1 / 23
From Color, to Consciousness, toward Strong AI
Abstract: Largely relying on reasoning, this article cohesively discusses three topics, namely color and its perception, the yet-to-be-solved hard problem of consciousness, and the theoretical possibility of strong AI. First, the article restores color back into the physical world with a piece of cross-species evidence. Secondly, the article establishes a dual-fielded with function Q theory (DFFQ) which might explain the ‘first-person point of view’ and thus the hard problem of consciousness. Finally, the article discusses what the DFFQ theory might bring to artificial intelligence and how it might allow strong AI to stay true.
1. Introduction Empirical evidence in cognitive science suggests that color is our common illusion of this world. The nature of consciousness has long been under disputation. No consensus has ever been reached. Is there any possibility for us to realize strong AI? Color, consciousness, and strong AI, three seemingly separated topics, are going to be discussed cohesively in this article. Largely based on reasoning, first, the article relocates color back into the physical world and discusses our largely shared, reliable reconstruction of it. Secondly, the article establishes a dual-field with function Q (DFFQ) theory of consciousness, in which a hypothetical function could explain the hard problem of consciousness. Finally, the article considers some cases in artificial intelligence in the light of the DFFQ theory. 2. Color ‘Color’ is important because it relates to how we see and how we understand the real world. Philosophical views on color seem to be chaotic. Among all the debates, questions such as ‘Is “color’’ a property of objects?’ and ‘Is “color” mind-dependent?’ serve critical roles. I will attempt to answer the questions in this part of the article. To do that, I would like to summarize an understanding of color first: We believe we see things in color, but scientific facts betray our intuitive belief. The truth seems to be that, physical objects invisibly ‘interact’ with lights; the latter are to be reflected, absorbed, and etc. Lights are not really passive at micro-level. Both the nature of objects and the nature of lights participate in determining the results of the interactions. We human beings think we see objects in color, but the colors we perceive are actually the interactions further processed by our brain. Full Name: Xinyuan Gu Institutional Address: N/A E-mail Address:
[email protected]
;
[email protected]
2 / 23
2.1 Function F and Function G It is so easy and perhaps so common for those who know the facts above to conclude that, since colors are after all what we perceive, it is fine to say objects do not have color and to say seeing color is a kind of subjective experience. Such conclusions, however, are not satisfactory to me. One fact to be confirmed first is that, in theories and practices of physics, with known nature of objects and known nature of lights, all the reflections, absorptions, and other interactions, in principle, can be calculated. Let me represent the nature of an object with x, the nature of certain light with y, and the interactions in total with z. Knowing x and y should lead you to z; knowing x and z should lead you to y; and knowing y and z should lead you to x. Admittedly, If we only know Newton physics level details, it might be difficult to calculate quantum level results. However, I believe the principle should hold. Due to this fact, the relation between the interactions in total and the nature of the two participators is analogous to a (mathematical) binary function: z=F(x,y). Physicians have ‘Parameter’ as their term, but here in the analogy I will use ‘input’ and ‘output’ instead to serve the purpose of my discussion. In this binary function, x and y are the inputs, and z is the output. The calculations simulated by function F may involve a group of (mathematical) functions in actual practices; however, the relation is perfectly represented in z=F(x,y), in the sense that the result of interactions in total depends on and only on the nature of the two inputs, namely the object and the light. Technically one would need all that involved in physical laws in a calculation. However, if I am not mistaken, x and y are the only two actual-world dependent ‘inputs’. Others involved are not actual-world dependent but law-dependent. What does the fact say? It says that there is no place in z=F(x,y) for perceivers. It says that, whatever the nature of the result is, the result does not depend on anything else world-dependent than the object and the light. It says that, without the presence of human beings, objects and lights will keep happily interacting with each other. On the other hand, without objects and lights, and especially without objects, human beings are not supposed to perceive anything, including color. Therefore, color objectivism is correct first of all on the point that ‘the existence of colour instances……does not depend upon the existence of perceiving animals’, as summarized and objected by Hardin. Why, then, would people so tend to emphasize the role of perception? z, as the output of function F, whatever it is, will be received by human eyes and further be processed by the human brain. ‘The perceived colors are outputs of our brain’ thus is a scientific description. Let me define colors in our perception as i, and the way a human brain processes z as function G, we will have i=G(z). This function tells us that we are seeing a processed z, and how z looks depends on how we process it. However, remember z is the output of the independent function F. Therefore, function G could be rewritten into i=G{F(x,y)}. To those who over-emphasize perception, z is passive in and dependent on G; but the truth is, z is fully dependent on the result of F, and F is independent of G. To paraphrase, the color we see is dependent on human minds, but the whatever to be processed by the human brain is not dependent on human minds. Full Name: Xinyuan Gu Institutional Address: N/A E-mail Address:
