Turing, A.M. (1950) Computing Machinery and Intelligence. Mind 49 433-460
I propose to consider the question, "Can machines think?" This should begin with definitions of the meaning of the terms "machine" and "think." The definitions might be framed so as to reflect so far as possible the normal use of the words, but this attitude is dangerous, If the meaning of the words "machine" and "think" are to be found by examining how they are commonly used it is difficult to escape the conclusion that the meaning and the answer to the question, "Can machines think?" is to be sought in a statistical survey such as a Gallup poll. But this is absurd. Instead of attempting such a definition I shall replace the question by another, which is closely related to it and is expressed in relatively unambiguous words. The new form of the problem can be described in terms of a game which we call the 'imitation game." It is played with three people, a man (A), a woman (B), and an interrogator (C) who may be of either sex. The interrogator stays in a room apart front the other two. The object of the game for the interrogator is to determine which of the other two is the man and which is the woman. He knows them by labels X and Y, and at the end of the game he says either "X is A and Y is B" or "X is B and Y is A." The interrogator is allowed to put questions to A and B. We now ask the question, "What will happen when a machine takes the part of A in this game?" Will the interrogator decide wrongly as often when the game is played like this as he does when the game is played between a man and a woman? These questions replace our original, "Can machines think?"
I propose to consider the question, "Can machines think?" This should begin with definitions of the meaning of the terms "machine" and "think." The definitions might be framed so as to reflect so far as possible the normal use of the words, but this attitude is dangerous, If the meaning of the words "machine" and "think" are to be found by examining how they are commonly used it is difficult to escape the conclusion that the meaning and the answer to the question, "Can machines think?" is to be sought in a statistical survey such as a Gallup poll. But this is absurd. Instead of attempting such a definition I shall replace the question by another, which is closely related to it and is expressed in relatively unambiguous words. The new form of the problem can be described in terms of a game which we call the 'imitation game." It is played with three people, a man (A), a woman (B), and an interrogator (C) who may be of either sex. The interrogator stays in a room apart front the other two. The object of the game for the interrogator is to determine which of the other two is the man and which is the woman. He knows them by labels X and Y, and at the end of the game he says either "X is A and Y is B" or "X is B and Y is A." The interrogator is allowed to put questions to A and B. We now ask the question, "What will happen when a machine takes the part of A in this game?" Will the interrogator decide wrongly as often when the game is played like this as he does when the game is played between a man and a woman? These questions replace our original, "Can machines think?"
1. Video about Turing's work: Alan Turing: Codebreaker and AI Pioneer
2. Two-part video about his life: The Strange Life of Alan Turing: BBC Horizon Documentary and
3. Le modèle Turing (vidéo, langue française)
In his 1950 paper "Computing Machinery and Intelligence", Alan Turing addresses Lady Lovelace's objection, an argument against the possibility of machine intelligence for the following reason: A machine must only possess knowledge (and, too, the ability to do what it can) because a human has programmed it thus. It follows that making such a machine presupposes an understanding of the entire cognitive mechanism. Since we do not know how the mind works, we cannot create a machine that does exactly what a mind can do.
ReplyDeleteIn response, Turing suggests that it is not crucial to program all that (adult) minds could ever do. Rather, he understands an adult human mind to be composed of three elements: “(a) The initial state of the mind, say at birth, (b) The education to which it has been subjected, (c) Other experience, not to be described as education, to which it has been subjected.” The challenge thus becomes to create a machine that is like a child’s brain (one that can learn, much in the same way children’s do).
Certainly, if we created such a machine, we would have a solution to machine intelligence: we would have at least one model that passes the Turing test.
But, this conception of the original problem entails an understanding of what a child brain (a mind before any learning occurs) looks like. Turing presumes that the child brain is “something like a notebook”, but we know this to not be entirely true. Turing’s claim ignores innate knowledge. For example, while we are in utero we do not know what colour is, but we possess the capacity to learn this in the presence of visual stimuli. Too, we can learn to distinguish between colours. If we wished to create a Turing machine that could distinguish red from yellow, we could do that. Such a machine would point to an image and say “this (apple) is an example of a red object, and not a yellow object”.
Thus, child brain machines must, too, possess the sort of innate knowledge that the task of colour-distinguishing falls under. The child brain machines cannot be just a tabula rasa, as Turing’s claim suggests. The challenge becomes understanding what categories of things are required to be programmed into the child brain machine. Fact-learning does not appear to be of this nature, but a program that understand positive and negative response surely would be...
I would like to respond to your argument that Turing is using the Lockean "tabula rasa" idea -- that humans are born as "blank slates" -- in addressing Lady Lovelace's objection to the possibility of a learning machine.
DeleteFirstly, I do not believe that Turing is referring to the epistemological tabula rasa idea in assuming that the child’s brain is “something like a notebook”. Turing states that he hopes the mechanism of the blank notebook has “so little mechanism in [it] that something like it can be easily programmed.” Turing is not ignoring the human brain’s natural disposition to experience the world through physical sensation. He is just noting an infants lack of learning (or the lack of facts it “knows” about the world) prior to birth. Although it can be argued that we are ‘programmed’ to experience the world in color, our ability to distinguish between colors is a learned experience. One could argue a constructivist theory of knowledge, in which the child’s prior experience (in this case, learning what the color “red” looks like through experience in seeing the color for the first time), then determines the child’s learning to differentiate colors (seeing the color yellow for the first time, and using the prior knowledge of red, understanding that in fact red is different from yellow).
In the case of the color example, it is not impossible for a computer to “learn” in the same way that an infant can. One could program a computer to distinguish colors. Upon presenting the computer with the color red for the first time, it could register that an apple is red (mimicking the initial processes an infant undergoes when seeing red for the first time). Then, when presenting the computer with a banana one could ask: “is the banana red?” The computer would, using previous knowledge of what red is, would respond “no” (in the same way that a baby would recognize that yellow is not red upon seeing the color yellow for the first time after seeing red). The process of distinguishing, for both an infant and a machine, only requires experience with what the color is to then understand when a color is different.
Turing grapples with the idea of how to teach a computer, using examples of behaviorism as a method to teach imperatives (like “Do your homework”). It is there that he starts to answer the Lovelace problem in a way that even he notes can feel “paradoxical” at first. It does not ignore what you call human being’s “innate” dispositions.
I would just like to add, that I also felt like Turing neglected to address some of the innate knowledge that infants possess. Turing states that “the learning process may be regarded as a search for a form of behavior which will satisfy the teacher (or some other criterion).” Turing seems to focus on the fact that most learning requires a teacher and that behavior stems from the desire to follow sets of rules taught to us by our teachers. Turing does not address the things that children do not learn explicitly, like language. Children learn language on their own at a very rapid pace. No one explicitly teaches a child how to produce the sounds they hear adults make or how to correctly form grammatical sentences. Could a child machine have the capacity to learn things without being explicitly prompted?
DeleteTuring also talks about the ability of machines to learn via a punishment and reward teaching process. Due to Chomsky’s work, we know that language is not acquired using punishments and rewards like behaviorists (like Skinner) originally argued. In order to truly simulate a child’s learning, the education process that the machine child would have to undergo would need to include more than just reward and punishment processes. Language acquisition is a process that we do not even completely comprehend, how could we expect to simulate this type of learning using a “child machine”?
With all this being said, I agree with the point you made about the child machine also needing to possess some type of “innate knowledge” in order to then be able to learn using an education process described by Turing. In my opinion, the child machine would need to start off with certain things, that children are able to learn on their own, in order to be able to begin learning.
I disagree that Turing ignored innate knowledge. On one hand, Turing’ machines are explicitly described as having innate baggage, i.e. a set of rules and symbols kept in their “store” component. This knowledge was provided by their programmer when they were conceived as our brain already has the capacity to learn when we were born. The only difference is that the apparatus to learn is provided artificially to the machine while ours comes biologically. Both requires stimulation in order for “learning” to occur. Children that grow up without constant stimulation from their parents will suffer from cognitive delay as much as a machine kept in the corner of a room will not learn new stuff by itself. The real problem is that Turing’s original machine only has “definitions and propositions” in his store component which only allows for verbal symbolic learning. However, we know that sensorimotor processing in a huge part of cognition. This is why performance on the TT3 is better suited than the original test to assess such learning capacity.
DeleteOn the other hand, what Turing machines don’t have at the moment they are conceived is more of a meta-knowledge: knowledge about what it knows and how it knows it. But that goes far away from the original Turing Test. This brings us back to the Other Minds problem which Turing does not intend to solved nor to addressed in his test. Speaking of verbal and sensorimotor capacities solely, a machine that is simulated the right way, that being referred to as the “education processes” by Turing, and possessed the right baggage in its storage unit will have no problem learning at all.
The first skywriting was from Paul Ashkenas:
DeleteWhat is “programmed” is a general algorithm for learning. That can include built-in (unlearnt) capacities too. The “tabula rasa” is a simplistic misconception. And the Lady Lovelace objection misses the fact that an input/output algorithm is changed by its input (and the “programmer” cannot anticipate what the input will turn out to be).
Gili: Remember that the TT is just about doing, not feeling, i.e., input/output capacity. So color identification and discrimination would be “optical processing,” not “seeing.”
Amanda: Learning can be by trial and error with corrective feedback coming from the consequences of whether one has done the right or wrong thing (as in learning which mushrooms are edible). Learning from verbal instruction is more advanced; first you have to learn categories (what to do with what), including their names.
We’ll talk about Chomsky later, but that concerns a special kind of grammar, Universal Grammar, UG, which seems to be unlearnable, so it is one of the capacities that is built into the learning algorithm.
