modified Chinese rooms display the same proficiency in fluently speaking Chinese and both contain the same basic parts (most importantly John Searle). Also we didn't conclude John Searle is not intelligent or conscious as he doesn't originally speak Chinese.
We may only conclude that (if this test can be passed) both Searle and the machine demonstrably pass it.
"But now both the original and modified Chinese rooms display the same proficiency in fluently speaking Chinese and both contain the same basic parts"
DeleteNo, that is a bizarre non sequitur. You are simply proving how stupid are the crude behaviourist tests are.
The point of this juxtaposition is that if we think about what has been added here its information (an internal representation of Chinese dialect in John Searle's mind). Of course the internal representations imagined do not have to correspond to the real world, that is there is scope for imagination, logical truth and falacy. The only demonstration and effectively existence of mental states is in a form where they are used to solve intelligence questions. Either its an ability test similar to Turings or nothing can demonstrate it passes it (ever).
ReplyDeleteThis is not like a digestive tract where simulation is demonstrably not digestion.
SOmeone more patient than I will need to take over straightening LK out. Nic is doing fine work.
ReplyDeleteMeanwhile http://kenblogic.blogspot.com/2016/04/lk-on-computers-history.html
In none of my posts do I deny computers can perform those things, which can be done by mere symbolic manipulation by rules.
DeleteYou evade the issue: what is consciousness and what causes it?
Once more unto the breech.
DeleteYou have made two claims LK. Both are wrong, andf borderline silly, but that's not my concern here. My concern in this note is the implications of the pair of them.
You claim
1) no symbol processing system such as a Turing machine can ever be conscious because it's all just syntax
2) the Turing Test is a hopeless test of whether machines can be conscious.
Now the first implication of this pair is that you think that someday machines will pass the Tuuring test, and at every level of stringency.
Why? Because if no machine were ever conscious and no machine could ever pass the TT thant the Turing test would be a *perfect* test. Perfect. Therefore you think machine will pass the most stringent Turing tests, eventually.
Now that implication has its own implications. Assuming you live long enough you will be able to subject the K5000 to an amazingly stringent TT.
You will chat with it for hours, over a period of years. Tell jokes, make puns, test its recall of all you told it, teach it New Keynesian theory.
You will debate massimmigration for years, and gradually persuade it. You will agree pomo is crap and that strawberries are overrated. You will discuss the final chapter of War and Peace but agree to disagree.
You could have it teach you the basics of organic chemistry (if it claimed to be a chemist say), and reminisce about its old mum and growing up in Blackpool.
You will nake jokes about Bob Roddis and it will respond with shots at Hedlund. You will talk to it deeply and regularly for years like an old friend. Never once will its mask slip to reveal the silicon beneath.
At the end of all that you will be convinced it is a person not the K5000.
Until you learn it cannot actually shit, whereupon you will deny it has consciouness or can "handle meaning rather than just syntax."
No, Ken B, another silly comment that just confirms you seem to have a quasi-religious, dogmatic faith in the crude external behaviourist Turing Test.
DeleteIf we want to know whether an artificial intelligence can be constructed to be fully conscious like us, the crucial questions are these:
(1) what is human consciousness? What is its proper scientific explanation?
(2) when we have a proper and proven theory of consciousness (proven to the same extent as, say, the proposition that the earth revolves around the sun), we can then answer question (3).
(3) is human consciousness a process causally dependent on and requiring certain biological/physical processes, just like other physical/biological phenomena which current digital computer lack? (e.g., photosynthesis, digestion).
(4) If "yes" to (3), Turing machines cannot be consciousness, and we have striking confirmation of the truth of the Chinese Room argument of Searle, which was a very good thought experiment pointing the way to the truth.
If "no", and the CTM is true, then presumably sufficiently sophisticated Turing machines can be made to be consciousness.
(5) I argue more and more of the evidence is pointing to "yes" as the answer to (3).
-------------------
You argue "no" to (3) and defend CTM, but actually your defence is looking more and more like a dogmatic, quasi-religious act of faith.
Even now you can't even just quietly and rationally assent that no proper and definitive answer to the question "can a Turing machine attain human-level consciousness?" can be given unless we have a proven theory of consciousness.
Instead, you bizarrely think crude behaviourist tests can definitively establish the truth.
That you call it silly proves my point: you have done no serious thinking about the Turing Test. What I described IS THE TURING TEST. The TT is not asking a box to play chess or do a regression or a date from history as you and Callahn falsely and stupidly imply. The Turing test is what I described: you talk to the box as long as you want as often as you want about whatever you want in anyway you want and try to tell if it's human or not. You get to try puns on it, see if it can remember what you tell it, see if it can learn, see if it can explain, see if it can do every mental thing humans can do. If you did not envision a scenario like mine then you did not in fact understand Turing's test. And it is clear you did not.