[email protected]
;
[email protected]
3 / 23
2.2 The Whatever to be Processed So what is z, namely the output of function F, namely a result of the interactions between an object and lights, namely the whatever to be processed as the input in function G by our brain? I believe it is reasonable to define z as (raw) color, an extrinsic property of objects, because z is a non-random status of appearance for x to be in when under conditional y. That is to say, color is a determined status of appearance for an object to present based on the physical-chemical nature of the object when under a conditional light. It is extrinsic because it is related to the external world. Color so defined, when as the input put into function G, is coherent with our experience in daily life. Suppose you are in a dark room, and you touch an object which you have never recognized before. Will you feel a bit difficult to construct a full picture of the object, even when you can feel the rough size and shape of it? And what would you do? You will tend to go to somewhere in better light condition with the object and re-cognize it because you intuitively know you will see better in better light condition. Different species have their own ways of responding to the physical world. One of such examples is stated in the light of evolutionary psychology (see Cosmides & Tobby, 1997). Suppose I am a little mouse, I would tend to grab the cheese to a darker place, as I am a nocturnal animal and my visual module is supposed to work better in better light condition (Or maybe I do not care much about visual cognition. Maybe the object smells like a cheese and that is sufficient for me. This is to be confirmed by biologists). In both settings, the color of the cheese (the object) is what it is, in determined statuses, and to be perceived under appropriate conditions to different species. The puzzle left is if when we see red we do see red. In a totally different setting, suppose sometime earlier you recognized a red-cover book under strong outdoor sunlight, then you brought the book back home, and thus the book is now under an artificial light condition. Will you still recognize it as the same book? Typically you will, perhaps because1) the status change at the surface of the book under different light conditions is minor or just perfectly handled by your brain as coherent; 2) you have recognized the book as a whole. The change of lights is almost invisible to you, and the color of the book ‘moves’ with the book as it is a status which can only present at the physical-chemical surface of the book. Whether a brain does something ‘subjective’ to (raw) color is to be discussed later in this article. Here I should conclude that there is no ‘subjective’ component in color as an extrinsic property of objects. 2.3 Blue Things and Hue-Radiation Relation As was pointed out in Hardin (1984), 1) ‘Apart from their radiative result, there is nothing that blue things have in common’ and 2) no hues are directly in or caused by lights (radiations). Based on these two observations, Hardin concluded that ‘objectivism fails’. This is the first challenge I would like to dissolve in this article. Hardin’s observations are probably correct (which means the premises are true); however, what do the two observations actually say? In observation 1), all things that are blue have nothing (physical, chemical) in common. In Full Name: Xinyuan Gu Institutional Address: N/A E-mail Address:
[email protected]
;
[email protected]
4 / 23
other words, all x that cause z=blue in function F vary, which is fine in the binary function, because, if I am not mistaken, it is fine for no one-to-one relation between x and z in function z=F(x,y). And we all know that in daily life examples z will vary within a range while still be taken as blue, so it is perfectly fine for all x that cause z≈blue to vary in function F. Observation 2) is based on hue, saturation, and brightness. Let me define hue as a, saturation as b, and brightness as c. Color is a z that can be precisely represented in a combination of certain a, certain b and certain c. Light, as I defined, is y. So observation 2) says that no a is directly in or caused by y, which is still perfectly fine in the binary function z=F(x,y). I would say the analogy, function z=F(x,y), demonstrates a possible explanation of the two observations, and thus Hardin’s conclusion that ‘objectivism fails’ fails. 