Chloé: Although he puts it in terms of verbal instruction, Turing was already well aware of the potential of trial/error learning with corrective feedback from the environment. (By including robotic sensorimotor learning capacity, T3 already leaves the symbols-only world of T2 behind.)
I agree that learning by trial and error with corrective feedback is an important method of learning that the child machine would need to implement. Children, as well as the child machine, can learn a lot from the corrective feedback they receive from the consequences of whether one has done the right or wrong thing.
DeleteAs Professor Harnad has pointed out, learning from verbal instruction is indeed more advanced, but I believe it would be necessary for the child machine to be able to do if Turing is really trying to replicate a child’s brain. Children receive a vast amount of verbal input from their parents. Learning from verbal instruction is necessary and important for children. I believe that learning from verbal instructions is more advanced because it could be tied to the hard problem of cognitive science (how and why we feel what we feel). We often use language as a means to communicate how we feel and to explain to others why we feel certain ways. Getting a child machine to learn using verbal instruction could be more challenging because there are often emotions and feelings tied to the things we say. For example, a child may also learn not to eat poisonous mushrooms when he goes out in the backyard to play when his parent sternly instructs them not to pick up and eat any mushrooms that are look a certain way or else, they will get extremely angry. Children are able to pick up on intonation and cadence, which gives them insight to how a parent could be feeling while giving them verbal instructions. This would also cause the child to feel a certain way when receiving verbal instructions impacting the way the instructions are given. A child may take verbal instructions more seriously when they sense that their parent is upset or serious.
Finally, after further thought I am able to accept that Universal Grammar (UG), something that is currently seen as unlearnable, can be a capacity that is built into the learning algorithm. Although a small part of me still wonders whether Universal Grammar can actually be learned by children, but we simply aren’t capable of discovering how it is done right now.
“The machine has to be so constructed that events which shortly preceded the occurrence of a punishment signal are unlikely to be repeated”
ReplyDeleteIn this section, Turing is speaking of a child machine and of the possibility of a machine, a digital computer, to be able to learn as a child learns. The computer cannot learn typically as a child does as it has potentially no eyes and no legs; however, it can still learn through reward and punishment. When some children are punished, rather than stopping this behavior entirely, in some cases, these behaviors require multiple reminders to be removed or are perhaps never removed at all. Would a child machine being taught immediately stop the behavior being punished or could it replicate the slow progression of learning? And if it does stop immediately to observe whichever behavior is being punished, can it really be said to do whatever a “human machine” can do?
I also found the section in Turing’s “Computing Machinery and Intelligence” on learning machines, and specifically the idea of simulating a child’s mind and its development as opposed to that of an adult very interesting. In trying to figure out a response to your questions about the learning process of a child machine, I wrestled with a few ideas. The first is under the assumption that Turing machines have unlimited storage. I would think that the child machine would only need to be reinforced once before it learns that specific rule. With this being said, a human child brain does not have an unlimited memory let alone the amount of an adult so giving a child machine an unlimited memory would not be a realistic representation.
DeleteA difficulty I see with the child model, would be in the learning process itself. If the child machine had an infinite storage capacity then it would be able to learn “right from wrong” within one experience, given that there is either a punishment or reward associated with it. The issue lies with the fact that humans tend to generalize rules over many scenarios. For example, if a young child “x” punches their sibling “y”, their parents will punish them for this act. In normal circumstances, the child “x” will generalize that it is bad to punch people and won’t try punching their other sibling “z”. A computer however, as it learns everything on a case by case basis, will likely attempt to punch “z” even though it learned that punching “y” was wrong. Therefore, from this model we still won’t be able to understand how we apply our knowledge.
Additionally, Turing mentions that “one might try to make [the child machine] as simple as possible consistently with the general principles” or “one might have a complete system of logical inference built in” (Turing, 1950). I think that it would be important to have some sort of logical system built in, because throughout evolution, humans are innately born with a certain level of intelligence. For example, the paper Face Perception During Early Infancy shows that newborns as young as 53 mins prefer face like images as opposed to non-face like images (Mondloch et al. 1999).
I agree with the general idea of your commentary, but I don’t see the issue of child X as insurmountable. Of course, a computer that has an unlimited storage capacity could be thought to be infallible. However, it isn’t the case, and Turing defends this point of view himself in an earlier passage of his essay. He says: “The claim that "machines cannot make mistakes" seems a curious one” (Turing, 1950) and goes on to claim that a machine that is trying to imitate the way of learning or the way of reasoning of an adult human in the imitation game would “deliberately introduce mistakes in a manner calculated to confuse the interrogator” (ibid.). I think it is important to understand that, even though the computer can theoretically solve any kind of arithmetic problem, it doesn’t mean it will.
DeleteThere are even two different ways that I see an “infinitive capacity computer”, as Turing calls them, could successfully replicate the learning of a child: (1) the computer could be programmed to give itself only access to some amount of its storage space based on the age of the human it is trying to imitate, and to increase this access by itself at set times in the development of the child and (2) the computer could purposely commit mistakes to emulate the reasoning of a child (like overgeneralization, to come back to your example of child X).
Marine: Turing was just referring here to operant/instrumental (behaviorist) learning by trial and error, with corrective feedback from having done the wrong or right thing. (Think of mushroom sampling, with feedback from feeling sick or nourished afterward.)
DeleteAkhila: Infinite storage is not relevant (we don’t have it either). The learning algorithm has to learn to generalize across the same and different kind of input. I.e., it has to be able to learn to categorize (and for nontrivial categories this needs many examples, not one.
Marianne: An unlimited storage capacity is not necessarily an advantage! (Look ahead to Week 6a and read Borges’s "Funes the Memorious"
http://vigeland.caltech.edu/ist4/lectures/funes%20borges.pdf
and look up “hypermnesia” and Luria’s "The Mind of a Mnemonist"
http://arteflora.org/wp-content/uploads/2018/05/Luria-The-Mind-of-a-Mnemonist.pdf
In this paper, Turing reasoned his idea that machine can actually think and learn, like humans do, by first proposing the imitation game, and then developing his theory based on this game step by step, also refuting several objections that others may hold, and finally arriving at his conclusion that machine can learn.
ReplyDeleteFirst of all, Turing was indeed a giant who had incredible foresight and sagacity. Numerous arguments he presented at the end of this paper are exactly what is happening right here right now, in the field of computer science, more specifically, in machine learning. The idea of introducing a random element and punishments and rewards into the learning process is precisely what reinforcement learning relies on. And certain randomization used in regular algorithms has long been proved to be very helpful in cutting the time costs. Based on my understanding, reinforcement learning is mostly based on Turing's description of the teaching process to the machine. And it's indeed analogous to the real evolution, at least in a way, but more expeditious because the "unnatural selection" of what to keep and what to avoid is not random anymore, but also sort of intelligent.
With machines already learning, I believe we've really come to a point that we may need to worry about what if they really grow much more intelligent and powerful than human. I totally agree with Turing that "machines will eventually compete with men in all purely intellectual fields" -- and at least it has already been proved in Go. Probably Turing has never been upset about this since at that moment, it still seemed to far in the future (actually his anticipation was by the end of the last century). But now, as he said, "an important feature of a learning machine is that its teacher will often be very largely ignorant of quite what is going on inside, although he may still be able to some extent to predict his pupil's behavior.", which is so true that for most of the problems that have been satisfactorily solved by deep learning solutions, we can no longer be crystal clear of what the thinking process the computer has gone through to finally get the answers. For instance, AlphaGo, which played a move that not a single person would have never expected, and won the game. Well, "he" might have stumbled on the move by just random search, if I may guess, but random search do not explain all the cases.
Furthermore, also as Turing pointed out, "Another important result of preparing our machine for its part in the imitation game by a process of teaching and learning is that 'human fallibility' is likely to be omitted in a rather natural way", which immediately leads me to the question: then who is to blame? This is such a complicated but significant problem that must be solved first before any self-learned machine can be really released and functioning on their own, otherwise once something dreadful happens, the assignment of liability will instantly run into a mess.
At the end, I actually have a question which I still cannot fully understand. Near the beginning, Turing said, "May not machines carry out something which ought to be described as thinking but which is very different from what a man does? This objection is a very strong one, but at least we can say that if, nevertheless, a machine can be constructed to play the imitation game satisfactorily, we need not be troubled by this objection." I'm not grasping why we need not be troubled by this objection if a machine can perform indistinguishably as a man.
The point of the Turing machine is not that it *be* identical to a human (which would indeed be an outrageous requisite). The requirement of the TM is just that it can achieve the same goal: it need not be constructed of the same materials (in Harnad's words, we are not looking for a T4 or a T5 machine) but it must be able to simulate human capacity (T3).
DeleteI'm not sure I entirely understand the issue that you are taking... If a TM could perform to the level of a T3 robot, the problem of machine intelligence would be dealt with: we would have (at least) one successful model for cognition in some sort of dynamical/computational system. The process by which the machine "thinks" may very well be completely different from human thinking.
The question is... If it can do all the things we can (if we can simulate human capability), does it matter that we don't have *the* explanation?
“At the end, I actually have a question which I still cannot fully understand. Near the beginning, Turing said, "May not machines carry out something which ought to be described as thinking but which is very different from what a man does? This objection is a very strong one, but at least we can say that if, nevertheless, a machine can be constructed to play the imitation game satisfactorily, we need not be troubled by this objection." I'm not grasping why we need not be troubled by this objection if a machine can perform indistinguishably as a man.”