DeleteTell me LK, how is the scenario I explained a "crude behaviourist tests[]"? You do not understand the Turing test.
Delete(1) "That you call it silly proves my point: you have done no serious thinking about the Turing Test. "
DeletePathetic.
Yeah, yeah. I've done *no* serious thinking about it -- even though I did a cognitive science subject as an undergraduate and we did a MASSIVE study of it, with a big literature review.
Even though I just carefully analysed Turing's original article in a previous post.
But clearly I've done no serious thinking about it.
(2) "Tell me LK, how is the scenario I explained a "crude behaviourist tests[]"? "
Of course it is a bloody "crude behaviourist test" -- precisely because it is fails to examine internal states of the system to compare them with ours.
If we took a person from the 18th century and had them play chess with a computer in a hidden room with moves relayed by a person, and we asked that person: "were you playing chess with a conscious human"?, the crude behaviourist test would lead them to say "yes".
The lesson: crude behaviourist tests are insufficient to know whether a machines is conscious.
(3) it is also quite bizarre you continue to militantly reject my fundamental questions:
(1) what is human consciousness? What is its proper scientific explanation?
(2) is it true that only when we have a proper and proven theory of consciousness (proven to the same extent as, say, the proposition that the earth revolves around the sun) then we will be able to definitively answer the question "can an artificial intelligence be constructed to be fully conscious like us?".
Bloody crude behaviourist tests are insufficient to give a definitive answer. You can argue that the passing of a Turing Test by a true AI is a necessary but not sufficient condition for definitively answering the question whether it is really conscious.
But even this is debateable. In fact, a Turing test might not be a good test of consciousness intelligence even in humans. What if children and mentally impaired people were to repeatedly fail a Turing test?
You write that we cannot answer the questions "unless we have a proven theory of consciousness." Proven.
ReplyDeleteSo let's ask how it's proven? Say you present a wonderful bit of work showing how consciousness can arise from cells in concert.
Is that the proof you need? It is not. Because all it proves is that consciousness can arise from sells in concert, not that it can arise only from cells and nothing else.
So how do you prove that it cannot arise from machines? It's not clear to me how you *prove* it (so we see your entire demand is just special pleading) but at the
very least you need to look at all the machines presented to you and show they lack it. After all you insist -- when charged with prejudging -- you don't deny conscious machines are possible.
So you'd need to show -- I won't say *prove* -- any particular one isn't conscious.
And how I ask do you do that? "Oh, they aren't cells QED" is circular. "Oh they are not LIKE cells" is circular. "Oh they are not carbon based" is circular.
At some point you need to watch them do their stuff and convince people they are not conscious. "Look at what they do, that doesn't seem conscious does it?"
Now that sounds a but Turing like, doesn't it. "Look and see." (Empiricism is not behavioursm.)
The difference is Turing insists on a bias filter: you need to do it black box so that circular assumptions cannot pollute the judgment.
And you insist there should be no bias filter.
No, Ken B.
Delete(1) "So let's ask how it's proven? Say you present a wonderful bit of work showing how consciousness can arise from cells in concert.
Is that the proof you need?"
No. And I've stated again and again and again, we need more than that. We need to know if human consciousness is casually dependent on and even, scientifically speaking, identical with certain necessary physical processes. How many times do I have to tell you this?
(3) "So how do you prove that it cannot arise from machines? It's not clear to me how you *prove* it "
At a conceptual level, it's bloody easy, Ken B.
On analogy with current arguments we can now make to prove that current digital computers cannot reproduce photosynthesis:
(1) what is photosynthesis?
(2) we know, and it is proven, that photosynthesis is a process by which living things take carbon dioxide and water and produce glucose and oxygen.
(3) we now know that you *physically need* to take carbon dioxide and water and produce glucose and oxygen to reproduce photosynthesis -- this is physically necessary.
(3) digital computer cannot and do not reproduce photosynthesis in their current form. This is now proven empirically from their very structure and nature.
---------------
We will be able to make much the same types of arguments about consciousness in the future once we have a proper scientific understanding of what conciseness is.
Note carefully: none of this means we will never produce conscious artificial intelligences -- only that, if consciousness requires certain physically necessary process that Turing machines lack, current digital computers will be shown to be totally unable to reproduce consciousness.
"Even though I just carefully analysed Turing's original article in a previous post."
ReplyDeleteTwo comment.
1. I am laughing at you now.
2. You are not just ignorant but proud of it.
Apologies to Hedlund! You were right all along!