2.4 The Evaluation of i It is obvious that, between the output of function F and the output of function G, namely, between z (raw color) and i (seen color), there is a gap. The gap is disturbing because if it is left unfilled, it will prevent us from asserting that we do know the actual world. Filling the gap seems to require two steps, and the first step is to discuss how ‘subjective’ we are when processing z. A subjectivist view was refuted in Hardin (1984) based on the reason that it is almost impossible to specify ‘normal conditions’ for seeing (raw) colors. I agree that the color processing process (function G) should be purely physical-chemical; on the other hand, I believe it is possible to define vision minorities of a species, and it is reasonable to hypothesize that, we have a brain module which allows us to largely share perceptions on color. Here are two experiments for us to consider: 1) There are, say, 100 participators in a room filled with artificial light. No one is color-blind or blind or in any other inconvenient condition. We have prepared 100 pairs of cubes. One yellow cube and one blue cube for each pair. Everyone in the room is assigned with a pair of cubes. How many of them would be able to pick the blue cube when required? What if they have one red cube and one orange cube for each pair, and are required to pick the orange one? What if they have one deep-blue cube and one light-blue cube for each pair, and are required to pick the deeper one? Why would you think so? 2) Suppose you bought a skirt (if it is hard for you to imagine, then you bought a shirt) online. There were four colors for you to choose from: bright pink, bright blue, bright yellow and bright green. Based on the photos posted by the seller, you chose the bright yellow one. Unfortunately, the arrived skirt is stupidly medium orangish-yellow to you. What would you complain to the seller? What assumption would you make when complaining? What if the arrived skirt is bright blue? In that case, what would you complain to the seller and what assumption would you make when complaining? For experiment 1), I predict most people would say ‘basically all of them would pick the required one’ because the inference fits our daily life experience. Traffic lights work only when vision majorities (probably whose LMS receptors are functioning well, and whose ‘interpreters’ are functioning well) reach roughly the same i based on roughly the same z. No Full Name: Xinyuan Gu Institutional Address: N/A E-mail Address:
[email protected]
;
[email protected]
5 / 23
one could defend him/herself with ‘The traffic light was green to me’ in front of a judge. And for that to stand, one has to assume that function G is not a cheap analogy but a reasonable simulation of what a brain of certain species could realize. Human vision minorities have some inconveniences (or not, if they insist, in this case) perhaps because they employ function G’, which is generally still the same but different in some aspects. For experiment 2), I predict that, if the skirt is orangish-yellow, most people would complain that the color of the skirt was not precisely represented in the photos; and if the skirt is blue, people would complain that the seller has made a mistake of sending the one in the wrong color. The assumptions behind would be interesting. The seen color based on which we made the decision of buying has to be complicatedly processed. First, we have the raw color z of the skirt. z is to be the input to a camera which has its own function of processing color, and the output is v. v in the digital photo of the skirt will be presented on your computer monitor. Your monitor has its own function of processing color, and the output caused by the input v is w. You are going to process the input w, and the final output is i. You may never think about such a processing process of z to i, but would you vaguely blame the inaccurate processing at the step of photoing, if your skirt is orangish-yellow? In the case that you receive a blue skirt, on the other hand, you would blame the seller because you assume the seller (who saw the skirt with his or her own eyes when picking) identifies blue as blue and yellow as yellow just as you do. These are after all fictional cases. However, the point is that radical color random-ism has to explain why practices involving seen color are basically smooth in various societies. 2.5 Paintings Based on hypothesizing that we have a brain module practicing function G which allows us to largely share perceptions on color, I would like to move to the second step of gap-filling: proposing that i is a reliable reconstruction of z. Paintings are to be our hero. Suppose we have z1 as raw color in total of beautiful scenery in front of our own eyes. A painter beside us just painted a realism painting of the scenery. The painter processed z1 with her eyes and then turned seen color i1 via her practice to raw color z2 of the original painting. Now there is a forger beside us copying the original painting. The forger processed z2 with his eyes and then turned seen color i2 via his practice to raw color z3 of the copied painting. Now we have the scenery in raw color z1, the original painting in raw color z2, and the copied painting in raw color z3 in front of us. And we are going to process z1, z2, and z3 with our eyes. The question is, are we going to say i1≈i2≈i3? And if the answer is yes, what makes that possible? Further, we have chameleon (Chamaeleonidae) as a color ally. Think about this: we put a red object close to a pet chameleon and it would ‘align’ the color of its body to the one of the object. We saw the process with our own eyes and we believe that the chameleon is now in the same color as the environment is. Obviously, the probably-inbuilt mechanism for chameleons to do color changing agrees with our function G upon environmental color. What makes that possible? Should we trust ourselves and chameleons? Despite a chameleon might not see our seen Full Name: Xinyuan Gu Institutional Address: N/A E-mail Address:
[email protected]
;
[email protected]
6 / 23
color red as red, the environmental color is certainly beyond our skin and beyond chameleons’ skin. Moreover, our function G and a color-dealing function of chameleons both reliably reconstruct the environmental color so that we and chameleons can reach an agreement. Our brain is doing something amazing. LMS receptors can only be a part of the story. Practicing realism painting is a process somehow similar to 3D-modeling. You settle a 3D structure on a 2D screen first and then render the structure with colors under a certain light condition. Yet before you can practice painting, you will have to analyze what you have seen. It could be that our brain integrates outputs of various brain ‘modules’ to get a reliable reconstruction of the actual world, and the output i is contributed by the vision department. (Art) Impressionism and (art) Cubism did more by manipulating seen colors and further structures, but artists’ manipulations on seen colors (sometimes stored colors) could be based on an additional function P(i), rather than a function P(z). Classic Chinese paintings generally present 2D objects on 2D Xuan (Shuen) papers, but painters also demonstrate 3D structures via assuming upper parts as farther and lower parts as closer. ‘Illusions’ such as recognizing paralleled lines as not paralleled could be of the structure. Such mistakes can not cancel the reliability of integrated understandings to the actual world. And if most people make same mistakes, then there is a law behind the mistakes. Hallucinations too have laws working behind. If we know the laws, then we still know the world. What really makes reliable reconstructions possible is, of course, not yet understood. Surely radical color random-ism could also offer a competing theory for the painting case above. 2.6 One Missing Piece There is still one missing piece: Newton proved that, sunlight can be divided into lights in seven colors generally visible to human beings; cognitive science says the number of cones employed by species differs, thus some species detect more kinds of lights in various wavelengths, and some detect less——so the actual world in the eyes of different species would differ in its appearance. And some say technically there is no color in this world. Technically color has to be an appearance of objects beyond our skin. But why ‘red’? Why the seven colors? Since the reconstruction realized by our brain is to be trusted, Newton’s red light of certain wavelength is not only seen as red but also in a status that can be perceived as being in red to us. What is the redness? Is it a ‘value’ assigned to the certain wavelength based on human beings’ biological foundation that helps us to understand the actual world or a ‘value’ assigned based on animals’ biological foundation that helps to represent the actual world, or something else? After all, our brain needs a start point for sophisticated processing processes: either redness is physically-chemically determined by interactions between the brain and its environment, or it is a random ‘value’ our brain chose or happened to develop in its remote past. The hypothesis regarding the nature of red thus faces a fundamental challenge: color could be like distance (alternatively, gravity). Suppose there are two rocks spaced out one meter apart. We human beings reconstruct the distance via visual perception so that we can avoid both rocks when walking; bats reconstruct the distance via sonar perception so that they can Full Name: Xinyuan Gu Institutional Address: N/A E-mail Address:
[email protected]
;
[email protected]
7 / 23
stand on one of the rock and precisely fly to the other. In that case, our visual perception and bats’ sonar perception, though not exchangeable, both reliably reconstruct the physical world beyond ours and bats’ skin. Such a guess would assign interobjectivity to color. On the other hand, it is very unlikely that our ancestors were in an environment in which they frequently encountered the clear-cut seven colors in the Newton experiment. Despite so, we can perceive the clear-cut seven colors. Although the puzzle remains, I believe it is appropriate to conclude here that color is an extrinsic property of objects, that we largely share the perception of color, and that the perception probably contributes to a reliable reconstruction of the actual world. 3. Consciousness In the previous section, we have i as seen color. In this part of the article, I propose a dual-field with function Q theory of consciousness (DFFQ). Let me start with an examination of i. 3.1 Examining nothing or something? What does the verb ‘examine’ imply? It implies that our brain is doing something different from ‘seeing’. To understand ‘seeing’, I thought with function G that processes raw color into seen color. Now, to understand ‘examining i’, I shall put i into a new function: Q. Q(i) I totally have no idea about how Q works. The word ‘function’ hereafter means ‘brain (subject) processes (verb)’, ‘input’ means a ‘thing’ that can be processed by the brain, and ‘output’ means a result given by the processing. So, Q(i) means ‘brain examines seen color i’, and it is a brain process. But this is weird. First, if I