DeleteI think the idea behind Turing dismissal of this (reasonable) objection is related to the very idea of the “imitation game” (aka the Turing test), which is only a “weak equivalence” test. Indeed, if the input-output processes are the same in the computer and the human, the computer has done its job. Prof. Steven Harnad pointed out in “The annotation game” that the test is not about thinking “as such”, but about performance capacity of a machine. The test is worried about the results, not the recipes.
To add to your comment, I also think that the matter of the importance of the objection is related to what we want to do with the Turing test. If you want to ask the question “Can a machine perform like a human?”, then it is right that we do not care if it’s not **the** explanation, as long as the machine can pass all the tests we want (up until T3 as it was suggested in the skywritings). But, the question is still relevant if you flip the question on its head and ask “Is human thinking just like any other machine thinking?”, i.e. “Is computationalism true?” Then, the question about **the** explanation is relevant, because it would tell us how and why humans act and think this way (the “easy” problem of cognitive science). Nevertheless, because of the underdetermination problem (there can be more than one way, more than one algorithm/recipe, to imitate perfectly humans at the level of T3), I doubt that having a machine succeeding the T3 would give us any real insight about human thinking.
There is simply no way to prove that the explanation is **the** one. Thus, to check if humans are “just like” machines, you need to understand what humans do without already assuming that it is like a machine (because it is precisely what do you want to check). If you can’t use computation ideas for the easy problem, then the question about the similitude between humans and machines is not as relevant as computationalism would like to see.
Junlin, in a model, "selection" is of course unnatural, because in real evolution and learning it is the external world that "selects" (based on the consequences of doing the right or wrong thing). That sense of "unnatural" does not imply that the model is incorrect. A learning algorithm inside a T3 robot is really learning what to do (in the world).But "intelligence" (the capacity to do and learn to do the right thing) is a feature of the learner, not the world (except in the zpecial case of verbal instruction [i.e., the power of language] -- but we are not there yet, in machine trial-and-error learning algorithms, whether unsupervised or "supervised").
DeleteTuring did not predict that the TT would be passed at the end of the last century -- just that by about the turn of the centuray 7 out of 10 judges would mistake a TT model for a real human if testing purely verbally for 10 minutes. (That is the idea behind the Loebner Prize, but it's not the TT!)
Can you elaborate on why you think it is a cause for worry if computation can do things human brains cannot? And if a T3 is more intelligent than most or all people, is there more reaon to worry than if a human is more intelligent than most or all people?
In general, random (as well as exhausitive) search is not thought of as "intelligence" but as brute force.
The question of whether we should trust computations is not the same as the question of who is to blame. If I trust a newspaper article, or Trump, and the information turns out to be wrong, then I am to blame for trusting it.
Turing's reflection about human-like and non-human-like thinking is easily settled once we remember that thinking is only thinking if it feels like something to do it (hard problem). Otherwise it's just computation i.e., (doing; the "easy" problem). If there is some sort of internal process that is unlike human thinking, but that produces results exactly like what thinking does, and it feels like something to do it, why would you want to find anther word for it than "thinking" too?
Paul, a TM (Turing Machine) cannot be a robot, whether a full T3 robot or just a Khepera. A robot is a dynamical system. It does physics and physiology ooo; and it is not just the hardware of a TM.
Antoinette, yes, the question about which way of doing what thinking does is "right" is rather like asking for strong rather than "just" weak equivalence, and that in turn is related to the "problem" of underdetermination. The question is: how and why does it matter, if it does not make a difference to what a system can do (and we cannot be any the wiser, because of the other-minds problem -- which Turing explictly bracketed, restircting himself to the "easy" pronlem)?
“The new problem has the advantage of drawing a fairly sharp line between the physical and the intellectual capacities of a man [...] May not machines carry out something which ought to be described as thinking but which is very different from what a man does? This objection is a very strong one, but at least we can say that if, nevertheless, a machine can be constructed to play the imitation game satisfactorily, we need not be troubled by this objection.”
ReplyDeleteThe notion that the ‘fairly sharp line’ Turing is referring to cleanly divides intellectual from physical capacities seems uncertain. Evaluating an entities success in holding up its end of a casual conversation fails to test its facility in other important domains; that it might fool some human invigilator says little about whether it’s brand of thinking could generalize to solve problems and do things that brains can but digital computing hasn't yet. I’d be hard pressed to say my ability to essentially catfish someone into thinking im human via text --to thrive on one side of the line Turing drew here-- employs more than a smattering of the meaningful activity brains perform, what this problem Turing is starting to tackle is after.
If I understand your comment correctly, your objection to Turing’s argument is his proposal to replace the question “Can machines think?” with “Are their imaginable discrete state machines which would do well in the imitation game?”. You state, “I’d be hard pressed to say my ability to essentially catfish someone into thinking I’m human via text -- to thrive on one side of the line Turing drew here -- employs more than a smattering of the meaningful activity brains perform, what this problem Turing is starting to tackle is after”. In class, we discussed the four levels of Turing tests. T0 is a machine that simply perform an arbitrary task, such as playing chess. T2 is a machine that is indistinguishable from a human through messaging. T3 is a machine that is indistinguishable in sensorimotor performance capacities, such as talking, reasoning, and looking like a human. Lastly, T4 is a machine that is physically identical to humans with neuro-behavioral equivalence. If I understand, your objection is that even if a machine can pass T2, you do not believe it is enough to say it can do what humans do when they think. I don’t believe this objection is really relevant to the heart of Turing’s argument. Turing asks not whether there is a machine that can pass the imitation game, but whether we can conceive of a machine that can pass the imitation game. We can conceivably imagine a machine that looks, speaks, and behaves like a human, and thus can pass T3. Turing only discusses T2, I believe, to solve the problem of appearance differences between machines and humans. So the true question that can be proposed from Turing’s paper is - if a machine and a human are indistinguishable in sensorimotor performance capacity, can we say it can think as humans do?
DeleteKevin, the TT is not "catfishing." imitation, or a game. It is the reverse-engineering task of cognitive science (according to Turing) of finding a mechanism (whether purely coputational or other) that can generate the capacity to do anything and everything a human can do, indistinguishably from a human, to a human, for a lifetime.
DeleteThe dividing line is not between the intellectual and the physical (that's just the software/hardware distinction). It is between (EP) doing , whether by computation or by dynamics, or both and (HP) feeling.
Taylor, Turing's test is not about conceiving whether we could build something that can pass the TT, but about actually building it. T2 connot do everything T3 can, so in that snse it's not enough. (But "Stevan says" only a T3 robot could ever pass T2 -- because of the Symbol Goruning Problem, week 5).
“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. Constructing instruction tables is usually described as "programming." To "programme a machine to carry out the operation A" means to put the appropriate instruction table into the machine so that it will do A.”
DeleteHere Turing explains the process of reverse engineering necessary to build a machine that is behaviorally equivalent to a human being. Reverse engineering could go both ways; if we figure out how the brain does everything it can do, we can then wire a computer in a similar way OR if we end up creating a machine that can do everything we can do, we can use that information to reverse engineer human cognition. Since our focus is on explaining the causal mechanisms of human cognition, the latter is the obvious choice. I’m assuming that in this case, the “human computer” Turing speaks of refers to the human brain/whatever it is that allows humans to do what they can do.
However, in order to build a machine that can do everything that we can do, we first have to know all that we can do. This seems obvious, but I think there is still much research to be done surrounding aspects of human cognition that happen underneath the surface of consciousness (or even if we are conscious of it, we don’t quite understand). I think this is what Turing means when he says that having a “book of rules” or any set number of operations for a human computer is “a convenient fiction”.
Re: “Reverse engineering could go both ways; if we figure out how the brain does everything it can do, we can then wire a computer in a similar way OR if we end up creating a machine that can do everything we can do, we can use that information to reverse engineer human cognition.”
DeleteI don’t think that’s exactly true. A major problem of reverse engineering is the problem of underdetermination – we can understand the processes behind a functioning system, but that is just one possible explanation. So if we manage to create a machine indistinguishable from humans (which is very ambitious and would rely on a great deal of luck to accomplish if we don’t explicitly set out to do so), it would only explain how the machine does it, and transferability onto humans is questionable.
Going the other way around, understanding how the human brain does what it does (which is another ambitious undertaking) does not mean we would be able to replicate it and “wire a computer in a similar way.” Furthermore, saying that understanding the brain means understanding cognition implies rejecting a mind/brain dualism and arguing that mental life is the result of brain activity exclusively. While that may be true, we have yet to explain how and why things feel the way they do.
“Thinking is a function of man's immortal soul. God has given an immortal soul to every man and woman, but not to any other animal or to machines. Hence no animal or machine can think. […] Should we not believe that He has freedom to confer a soul on an elephant if He sees fit? We might expect that He would only exercise this power in conjunction with a mutation which provided the elephant with an appropriately improved brain to minister to the needs of this sort”
ReplyDeleteIn this argument, from what I understand, Turing is saying that if God can grant humans an immortal soul, then it is within His power to grant a soul to an animal, provided it developed cognitive capacities similar to that of humans. In a similar fashion, a machine should be worthy of a Soul too. If God would grant a soul to an animal if it had the cognitive capacities of a human, then it would grant a soul to a machine with the cognitive capacities of a human as well. Thus, if thinking is a product of the immortal soul, then a machine worthy of a soul can also think.
However, I think that Turing overestimated how easily theologians would accept the premise of his argument that God would grant an elephant a soul if it were to develop sufficiently advanced mental abilities. In fact, in the "Heads in the Sand" Objection, Turing himself states “We like to believe that Man is in some subtle way superior to the rest of creation. […] The popularity of the theological argument is clearly connected with this feeling.” Thus, he is aware of religious propensity to elevate humans to status different than those of animals. A religious person would not agree with the idea that God would grant a soul to an animal, much less to a machine, regardless of its cognitive capacities. He would argue that Man was created in the image of God, which is what makes us different from animals first and foremost, not cognitive capacity.
While I don’t agree with the theological objection, it seems clear that supporters of the theological objection would not be convinced by Turing’s argument.
The "soul" is a red herring here. The real issue is feeling. (That, and the hard problem, is the origin of the notion of "soul" anyway.) And theology is irrelevant... Turing just wants to point out that there is no reason to believe that entities other biological ones could feel.
Delete“A better variant of the objection says that a machine can never "take us by surprise.” […] Machines take me by surprise with great frequency. This is largely because I do not do sufficient calculation to decide what to expect them to do.”
ReplyDeleteI agree with Turing’s response to the objection that machines cannot create anything new or take us by surprise. Turing makes it clear that due to his own lack of knowledge or preparation, he is often taken by surprise by machines. In my opinion, that machines cannot surprise us is a strange objection, because being taken by surprise doesn’t always require an action on the part of the surpriser, but just the absence of information, or the presence of false information on the part of the surprisee. For instance, I can be surprised by an inanimate object if it is not in line with my expectations. I can pick up an empty milk carton, while believing it to be full, and be surprised by its light weight.
However, I am inclined to believe that the objection is more about a machine surprising us by doing something new. It seems like it is an objection against creative thinking for a machine. Although, it is not a claim that machines can’t follow a set of rules and perform according to those rules, the objection, in my opinion, revolves around divergent thinking. That is, that machines cannot think outside of a set of rules.
However, I would disagree with that as well, because divergent thinking involves framing a problem or concept differently than you have currently been framing it. So, assuming that a machine has a set of steps to follow to solve a problem, but that these steps aren’t sufficient to solve the problem, it can be programmed into the machine that if the problem cannot be solved using sequence of steps X, for example, adjust and solve the problem using sequence of steps Y, and so on and so forth. This is similar to when humans have the proverbial “Aha moment”. It simply involves cognitively reframing the way you are looking at a problem, which machines should be capable of doing if programmed to do so. Thus, if a machine were to reframe a problem in a way that no human has though of before, it will have effectively surprised us.
"Surprise" here partly means that you could not have predicted what TT would do even though you had written TT's algorithm.
DeleteThe first answer is that even if you write an algorithm, you may not know everything that algorithm can do -- and especially not if it's a learning algorithm, which changes its own state based on what it has learned, because you cannot predict what its future input history will be.
But another intuition behind these thoughts about originality or creativity is that inasmuch as what a TT does is determined by rules (whether or not the designer can predict where they will lead), the rules themselves determine where they will lead, and in that sense there can be nothing really new.
Well, either that's wrong too, and the right answer is that future inputs are still unpredictable and not determined by the rules -- or, if everything is rulefully determined, whether by the TT's algorithm or the Big Bang that started the causal chain of everything in the universe, then there can be no such thing as creativity or originality, either by TTs or by people.
That is of course nonsense. There's always chance events that send things in a different direction (including inside the head of a person or a TT). And "surprise" can be taken to refer to predictability itself, so that if an unpredicted, unpredictable outcome occurs, it is suprising even if it was all a nonstop determinate causal series of events ever since the Big Bang. (That too is wrong, because of quantum mechanics and even statistical physics.)
The bottom line is that the overall, generic chances for originality are no different for people and TTs.
Though they are definitely different for giants and pygmies....
It's easy to see that Turing is a giant. But I find it interesting that his argument, as he admits, is mainly shooting down counterarguments. I rather enjoyed that style. At two points I had ideas which I think merit a skywriting. First, "I believe that in about fifty years' time it will be possible, to programme computers… to make them play the imitation game so well that an average interrogator will not have more than 70 per cent chance of making the right identification after five minutes of questioning." This reminded me of the Turing Test competition held today, where the winner has to fool 75% of judges, like we've discussed in class, after 30 minutes. Now, this obviously isn't the life long T2 passing pen pal that the test is supposed to be, but I feel like the idea is similar to this quote. Turing seems to set the bar pretty low, with machines having a 30% success rate after 5 minutes, so in present day we've stepped it up a bit. But I also enjoyed reading Turing hypothesize about machines of the future and their capabilities; guesses at technology of the future in any era are fun to read, either for their astounding accuracy or for how ridiculous they seem. Secondly, in Lady Lovelace, Turing's counterpoint is that nothing is new under the sun. Thus, one can't say anything is original, invalidating Lady Lovelace's point that machines can't create anything new. My initial reaction I think is mostly rooted in my inexperience in pointing out fallacies. My first thought was that Turing was being rather dismissive as his point could be said for anything regarding originality. Of course everyone has their minds molded by their teachers, which could lead to an incredible invention or innovation. I don't agree that using those teachings invalidates any originality of thought. Though, at the same time, I do understand to a degree that he is using that idea to effectively counter the Lady Lovelace point. My qualm I feel lies more in the general dismissiveness that can be applied to any point rather than his use of it in this instance. All this said, Turing is a giant, so I'll take care to not wander under his shoes.
ReplyDeleteThe Loebner Prize is certainly not the Turing Test; it's just a test of how much progress we are likely to make in about a half century -- and Turing was pretty close to the mark...
DeleteSee the reply to Johnny, above.
"Most actual digital computers have only finite store. There is no theoretical difficulty in the idea of a computer with an unlimited store. Of course only a finite part can have been used at any one time. Likewise only a finite amount can have been constructed, but we can imagine more and more being added as required. Such computers have special theoretical interest and will be called infinitive capacity computers."
ReplyDeleteAlthough I agree that digital computers have finite store or “memory capacity”, and if we are comparing digital computers to human computers, the latter likewise have finite store. However, I am not sure that we can compare digital and human computers in terms of their ability to access or recall stored information.
A digital computer can be provided with as much information as long as it doesn’t exceed the memory capacity. It can also access that information at any given point. In other words, as long as the digital computer has already been provided the information, it can access or recall that information; the digital computer doesn’t “forget”. However, with the human computer, we can provide it with as much information (again, as long as it doesn’t exceed the memory capacity). But its ability to recall that information depends on how well it remembers the information long-term. Human computers seem more prone to forgetting than digital computers, unless I am misunderstanding the concept of a human computer.
That being said, isn't the act of remembering information related to (or a part of) “thinking”? Although some things can be remembered more “automatically” by humans (e.g., naming your third-grade teacher), some information might require thinking in order to be remembered. Would it not be the case that all a computer does is automatically access information once provided? Can we call this “remembering” or is it simply a storage of information that can be generated once prompt? If we can’t call this remembering, is the computer even thinking? Or does it not matter because of (weak) behavioural equivalence? And in that case, does it matter if a computer can think at all?
Humans are not computers. And we don't yet understand how human's remember. In principle, computer memory can always be increased -- by adding more memory cards and more computers. Finding data is a matter of processing time.
DeleteTuring proposed:
ReplyDelete"Instead of trying to produce a programme to simulate the adult mind, why not rather try to produce one which simulates the child’s? If this were then subjected to an appropriate course of education one would obtain the adult brain".
If this were the case, we may not be certain that the child-program would yield the types of errors that a human child would. I tend to think that a child’s errors are more often based on "logical" confusions (i.e., you can logically see where they went wrong) rather than randomness. To counter my own argument, I think it would be interesting to modify the Turing test such that the interrogator would be a teacher, and they would have to distinguish between a child-program and an actual child. If the criterion to “pass” the test is weak behavioural equivalence, then I would expect the child-program to learn as a child does and essentially "do" what a child does through verbal (typewritten) communication.
Passing TT means being able to do anything (cognitive) that a human can do, indistinguishably from a human. The same would apply to a child-TT.
DeleteAnd much of learning is not by verbal instruction but by trial and error and feedback.
“An interesting variant on the idea of a digital computer is a "digital computer with a random element. (…) Sometimes such a machine is described as having free will."
ReplyDelete“It is probably wise to include a random element in a learning machine. A random element is rather useful when we are searching for a solution of some problem. (…) Since there is probably a very large number of satisfactory solutions the random method seems to be better than the systematic.”
“Intelligent behaviour presumably consists in a departure from the completely disciplined behaviour involved in computation, but a rather slight one, which does not give rise to random behaviour, or to pointless repetitive loops.”
Above are three paragraphs related to the random element. What I find very interesting is that there seem to be an inconsistency in the function of a random element. On one hand, a digital computer with a random variable is sometimes as having “free will” (although Turing himself would not use this term), and in a later example, “a computational problem” in Horswill’s words, a random element is said to be “useful” as it provides a “better” solution than the systematic approach. It seems that it is the random element that makes the “machine” more “human-like”.
On the other hand, it is argued that intelligent behaviour presumably deviates a bit from the completely disciplined behaviour involved in computation yet it does not give rise to random behaviour. Here, randomness is not considered as a “human characteristic”. In my opinion, it is randomness that makes human “more human-like”.
It reminds me of my comment for “What is computation?”, even though I was a bit confused with computation and computationalism, what I was thinking was exactly this random element that seems to model a human behaviour with a better approximation. Of course, human behaviours are not “completely random”, but what is the reason why we choose to do one thing yet there are many other options at the same time, which may not be equally possible, and for another time the choice may be completely different? There are a lot of factors in our decision-making process, but is the random element one of them?
I agree with your sentiments on the seemingly contradictory purposes of the random element. In addition to the distinct functions you mentioned in your post, I was confused by the concept that a random element "does not give rise to random behaviour". I may be misunderstanding Turing's explanation here, but doesn't the very nature of the random element, embedded in a machine's program, imply some degree of unpredictability? I would assume that as a result, once in a while, the machine will produce an output that is inconsistent with what we may otherwise expect.