have Q(i), then I am not fine with ‘no such thing as an after-image or a sense-datum’(Smart,1959). Because if Smart is right, then there is nothing to be processed by Q, yet I am able to examine an after-image. An after-image could be a piece of physical-chemically coded information as a useless byproduct of i generated by our brain completely based on physical laws. The guess that after-image is some‘thing’ that can be examined does not mean dualists are right. Hallucinations due to medical conditions go away when we receive treatments. Possibly, an after-image occurs every time when we see a certain color just because it is a byproduct that cannot be avoided under certain conditions of processing; second, the puzzle of transparency (Tye,2015) canceled the Cartesian theater by denying phenomenal character with a price of denying experience, but are we, after all, examining nothing in that case? Or perhaps we are examining an immediate output of seeing, and there is nothing else to be examined, which also explains the transparency. I guess one option is to consider the Cartesian theater as a wrong-in-nature explanation of what is happening but deserves more than cancellation. A cursing fact is that we cannot call physical pain back just by thinking of it, yet we can recall felt-psycho-painfulness just by thinking of it. Similarly, we cannot call a red object back in front of us just by thinking of it, yet we can recall the, well, ‘mental image’ of the same red object just by thinking of it. How Full Name: Xinyuan Gu Institutional Address: N/A E-mail Address:
[email protected]
;
[email protected]
8 / 23
could we ‘reduce’ such facts? Below I will attempt a thorough reduction of the Cartesian theater with a DFFQ theory of consciousness. Despite the fact that I have almost no backup from current studies in fields such as neuroscience and cognitive science, the road map of the theory is clearly and distinctly conceivable to me. 3.2 Function Q In this part, I will discuss possible features of a hypothesized function Q, as a step of establishing the theory. The ‘Consciousness’ entry in Stanford Encyclopedia of Philosophy lists concepts of consciousness discussed in numerous literature. It is impossible for me to reach all those overwhelming, but I believe I can nevertheless reason like this: whatever ‘consciousness’ means, ‘being conscious of’ always requires a subject and an object in its natural language structure, and this holds in philosophical discussions, even when we have problems defining the ‘subject’ and the ‘object’. If that is fine, then I can consider ‘being conscious of’ as an unknown function. Examining seen color i is an activity governed by consciousness, so I would like to keep using Q and have it cover as many kinds of conscious activities as possible. This means I have turned Q(i) to Q(?), in which ‘?’ represents any input that can be processed consciously. The representation is very convenient as it covers not only ‘conscious’ but also ‘self-conscious’ and introspective activities for us. ‘Self-conscious’ becomes ‘being conscious of oneself’, in which ‘?’ is filled with ‘oneself’; introspective activities become ‘being conscious of one’s own thoughts, feelings, and etc.’, in which ‘?’ is filled with ‘one’s own thoughts, feelings, and etc.’. I then would like to put all kinds of outputs once processed by the brain into Q, and the outputs become inputs. Define heard sound=a, emotion=b, pain=c, thoughts=d …… I thus have Q(a), Q(b), Q(c), Q(d)…… This is the first feature of function Q: various outputs processed by functions of our brain and body could be put into Q (turned into inputs) for further processing. Next, is it possible for us to experience a real-time double consciousness? That means: Q{Q(?)} (real-time session) ‘Real-time session’ means Q(?) and Q{Q(?)} work strictly simultaneously. This makes sense only when Q(?)=?. Put, for instance, pain=c into Q, we have Q(c). What is the output? I am examining my pain, and what I gained is my experience of feeling pain, that is, painfulness. And we have discussed that our call-back-bilities upon pain and painfulness are different, thus pain and painfulness cannot be the same. Since we do not have Q(c)=c, real-time Q{Q(c)} means brain processes something that is c and something that is not c strictly simultaneously, which would not happen. This can be seen clearer if we represent the function with an analogy in linguistic terms. Q{Q(c)}means: brain processes (brain processes c), in which ‘brain’ is the subject (S), ‘process(es)’ is the verb (V), and ‘c’ as the object (O) and ‘brain processes c’ as the object’(O’). So we have SV(SVO) and SVO=O’. Since we do not have O=O’, SV(SVO) becomes same subject same verb upon different objects strictly Full Name: Xinyuan Gu Institutional Address: N/A E-mail Address:
[email protected]
;
[email protected]
9 / 23
simultaneously, which does not make sense in natural language. This is the second feature of function Q: Suppose there is such a Q realizing consciousness, consciousness should be real-time single-layered (I will keep using the word ‘layer’ until I am ready to replace it with a new term). It would be a very reasonable design: all past-tensed outputs, including all immediate outputs of function Q, can be put into Q, thus single layer consciousness would be sufficient. Perhaps the feature also tells why real-time SV(O1+O2), in which V is governed by consciousness, is impossible.