DeleteFrom last week's reading, I interpreted computation as the process that produces a set of outputs from a set of inputs - resembling a recipe or instructions that may produce a predictable result. Of course, as Turing stipulated, "intelligent behaviour" or human nature is not completely disciplined, but if computation is as we defined it in class, we can more or less anticipate what will happen if an ingredient were missing or if salt was added instead of sugar. However, we know that the random element "is rather useful in searching for a solution of some problem" and it does not give rise to "pointless repetitive loops".
Therefore, as a continuation of these points and in conjunction with Turing's assertions, could someone clarify that the random element be characterized as "human fallibility"? It may be impossible to discern or predict the "mistakes" made as a result of the random element, but the behaviour of the machine in general still follows its intended programming.
Turing computation is rule-based and completely determinate; variants can have probabilistic or statistical rules, and in some cases these can do things deterministic rules cannot. Chance (whether internal or external) can also joggle things in a (by chance) more fruitful direction than determinate rules.
Delete"Another important result of preparing our machine for its part in the imitation game by a process of teaching and learning is that "human fallibility" is likely to be omitted in a rather natural way, i.e., without special "coaching.""
ReplyDeleteTuring suggests that learnt processes naturally do not produce 100% certainty of result. As this property permits unlearning, he asserts that teaching 'human fallibility' to a machine is a necessary condition for it to be capable of passing the imitation game. While Turing admits that he is unsure of how to program this child learning machine, I argue that the directions he does propose conflicts with the condition that human fallibility be taught "in a rather natural way, i.e., without special "coaching". For example, he proposes incorporating 'unemotional' channels of communication (i.e., reward and punishment) to expedite the process of learning between teacher and machine. For the child program itself, Turing plays with the idea of integrating a complete built-in system of logical inferences. Both these proposed components function to bound the machine to display a specific output the next time it is given the same input. How then can the machine be capable of mistakes in a way that fulfills the condition of "human fallibility" without special coaching to combat this rigidity? While it is not clear (to me) that he links this issue of human fallibility to his recommendation to include a random element in the learning machine, could adding stochasticity to the system via this element serve as a potential solution?
Reinforcement (or "supervised") learning is based on trial and error and corrective feedback from the consequences of having done the right or wrong thing. Both humans and other animals are capable of this kind of learning.
DeleteLearning by verbal instruction comes only later (and only in humans).
A few stray comments about this paper: (1/2 - my comment was too long to publish in one post)
ReplyDelete1. Turing mentions Laplace’s argument: “… This is reminiscent of Laplace's view that from the complete state of the universe at one moment of time, as described by the positions and velocities of all particles, it should be possible to predict all future states.”….
Does the idea of the universe as a program or a computer simulation preclude the existence of ‘random’? The existence of ‘random’ certainly would be evidence against a hard determinism, but would it also be evidence against the possibility of a “grand unifying theory of everything” – or a program that could capture all natural systems?
That begs the question as to whether ‘random’ is expressed in nature. I could be wrong, I don’t have a background in the natural sciences, but I think we accept random expression in gene mutation and sub-atomic physics – but is “random” or a non-pattern provable, or is it a convenient explanation for patterns too complex for humans to predict, or have not yet predicted?
Turing also writes “It is probably wise to include a random element in a learning machine. A random element is rather useful when we are searching for a solution to some problem…”
Is random programmable? How can we get ‘random’ from the result of an algorithm?
It seems to me that either random must be programmable, or random cannot exist (everything must be systematizable) for all natural processes to be simulatable. (Maybe these questions have already been answered).
2. Regarding the Argument from Consciousness:
“This argument appears to be a denial of the validity of our test. According to the most extreme form of this view the only way by which one could be sure that machine thinks is to be the machine and to feel oneself thinking. One could then describe these feelings to the world, but of course no one would be justified in taking any notice. Likewise according to this view the only way to know that a man thinks is to be that particular man. It is in fact the solipsist point of view”
Couldn’t one argue that it is “safe” to ascribe consciousness to another human, (through induction) because we were born out of the same (or similar enough) processes, and from the same material, , and there is no logical reason that I should be conscious while my parents are not? (As all of my genes – that is, assuming consciousness is ‘programmed for’ in DNA are inherited from my parents, and assuming that my perceived consciousness isn’t a random mutation – which I think is a safe enough assumption, as most people report that they experience consciousness), whereas on the other hand, a machine is made from other processes, and determined by programs different than our own DNA? Is it not then sound to be skeptical as to the program’s necessary inclusion of ‘consciousness’ but not to extend skepticism to the reported consciousness of other humans?
I suppose that is taking for granted that consciousness is programmable, that it is a result of our DNA program.
1. The universe is a dynamical system, not a computer or computer simulation.
DeleteThere are processes that can only be predicted and explained probabilistically in nature (statistical physics), and (according to some) quantum mechanics involves true randomness/indeterminacy, not just unpredictability.
Engineering systems and biological ones can have statistical and random components too.
One of the definitions of a random data series is a series for which the shortest algorithm that generates it is as long a the series itself. So predictability is only possible when there is an algorithm shorter than the series. (This is related to information and to scientific explainability.) So random strings are, by definition, unprogarmmable; only pseudo-random ones are, such as strings far out in the decimal expansion of pi, which "look" random (but of course are not).
2. Yes, it's reasonable to assume (for the reasons you mention, and others) that other humans feel (despite the other-minds problem). And we do assume it (and our assumption is right). The main basis is similarity and correlations. These work for all vertebrates, and also, I think, all invertebrates. But they break down with microbes and plants.
What about T3 (Gabe)?
Nor does it mean feeling (consciousness) is computational. Feeling, like growth, is a dynamic property (protein synthesis, etc.); DNA is only part of it. And DNA is not symbols of arbitrary shape, executed by a Turing Machine.
(2/2)
ReplyDelete3. “Whenever one of these machines is asked the appropriate critical question, and gives a definite answer, we know that this answer must be wrong, and this gives us a certain feeling of superiority. Is this feeling illusory? It is no doubt quite genuine, but I do not think too much importance should be attached to it. We too often give wrong answers to questions ourselves to be justified in being very pleased at such evidence of fallibility on the part of the machines. Further, our superiority can only be felt on such an occasion in relation to the one machine over which we have scored our petty triumph. There would be no question of triumphing simultaneously over all machines. In short, then, there might be men cleverer than any given machine, but then again there might be other machines cleverer again, and so on.”
I’m unsatisfied with Turing’s response to the mathematical objection: if we accept that there are types of questions that humans are better at answering than machines, and vice versa, but could it not be argued that that ability to accurately answer the types of questions humans are better at answering than machines, is the essential measure of “thinking” or “cognition”? And does it matter than machines are better at certain “types” of thinking: isn’t the point of the imitation game to prove that computation can explain human cognition, but not vice versa?
4. Regarding Lady Lovelace’s objection:
One of the common arguments I’ve heard made for a limit to artificial intelligence similarly to Lady Lovelace’s objection is that machines will never be able to “judge art” (or my interpretation of the argument: computers will never be able to simulate a “sophisticated” art critic’s response to art, or to create art that is worthy of the “sophisticated art critic”’s praise).
Remembering this, I searched for an article that I had read years ago about a “Deep Learning” machine that was input Bach chorales and output imitation chorales of its own:
https://www.technologyreview.com/s/603137/deep-learning-machine-listens-to-bach-then-writes-its-own-music-in-the-same-style/
This machine seems to not be Turing-test passing, but close: (if in this version of the Turing test, the instruction (rather than question) is "compose a chorale" and the candidates are a machine and Bach):
“When given a DeepBach-generated harmony, around half the voters judged that it was composed by Bach. That’s significantly higher than with music generated by any other algorithm.”
“Even when confronted with music composed by Bach himself, participants only judged that correctly 75 percent of the time.”
It seems that this is a valuable evidence against Lady Lovelace’s argument, however it is important to note that Bach is perhaps an easy target:
“This method is not only applicable to Bach chorales but embraces a wide range of polyphonic
chorale music, from Palestrina to Take 6,” say Hadjeres and Pachet.
“In many cases, that will be easier said than done. Bach’s chorales are highly structured and follow specific rules in their construction, albeit a great many of them. Other forms of music are not always so organized.”
It’s my hunch that if aesthetic sensibilities are programmable, this would be one of the more sophisticated / last frontiers of any Turing-test-passing machine.
3. We are not talking about question-answering and Turing Machines but about T3 (Gabe), who can only be partly computational, makes mistakes, in the slightly higher university-student-level range (if he is a pygmy, like me, or even fewer mistakes if he is a giant, like Turing). Humans vary in their abilities. The T3 is generic.
Delete4. Lots of people have bad artistic taste, but that would not make them fail T3.
Producing Bach-like chorales is not the Turing Test; and T3 certainly does not have to be a Bach (or any giant)!
Also, it's not the same to copy Bach as to be Bach. And Bach wrote a huge amount of masterpiece-level work, and some that is only gifted-mortal level. I doubt the computer-generated ones were masterpiece-level. They were just imitations of Bach's style that could be confused with some of his own lesser works (which are also less well known).
This paper begins with the question: Can machine think? Turing replaces this ambiguous question with a closely-related scenario of imitation test, and asks if machines can pass it which set the central theme of this passage on “can and how can machines behave intellectually like human beings”.
ReplyDeleteThe scenario for the imitation game is to have a man and a woman in different rooms from an irrigator and the irrigator is asked to deduce their genders while the woman is telling the truth and the man wants to confuse and fool the irrigator. The imitation problem asks if a machine can successfully fool the irrigator if it takes the place of the man. The advantage of this question lies in that it can separate the intellectual capacities from the physical of a man. Although there may be difficulties in robots behaving like men, Turing maintains that we do not need to worry about this if the robots are constructed successfully in passing the test.
Turing then set the restriction to digital computers and gives an outline of the three main parts of computer, store, executive team, control namely. The special property of a digital computers to mimic any discrete state machine makes it a universal machine.
Turing then addresses and discusses nine possible objections against his idea such as “religious objection” and “heads in the sand objection” and takes pain to point out the fallacies in them.
Turing finally discusses about the endeavors that have to be made to have the machine passed the test. He proposes this idea of produce a program to stimulates child’s mind and develops its intelligence through teaching process, especially through rewards and punishments. Also, for the sake of searching for a solution of some problem, Turing argued it might benefit if we include random element to produce variance and mutation in the learning process. In the end, Turing expresses his hope in machines competing with men in all intellectual fields while admits that there are lots to be done.
I have some questions with regard to the latter half of the paper:
ReplyDeleteIn 6.3 "The best known of these results is known as Godel's theorem ( 1931 ) and shows that in any sufficiently powerful logical system statements can be formulated which can neither be proved nor disproved within the system, unless possibly the system itself is inconsistent." Turing defends his view by saying that this such kind of limitation is also applicable to human beings. But I am just wondering that if it is possible that the limitation of this principle applies to human and the computer to different extent, since the computer is constructed by intricate but stipulated rules of logical system, but human reasoning does not.
Also, in 6.4, the author maintains that "According to the most extreme form of this view the only way by which one could be sure that machine thinks is to be the machine and to feel oneself thinking." I agree with his argument against this 'extreme' form, but as Turing admits that he is attacking Jefferson by attacking the most extreme form of the problem, is there a danger of straw-man fallacy in his argument? As Turing himself contends in the Theological Objection that there are some differences between animate and inanimate objects, could there be a possibility of differences of consciousness that could be argued (or interpreted from Jefferson's argument) without eventually converging into solipsist?
There is no reason to think a human would be more or less bound by Gödel's theorem than a T2 or a T3.
DeleteThe other-minds problem is not solipsism. It is correct. And for that reason Turing says his approach only concerns doing, not feeling (mind).
While I understand most of the arguments in the Turing paper - point 5 has a sentence stating that all machines move continuously from one state to another. I'm finding it a little hard to wrap my head around this. Although I understand the argument that electric signals are more or less continuous, the storage unit of a machine relies on hardware that stores information as bits. Doesn't storing information as bits necessarily mean that the state of the hardware is also binary? Or am I missing something here?
ReplyDeleteAdditionally, for point 8 Turing covers the possibility of machines displaying informal behavior. Does this also include the possibility of the machine generating innovative and creative solutions in one sphere of knowledge based on another? In essence can machines have a 'Eureka' moment or a moment where rules from one modality are for the first time implemented in another modality to create a new outlook/solution? For instance, I understand machines can write poetry, but can they solve a non-empirical open math problem say in real analysis?
I think Turing meant successively, without interruption, not continuously in the mathematical sense. As a co-inventor of the digital computer, Turing knew better than any that it is a digital, finite-state machine. His Turing Machine already shows this.
DeleteYes, machines (including computers and humans) can do something new. See replies on this to other skywriters).
“I believe that in about fifty years' time it will be possible, to programme computers […] to make them play the imitation game so well that an average interrogator will not have more than 70 per cent chance of making the right identification after five minutes of questioning.”
ReplyDeleteTuring dedicates a large portion of his paper to describing the type of machine (namely, a digital computer) that could potentially pass the Imitation Game, while failing to elaborate on the characteristics of the interrogator – the person tasked with correctly distinguishing between human and machine. At first, I thought that this may be an oversight on the behalf of Turing, since – by virtue of being distinct people with unique personalities – any two people taking on the role of the interrogator would play the Imitation Game very differently (e.g. asking different questions, reacting differently to the answers given by the participants, etc.), which could result in certain people being ‘fooled’ much more easily by a particular thinking machine than others. Hence, one could then argue that the outcome of the Imitation Game will always be affected by the inherent subjective variability of its interrogators. After some thought however, I realized that the aim of the Turing Test is not to ‘fool’ humans by having a machine ‘mimic’ a human being’s cognitive capacities. The Turing Test is also not about evaluating whether some people are more or less susceptible to being ‘fooled’ by a machine. Rather, it is about assessing whether a machine can successfully reverse-engineer human cognitive capacities – if it does, the machine has passed the Turing Test. As long as a machine is capable of doing everything that a human can do (with the exception of verbal and physical abilities), then it automatically passes the Turing Test (T2) – regardless of who acts as the interrogator.
Gabe can email for a lifetime with anyone.
DeleteTuring brought up some contrary views in his paper to his thoughts on what he calls the "imitation game" (although as discussed in the other reading, it is unfortunate he used this terminology to describe his ideas), and refutes these alternate arguments. As I was reading his description of it in the beginning of the paper, I was thinking about some critiques to it but he happened to later address all of them.
ReplyDeleteIn (9), Turing discusses the argument from extrasensory perception. He states that "the statistical evidence, at least for telepathy, is overwhelming. It is very difficult to rearrange one's ideas so as to fit these new facts in". Turing says that he believes this is a strong argument against the notion that machines think. I'm still a bit confused by what he means by telepathy in this regard, and not yet convinced I can agree with his opinion on this argument.
I also found the concept of learning machines very interesting. Turing proposes, "[i]nstead of trying to produce a programme to simulate the adult mind, why not rather try to produce one which simulates the child's? If this were then subjected to an appropriate course of education one would obtain the adult brain." This is a fascinating concept to me, although it is kind of a dystopian one. He compares a child's brain to a blank slate, and that education shapes their adult brain. To train a machine to think, perhaps it might be best to use a similar approach on it and make the machine easily programmable, and find effective rewards and punishments to facilitate learning (just as with a human child). The problem here is that these rewards/punishments with machines cannot be emotional. I was under the impression that emotional responses to reward/punishment for learning are rather effective, so I'm wondering how effective this method would be for a computer if emotion is absent?
At the end of the paper, Turing says "we may hope that machines will eventually compete with men in all purely intellectual fields." I'm left wondering why. Why do we want machines to compete intellectually with humans? I can understand if it is for something like improving fields like biomedicine or our environment, etc. But beyond that, I don't understand why we want to develop machines to out-think us humans.
Turing's strong suit is not what he says here about telepathy, for which all evidence is negative.
DeleteEmotion is felt. Turing has rightly renounced that. Today's computers are already capable of reinforcement learning; no feeling needed. Just postive and negative feedback.
By far one of the weirdest things Turing does in this article to me was at the end of the example, when he breaks down the components of the process that bring an adult human mind to the state it’s in and claims that the key to creating a computer that passes the Turing Test (I believe he’s talking about T1 here) might be to program a computer like a initial human mind (at birth, with only innate knowledge) and subject it to the education and non-educational experiences he mentions. In this line of argumentation, I don’t think Turing sufficiently addresses the lack of information we have regarding what is and what isn’t innate, and about the structure of the mind.
ReplyDeleteFirst (about what is and isn’t innate), we have excessively little to go off of when it comes to what information the brain already has. Beyond some examples of early language acquisition devices (the parameters of which are still incredibly blurry), we have very little to go off. I’d even argue that it would take thousands of years to get there.
Secondly, and forgive me if this falls under the head in the sand argument he was trying to refute earlier in the text, the human mind is structurally different than the input/output/instructions structure of a computer. Differential importance of processes plays a huge role in the fallibility of humans (which Turing wants to embed in these computers in a completely different way). All in all, I just don’t think that the type of fallibility that Turing wants to introduce wouldn’t work in the same way than it does in humans (and not just in terms of the invisible mechanisms of it, which we don’t care unless you’re doing T4: it would impact the real output of the machine).
It's not clear whether Turing is a computationalist. So it's not clear whether he thinks a computer alone could pass T2. It certainly could not pass T3. And T3 necessarily involves analog sensorimotor functions that are not Turing computation (though they can be simulated by computation).
DeleteFallibility is a non-issue: computers have it, robots have it, and so do human and nonhuman animals.
Turing had some strong views in which he persuasively states throughout his paper. As he begins with defining his version of thinking, Turing presents the audience with the imitation game. Turing then goes on to defend his points from every angle and ends with a fortification of his arguments once more. Throughout his defence, he makes a point to exemplify the relevance of his work and possibly the irrelevance of the objections points. However, I would state that some arguments of Turing were better than others. His defence for theological and mathematical objective seem sound. They both point towards a distinct potential flaw in both objections. In the theological argument, he points towards the fact that mothers do in fact bear children and that indirectly also means that she bore a soul. In the mathematical argument, he points out the limitations of the human mind. However, when it comes to the argument of consciousness, and informality of behaviour, Turing, doesn't seem to have a strong argument. For consciousness, he argues that it is not relevant as the question can never be answered. In contrast, his argument for informality of behaviour is that even if we don't know the answers, we shouldn't say that there is none. Turing seems to be doing whatever he can to avoid the question of consciousness and inconsistencies in the human mind.
ReplyDeleteYes, Turings rebuttals are not all of the same quality. (Neither are the arguments that he is rebutting.)
DeleteHe is right (and it is the core of the TT) that TT modelling cannot solve the hard problem of explaining feeling, just doing; and that, despite the other-minds problem, Turing indistinguishability cuts both ways: if we cannot distinguish T3 from a person through its doings, we have no better basis for affirming or denying that T3 feels than that other persons feel.