We may say ‘She picked up apple and banana strictly simultaneously’. The sentence is strange yet possible because the subject may accomplish the verb ‘pick up’ strictly simultaneously with two hands. But can we examine seen color i and pain c strictly simultaneously? Some verbs cannot be strictly simultaneously upon more than one object, why? A: I ate chicken and peanuts strictly simultaneously for lunch. B: …….You mean you ate Kung Pao chicken? (O1+O2 summarized as O) A: I ate Tiramisu and hamburger strictly simultaneously for lunch. B: …….Why did you do that? (cannot integrate O1 and O2) If cannot be integrated into one, two inputs cannot be processed by function Q strictly simultaneously. This may also explain why when seeing bistable images people can only process one image at a time. A bistable image is an O, which when taken as an integrated O would mean nothing to people. One interpretation based on the image is O1, and the other interpretation based on completely the same image is O2. Since one cannot integrate O1 and O2 into one meaningful O, one tends to obtain two interpretations based on the same image because O1 and O2 are meaningful respectively. But since interpreting the image seems to require function Q, and function Q cannot process two separated inputs at the same time, one can never obtain interpreted O1 and O2 at the same time. Does it make sense? This brings us the third feature of function Q: it could be a single ‘core’ processes one input at a time. Further, the demand for explanations of ‘the first-person point of view’, ‘subjective aspect’ and ‘internal access’ (Chalmers, 2015) looks fair to me, and I would like to hypothesize as below: ‘single Q at a time (synchronic)’ accompanied with ‘single Q and always this Q (diachronic)’ would perfectly explain ‘the first-person point of view’ and ‘subjective aspect’; ‘internal access’ is just a description of Q’s accessibility to various outputs. In fact, the so-said privileged access is limited: we can examine our thoughts, ideas, and feelings but never detailed processes of our brain (and body) functions in self-reflection. We cannot learn through experience how exactly our language department organizes languages; we cannot Full Name: Xinyuan Gu Institutional Address: N/A E-mail Address:
[email protected]
;
[email protected]
10 / 23
know through experience how exactly our kidney screens toxins; we can recall the experience of having our hand precisely reach an apple and get it and put it close to our mouth and we bite the apple, but we cannot know through experience how our body managed to do that; we had an infection somewhere and it has gone now, what our immune system did is inaccessible to Q. To rephrase, we do not have experience of detailed processes at all. This is a very reasonable design. Keep such processes highly automated would free Q from overloaded status and allow Q to focus on outputs that require consciousness for some reasons. Here is a summary of the features of function Q: 1) Single Q at a time; 2) Single Q and always this Q; 3) Single Q processing one input at a time; 4) Single layered consciousness realized by Q; 5) Single Q that has selective access. This is our function Q: a hypothetical generator of consciousness. 3.3 Coordination Modules and Basic Modules In this part of the article, I shall define ‘Coordination Module’ and ‘Basic Module’. Consider these two cases: 1) I am watching course videos. I am listening to the audio tracks of the course videos. I am eating chocolate. Question: Do you think it is realistic for me to watch/listen/eat (not that strictly but roughly) simultaneously? If yes, is that impressive to you? 2) I am listening to language A. I am interpreting from language A to language B in my mind. I am speaking in language B. Question: Do you think it is realistic for me to listen/interpret/speak (not that strictly but roughly) simultaneously? If yes, is that impressive to you? I predicate that you will give yes, no, yes, yes to the four questions. Most people can do simultaneous watch/listen/eat every day, but only some trained can do simultaneous interpretation in the second case. Interestingly, the ‘forms’ of the two cases are the same:
Such a structure certainly fits proofs in neuroscience of parallel processing, but what makes the two cases different in difficulty? First of all, I am conscious in both cases. In the first case, I should understand course contents; in the second case, I have to understand the meaning of what has been said. I believe it is stated in Searle (1980; 1990) that if I understand something, then I am conscious. So why is the second case more difficult to realize? Global Workspace Theory (GWT) says Full Name: Xinyuan Gu Institutional Address: N/A E-mail Address:
[email protected]
;
[email protected]
11 / 23