In his paper, Turing refutes what I would say may be the most common objection against the idea that machines can think which is the argument from consciousness.
ReplyDeleteThinking , is a byproduct of consciousness, thus for a machine to be able to think , or successfully play the imitation game, it should be conscious and generate its own thoughts. Turing’s answer to this view is to refute the solipsist position by saying “ A is liable to believe “A thinks B does not “ whilst B believes “B thinks but A does not”. Instead of arguing continually over this point it is usual to have the polite convention that everyone thinks. “
This, in my opinion is a very expeditious way to reject the argument from consciousness by saying that the only position that claims to verify it, is the solipsist position. Even if I don’t know what it is like to be my father, I can still say that he is a thinking being since we are both humans, have biologically similar brains and nervous systems and generate similar behaviours. We don’t need to take the solipsist point of view.
Thinking is expressed by language, the generation of new ideas, behaviour, taking decisions, act, and by that fact that “there is something it is like to be a being”. A machine can’t feel or like Lady Lovelace said, “generate anything new”, the only thing it can do is to provide outputs to previously entered inputs. In the case of Turing Machine, it can only write answers, which is not sufficient to categorize it as thinking.
Turing does not refute the mysteriousness of consciousness altogether, he admits that "there is a mystery about consciousness" embedded in the question of localizing it. The fact that we do not know how exactly consciousness works and what is going on when we are 'thinking' is what makes it difficult to argue that someone else is a 'conscious thinking being'. We may still expect and go about our lives as though everyone else is a 'conscious thinking being' - as we do. However, how can you be sure of someone else's thinking ability if what supports your own being a conscious thinker is that you are aware of your own thoughts?
DeleteAdditionally, I want to address your last point where you state that a machine cannot feel or "generate anything new" (Lady Lovelace). You state that the only thing it can do is provide outputs to previously entered inputs. I want to discuss this from the perspective of 'learned behaviour'. If we are in fact able to program and machine with a learning algorithm, and the machine is successfully able to learn in similar ways to us, can it not be said that whilst learning the machine is generating something new? If the machine is learning from historically programmed inputs which it uses later in combination to generate new inputs we can say that the machine has generated inputs which were not previously entered by the machine's programmer. The programmer cannot predict the machine's every output as it's outputs are now contingent on it's newly learned abilities. If these new abilities were not previously entered (by the programmer) inputs we can say that the machine has "surprised" us in generating something which was not explicitly programmed.
I assume "Unknown" is Paul Ashkenas.
DeleteAbout the other-minds problem, see the preceding reply (and others). There have also been many replies about Lady Lovelace on novelty.
"The criticism that a machine cannot have much diversity of behaviour is just a way of saying that it cannot have much storage capacity."
ReplyDeleteWhen I read this part of Turing's paper, I had a thought. What if someone were to upload their thoughts, memories, feelings, etc. into a computer or machine of some sort, and the computer was able to learn how to be a social 'human' from those inputs? So, the computer would have all the necessary inputs to learn how to communicate from and base their social learning of others on. This is hypothetically only possible once we know how to understand feeling, or emotions at a deeper level. Imagining that it is achievable, a new line of thinking is introduced concerning the question we discussed in class about whether we would kick Gabe the robot from MIT. If the robot is indistinguishable from everyone else, but we were aware that there’s something underlying them that makes them not human, there’s something about Gabe that makes us feel like he’s invalid in some way. However, if Gabe were now also to contain the ‘real human’ qualities of another person, it makes this question of our feelings towards Gabe more difficult to answer.
Following this, to be able to upload a person’s entire brain contents concerning communication, emotional reasoning and understanding, etc., the computer would need to have a large storage capacity. This is a point that’s brought up by Turing a lot in his paper, that storage capacity is an important part in determining what a computer is capable of doing.
In the quotation above, Turing is acknowledging an argument of various disabilities that argues that there is a lack of diversity in the behaviour of these machines, which he refutes by arguing that diversity in behaviour and storage capacity go hand in hand.
Now, we have computers with immense amounts of storage capacity, but we're still unable to achieve this goal that everyone has in mind when it comes to having a computer that's indistinguishable from the regular person. So, I would like to argue that although storage capacity is necessary in order to hold all the information we want, it's not the main feature contributing to the ability of a computer. Further from this, relating this to the Turing Test, I don't think passing this test is related to the storage capacity of the machine in an important way either. Rather, the inputs given to the machine are what is important. If the inputs were to cover every possible situation of communication, creating a machine that passes the Turing Test becomes a less complicated task.
To connect this to the discussion of feeling from class, I don't think we need to be able to understand feelings and emotion in order to have a machine pass the Turing Test. I don’t have a full understanding of what my opinion is in regards to this, but I have an example that I would like to share that’s driving me to think this way. So, there are human beings who are able to communicate effectively with others without any doubt of them being human, but that have disorders that prevent them from feeling emotions like others do, or behaving the same way as others in social situations. For example, a psychopath doesn’t feel the same pleasant emotions as others such as empathy, and although this prevents them from having a ‘human’ characteristic, it doesn’t prevent others from questioning whether their insides are biologically the same as ours, in the general sense. Perhaps, attempting to model a computer after an individual with a disorder such as this would bring us closer to being able to produce a machine like the one we're envisioning. However, again I haven’t made my mind up about this thought yet, but wanted to share it anyway.
How do you "upload" feelings?
DeletePassing the TT is not a problem of storage capacity. T2 cannot be passed by computation alone, and T3 is not just a computer.
A psychopath may not have the feeling of compassion or empathy or guilt, but they do feel (pain, heat, hunger, anger, etc.)
What struck me the most in Turing’s paper was his idea of producing a programme to simulate a child’s mind. “Presumably the child brain is something like a notebook as one buys it from the stationer’s. Rather little mechanism, and lots of blank sheets.” This builds on the tabula rasa idea, and “hoping that there is so little mechanism in the child brain that something like it can be programmed” depends too much on the idea that everything about our cognition is acquired through learning and experience. Here Turing seems to ignore the fact that humans follow a different process in thinking and learning and it’s not simply reward mechanisms or “programming”.
ReplyDeleteEven if the “child programme” and the “education process” are provided, and even if such programme yields the desired “output”s similar to human development it’s not good enough to conclude that such machine has cognition or a “thinking” process but simply computation.
Turing does not propose to simulate a child but to implement it. Nor is it clear that Turing is a computationalist. Turing proposed the TT as a methodology, but he never actually tried to design a machine that could pass it.
DeleteBuilding on this somewhat, modern machine learning and deep learning have built a similar reputation for building "children" algorithms and teaching them to become progressively more sophisticated.
ReplyDeleteSome of these algorithms have been hugely successful. In the example I've linked from CGPGrey, it's explained that it's somewhat easier to create a machine "dial adjustment bot" that tweaks parameters until the desired result is achieved. While a human could theoretically be competent at modifying parameters until the desired outcome is achieved, the scale of the number of combinations quickly overwhelms a human's patience whereas machines have infinite patience to train.
https://www.youtube.com/watch?v=wvWpdrfoEv0
Bringing this back to the Student/Teacher example you brought up, perhaps by a similar logic it would be easier to create a teacher machine than to create the student. This would allow the student to train at a speed humans could never comprehend until a satisfactory result emerged.
Computers can already do reinforcement learning: trial and error, followed by corrective feedback, positive if they did the right thing, negative if they did the wrong thing. No need for feeling. Just a system whose connection weights keep getting adjusted by every trial/error/feedback to make them do the wrong thing less and the right thing more.
ReplyDeleteTuring soundly argues against each of the nine objections. However, I am still unconvinced by a few aspects of his rebuttal of 'Lady Lovelace'. Lady Lovelace states machines cannot ‘originate’ anything… ‘it can do whatever we know how to order it to perform’. Turing equates this to “a machine can never take us by surprise.” This interpretation puts the emphasis on the observer, while Lady Lovelace’s statement focuses on the actor. The observer subjectively acts surprised. The actor objectively creates original content (outside the realm of what it was programmed to do). Though later on, Turing talks about probabilistic algorithms (in continuity in the nervous system) which, in my opinion, offers a better refutation of Lady Lovelace’s originality objection. Slight deviations of output through probabilistic means can create originality. I am not too sure if this is the degree of originality Lady Lovelace intended. I do suspect she meant something greater... (inventing a new school of thought)
ReplyDelete“Not until a machine can write a sonnet or compose a concerto because of thoughts and emotions felt, and not by the chance fall of symbols, could we agree that machine equals brain-that is, not only write it but know that it had written it. No mechanism could feel (and not merely artificially signal, an easy contrivance) pleasure at its successes, grief when its valves fuse, be warmed by flattery, be made miserable by its mistakes, be charmed by sex, be angry or de- pressed when it cannot get what it wants.”
ReplyDeleteTuring presents this quote as “The Argument from Consciousness”, refuting this argument through connecting it to the “other minds problem”: We can never know if another entity (even a human) is feeling unless we were to become that entity. In order to refute a solipsistic view, we infer feeling through a quasi-Turing Test of our own. By asserting that a machine is not acting out of feeling and emotion, this argument rejects our ability to make the assumption of feeling to any entity; Turing therefore renders this argument invalid as the solipsistic view is not widely held.
I also connected this quote on the consciousness argument to “Lady Lovelace’s Objection” that “The Analytical Engine has no pre- tensions to originate anything. It can do whatever we know how to order it to perform”. I think that “[writing] a sonnet or [composing] a concerto because of thoughts and emotions felt” plays into the ability of original thought, as complex emotions seem like experiences uniquely self-generated; the expression of those feelings into something creative and striking seems similar to the ability to surprise. I accept Turing’s rejections of these arguments, but I am just noting their overlap.
Re: Can machines think?
ReplyDelete“Instead of trying to produce a program to simulate the adult mind, why not rather try to produce one which simulates the child's?”
This has partially convinced me to believe that this could be the key to figuring out how we cognize throughout our lives. It also would tell us more about the learning process and why it is evolutionarily important. Additionally, I think it is beneficial to work backwards from an adult human brain rather than to build from the bottom up. Turing gives the “skin-of-an-onion” analogy, which supports the idea that if we want to create a machine that is a perfect model of the adult brain, then we need to peel back the layers and see how much we can encode based on our current knowledge of the brain and cognition. By stripping off this “skin”, we have the potential to find the “real” mind.
On the other hand, I would argue that the development of a child's mind more closely resembles supervised machine learning than any sort of programmable "education" that Turing suggests. In starting with the child-like mind and expanding its knowledge base by way of education, I’m hesitant to conclude whether the desired adult-like minded machine would result. The development of a child's brain seems less of a notebook of “blank pages to be written in, and more of a series of attempts to write correctly.
Turing is quick to point out rewards and punishments as a sufficient teaching mechanism. Although, humans learn by methods other than the simple stimulus-response relationship, such as by observing others around them. In order for a machine to learn in this way, it would have to be able to maneuver around the world. With that being said, I could argue more for the social-facilitative and modeling behaviours through which children learn to imitate the actions of those around them.
Hi Pauline!
DeleteIf I understand correctly, you say that it would be better off to try to simulate a child’s mind, because it would give us a basis of understanding regarding how we, as humans, cognize throughout our lives.
I don’t think the that the effort to simulate a child’s mind would give us anymore knowledge than trying to simulate an adult mind. Yes, maybe it might teach us more about how the mind/brain works specifically in the earlier years of the life, but the issue that we are facing regarding the efforts to simulate is “the hard problem”. Even if our interest is a child’s mind or an adult’s mind, the hard problem still exists even if we simulate it, we wouldn’t know how to create feelings. I think that trying to simulate a child’s mind would not give us more insight about how to solve the hard problem, ie. how to do more than just simulating and give that simulation the ability to “feel” the things that it is doing.
To take it further, if we have in mind to reverse engineer by means of going back in the development of a mind, we need to go back to birth of that child, we need to understand the genes, we need to investigate how those genes are interacting with each other at physical level… So, there would be many layers that we will have to peel of in order to be able to fully understand how the mind works. Maybe those layers are as many to bring us to the level of quantum interactions (which was the idea behind Penrose’s quantum cognition argument). But then, how is the bare quantum mechanics relevant to the problem regarding something like the “feeling” which is a way higher level of phenomenon.
Like Pauline, I think that studying a child's mind would be useful in understanding how a child forms new categories, concepts about the world, assigns meaning to what he learns, observes , makes associations which would help us reverse engineering the mind. Here it is not about the hard problem; it is about finding methods human use to replicate in computer machines.
DeleteAfter having read Computing Machinery and Intelligence, I’m left with many questions. First, would succeeding at passing the Turing Test be significant beyond its indication that the algorithm has extensive verbal abilities? Does passing the Turing Test indicate that computers exhibit intelligence? Can an algorithm be intelligent without a body? Would linguistic fluency be the product of mimicry or machine learning? Does language competency represent insight into human culture? How would tech like Siri and Alexa perform on a Turing Test?
ReplyDeleteHaving read the paper over again and considering our in-class discussions, I understand the Turing Test to be a benchmark or standard used to evaluate an algorithm’s mastery of human language and thus a means to begin evaluating the abilities of machinery in relation to our own. I understand it as a necessary but insufficient evaluation of generalized intelligence, given T2’s unilateral focus on verbal performance. As for the role of a body, when taking into account the symbol grounding problem, the answer would be no: grounding of words relies on direct interaction with the external objects (referents) by a dynamical system with sensorimotor capabilities. “The necessity of groundedness, in other words, takes us from the level of the pen-pal Turing Test [T2], which is purely symbolic (computational), to the robotic Turing Test [T3], which is hybrid symbolic/sensorimotor” (Harnad 2000, 2007). The mystery of consciousness then arises as symbol manipulation only takes on meaning through the identification of referents (groundedness) – a function of the brain.
The Turing Test is supposed to be a test which determines whether or not a computer (AI/Turing machine) is capable of thinking like a human being. Therefore, it is supposed to asses if a computer is cognizant. The Turing test is about *doing*and assess only what a computer can *do*. If a computer passes the Turing test it would be said that that computer is cognizant however, this decision that it is cognizant would be based only on whether it performed or *did* things in the right way (as to 'trick' the outside person into thinking it was a real human). If a computer passes the Turing Test then we are supposed to conclude that that means that the computer is 'conscious'. However consciousness = feeling. And Turing has nothing to say about feeling. The Turing Test has no way of dealing with feeling and does have any ways to distinguish if a computer feels or not. Therefore, the Turing Test is not an appropriate way of concluding whether or not a computer is cognizant.
ReplyDelete"Instead of trying to produce a programme to simulate the adult mind, why not rather try to produce one which simulates the child's? If this were then subjected to an appropriate course of education one would obtain the adult brain." (Turing, 1950).
ReplyDeleteThis quote explains the theory of Learning Machines which states that if instead of designing an adult brain we should strive to create (or simulate) a child's brain and then give it the ability to learn. If we were to simulate a child's brain and then subject it to the appropriate course of education we would perhaps have a better understanding of how it is that we can do what we can do. Additionally, it may be much harder to simulate an adult brain because you would have to simulate all the fully developed thoughts and abilities that an adult has. on the other hand, if you simulate a child's brain it would be much more of an easy task because you would have to simulate a fairly 'empty' brain and allow it to create meaning and thoughts for itself through sensorimotor capacity and education. Turing talks about the child's brain as a 'blank notebook' full of 'empty white pages' - this seemingly would be a lot easier to simulate. However, I don't think it's as simple as being a blank notebook waiting to be filled - there are some things are innate or certain abilities that maybe are already present but just haven't been activated or tapped into yet.
In his paper, Turing refutes what I would say may be the most common objection against the idea that machines can think which is the argument from consciousness.
ReplyDeleteThinking, is a by-product of consciousness, thus for a machine to be able to think, or successfully play the imitation game, it should be conscious and generate its own thoughts. Turing’s answer to this view is to refute the solipsist position by saying “ A is liable to believe “A thinks B does not “ whilst B believes “B thinks but A does not”. Instead of arguing continually over this point it is usual to have the polite convention that everyone thinks. “
This, in my opinion is a very expeditious way to reject the argument from consciousness by saying that the only position that claims to verify it, is the solipsist position. Even if I don’t know what it is like to be my father, I can still say that he is a thinking being since we are both humans, have biologically similar brains and nervous systems and generate similar behaviours. We don’t need to take the solipsist point of view.
Thinking is expressed by language, the generation of new ideas, behaviour, taking decisions, act, and by that fact that “there is something it is like to be a being”. A machine can’t feel or like Lady Lovelace said, “generate anything new”, the only thing it can do is to provide outputs to previously entered inputs. In the case of Turing Machine, it can only write answers, which is not sufficient to categorize it as thinking.
I have a question with the Turing machine. How can a T2 machine answer any question and have a written email conversation with someone if it doesn’t understand the meaning of the conversation, the questions asked? Not all answers will be stored in his memory storage, so does it generate random answers for preprogrammed questions all the time? Was Turing thinking about any kind of answer it was programmed to do ? Because when I answer a question, I think and ground it to something I know, however T2 machines just perform computations they don’t understand and attach meaning to anything thus they couldn’t haven any kind of conversation.
ReplyDelete“There are, however, special remarks to be made about many of the disabilities that have been mentioned. The inability to enjoy strawberries and cream may have struck the reader as frivolous. Possibly a machine might be made to enjoy this delicious dish, but any attempt to make one do so would be idiotic”
ReplyDeleteThe last part of this paragraph is absurd, I think. I don’t agree at all that it is idiotic to make a machine enjoy strawberries and cream because being able to enjoy something is a proof of interpretation and interpretation is related to computation. Therefore, seeing if a machine is able to do it helps us understand the principles of computation.
Also, Turing weirdly believes in paranormal activities like telepathy and telekinesis, but he doesn’t believe that the mysteries surrounding consciousness should be solved. To me, these statements lead me to have less credibility in him. Telepathy has no solid foundation, and consciousness does lack for some scientific proof, but it has more proof than telepathy.
The definitions of the terms “machine” and “think” are many and varied, especially when accounting for the more lay use of the words and then those varying between disciplines (for example computer science vs philosophy). Turing provides many restrictions which result in only digital computers fulfilling the criteria of a “machine”, and to avoid circularity of argument mentions it consisting of 3 parts: store, executive unit and control.
ReplyDeleteThe paper focuses on the imitation game where a machine takes the place of a participant in an online interaction with a person. Turing refers to this as the New Problem. Based on this he discusses the question “Can machines think?” He is of the opinion that the question is “too meaningless to deserve discussion” and that at some point in the future language will have evolved enough that speaking about the thoughts of machines will be normal. Previous readings have covered this topic in that concepts and language and the meaning of words develop over time and come to mean different things. From this perspective Turing’s point is very conceivable. What is difficult to foresee is what the meaning of those words will evolve into, specifically what “thought” will mean in reference to machines. Will it evolve to mean something different than what we currently think of as human thought (while this alone is already a contended subject).