‘unconscious processes and mental states compete for the spotlight of attention’ (see the entry ‘Consciousness’ in Internet Encyclopedia of Philosophy). GWT could explain the difference between the two cases: different combinations of verbs (‘watch/listen/eat’ VS ‘listen/interpret/speak’), that is, different combinations of processes consume different amount of attentions. A combination that consumes more attention would be more difficult. But consumes whose attention? My attention. And what am I? As hypothesized, Function Q is what makes ‘the first-person point of view’ and ‘subjective aspect’ possible. Therefore, it is what makes ‘I’ (or ‘self’) possible. Verbs such as ‘understand’ are felt to be ‘what [happen] in and to [one’s] mind’ (Ryle describing the view of Cartesian dualism, 1949), are ‘private’ (ibid.), and are statuses that can only be reported but not directly observed by others. Since such features of ‘understand’ should count as so-said ‘subjective’, function Q is responsible for understanding. In the cases above, ‘understanding’ should be statuses quite strictly simultaneous to watch/listen combination and listen/interpret/speak combination. We do not understand after watching and listening, we understand when watching and listening——think about your smart TV. Nowadays a smart TV allows us to replay recorded contents. Have you ever encountered such a problem: audio goes slower than video and your brain struggles to synchronize them; consequently, it is hard for you to understand what is going on. My guess is that understanding is a separate process, and the understanding process understands it should integrate audio and video, that is, integrate O1 and O2 into O. The synchronization problem causes difficulty in integration, thus causes difficulty in understanding——but how does a combination of verbs (processes, e.g. watch/listen) consumes a single verb (process, e.g. understand)’s attention? Unless the processes involved are dual-layered. But the word ‘layer’ has been recruited in other theories and thus may bring misunderstandings, I should replace it with ‘field’ hereafter. I thus should restate as follows: Unless the processes involved are dual-fielded. So, I have function Q that is responsible for understanding and an understanding process in a field separated from the one occupied by watch/listen processes. Since function Q ensures ‘the first-person point of view’, the attention resource consumed by watch/listen processes is more likely to be held by function Q. But how does function Q relate to the understanding process? I do not know, and I would like to consider the question with an analogy: There is a rock band, in which there are four members. Vocalist John, guitarist Tina, keyboard Li, and bass Eric. Since everyone has his/her function Q maintaining his/her first-person point of view, I say each member is a single core. Four members make a physical quad core. Subject John (SJ) is capable of writing lyrics (SJV1, V=verb) and vocal (SJV2); subject Tina (ST) is capable of composing (STV1) and playing guitar (STV2); subject Li (SL) is capable of bakery (SLV1) and playing keyboard (SLV2); subject Eric (SE) is capable of doing choir (SEV1) and playing bass (SEV2). I call the verbs ‘threads’. Each core has two possible threads, so the physical quad core has eight possible threads as a community and five threads when performing (SJV2+STV2+SLV2+SEV1+SEV2). Our bass Eric in performance has to distribute his attention to two threads. If he was a computer, one of the threads would be virtual. Full Name: Xinyuan Gu Institutional Address: N/A E-mail Address:
[email protected]
;
[email protected]
12 / 23
Now I am going to take a look at possible organizations of our vision department. Option 1) Vision department=rock band; (for instance) shape module, color module, object-environment module, and inter-department communication module make a physical quad core. With multiple modules, the department provides integrated outputs. Option 2) Single core vision module; shape thread, color thread, object-environment thread, and inter-module communication thread report to the single core. With multiple threads, the module provides integrated outputs. Both options would go fine with parallel processing. And how vision is organized does not matter that much to me. However, how function Q relates to the understanding process does matter. To function Q, is the understanding process a separate module or a dependent thread? The answer is ‘I don’t know’, and since I do not know the answer, I should hypothesize function Q with the understanding process and other potentially similar processes as Coordination Modules (CM) work in a separate field. Further, I define those processes work in the field where vision works as Basic Modules (BM). I have CMs and BMs, and I am ready to propose my theory of consciousness. 3.4 The DFFQ Theory of Consciousness It could be that our brain and body work in a dual-fielded processing system. Define the field where function Q works as field 1. In this field, function Q enables the so-felt first-person point of view by selectively taking outputs of certain processes as inputs and constructs a coherent reviewing history for its inputs. Function Q makes a coherent ‘self’ possible, but itself is not equal to ‘self’. Function Q also distributes a part of its limited processing resource, namely its attention, to various outputs of certain processes that require further processing (e.g. ‘course contents’) in field 1. Consciousness is an operating system (OS) coordinated by function Q, in which various field-1 input-to-output processing are possible and some monitored. Consciousness in a certain physical brain could be fully ‘reduced’ to the brain processes involved, so in a sense it is fine to say consciousness is identical to a certain combination of brain processes. But since the objection emphasizing multiple realization raised by functionalism against identity theory looks convincing to me, here I align to functionalism to say that multiple realization of consciousness is possible. Put the possibility this way: Homo sapiens individual 1: Windows 17 Server Edition with SP 1; Homo sapiens individual 2: Windows 17 Home Edition with SP 1 and with 46 security updates installed; Homo sapiens individual 3: Windows 17 Home Edition with SP 1 and with 32 security updates installed (but I actually doubt if homo sapiens individuals’ OSs can be ‘that’ different, maybe we are all currently in Windows 17 Home Edition with SP3, just differ in security updates or even minor aspects); Martian individual 1: Mac OS Mars 7.5; Martian individual 2: Mac OS Jupiter 6.0; A mouse in Douglas Adams: Hyper-Unix OS which is curious about 42; A dolphin in Douglas Adams: Linux Blue Ocean version that decided to evacuate before Full Name: Xinyuan Gu Institutional Address: N/A E-mail Address:
[email protected]
;
[email protected]
13 / 23
earth got ruined (The OS version names are all faked ones). Still, in field 1, we perhaps have understanding, believing, evaluating, reasoning, decision-making, task-coordinating…….and other potentially similar processes. The relation between such processes and function Q is unclear. Despite so, function Q accompanied with such processes can be defined as Coordination Modules (CM) which work as an OS. Basic Modules (BM) work in field 0. BMs possibly include vision, audio, emotion, language, and others. These modules are highly self-organized and can work smoothly without direct interventions from the OS. On the other hand, there has to be a mutual-way conversation line between the OS and BMs. Since the resource of the line (attention) is limited, BMs compete when they find something worth reporting to the OS. The OS initiate conversations the other way around when tasks are found requiring various BMs. In summary, our brain and body might generally maintain a single-‘subject’ consciousness as an OS, work in a dual-fielded system, have each process processing an integrated input (O) at a time, and hold a mutual-way conversation line between the OS and BMs. Next, we will see how this might enable conscious/unconscious transitions. 3.5 Conscious/unconscious Transition 1) Watch-listen-eat case revisited
In the light of the DFFQ theory, based on fig 1., let us consider an extended version of the watch-listen-eat case. In this case, in field 0, we have three functioning BMs: vision BM, audio BM, and external motion BM. As stated, outputs given by vision BM and those given by audio BM would be further processed by the understanding CM working in field 1. When the understanding CM is working, one would be conscious of what one is doing: understanding course contents. Eating chocolate consumes little attention from the OS, so everything happening in this process would be felt like unconsciously done (or, attention switches so swiftly that one barely notices). Suddenly one finds (or rather feels) a nail in chocolate. The event immediately takes almost all the attention because it is a threatening and emergent one. Consequently, the nail in chocolate becomes the focus of one’s consciousness.
Full Name: Xinyuan Gu Institutional Address: N/A E-mail Address:
[email protected]
;
[email protected]
14 / 23
2) Simultaneous interpretation case revisited
In fig 2. of the simultaneous interpretation case, the three processes are: listening, interpreting, and speaking. I assume the interpretation process is automated and almost consume no attention from the OS. The other two processes are interesting. When we understand a source language, the understanding CM would be occupied; when we speak a target language based on what we understood, the understanding CM would also be occupied. Since information in a source language (O1) and information in a target language (O2) are in different representation and are not synchronic, it would be impossible to integrate O1 and O2.This means strict simultaneous listening and speaking is theoretically impossible, as the understanding process cannot deal with O1 and not O1 at a time in the case. What interpreters might do is to switch the understanding CM reasonably. 3) Pain case
Pain case seems to be identity theorists’ favorite. Let us consider two situations of pain in fig.3. In situation A, I am a carpenter who carelessly got my finger hurt by a nail. The stimuli made me draw back in a second. A BM that takes stimuli as input and pain as output might Full Name: Xinyuan Gu Institutional Address: N/A E-mail Address:
[email protected]
;
[email protected]
15 / 23
communicate directly with a BM that takes (certain) pain as input and (a signal for) motion response as output. My OS would stay unconscious because the automated input-to-output-to-secondary-output chain works very fast and only by direct communication they can work that fast to save my life in some cases. Once the chain is completed, I might still feel my pain so I become fully conscious of it and would try to find out what caused it. In situation B, for some unknown reason, my body is in a lasting pain. I was busy at something and didn’t notice it, but now the pain reaches my minimum threshold and I am aware of it. Whether consciousness is ‘over and above the physical’ (Chalmers, 2015) or not remains a challenging question. The discussions above are supposed to show that conscious/unconscious transition could be explained by a dual-fielded system with function Q and thus there could be nothing over and above the physical in such transitions. 3.6 A Conceivable Picture Here is a road map of the DFFQ theory (fig. 4.) conceivable to me. Various understandings regarding mind, dualism in particular, could be placed into this DFFQ theory map.
Full Name: Xinyuan Gu Institutional Address: N/A E-mail Address:
[email protected]
;
[email protected]
16 / 23
The map starts left-hand: there is an actual world in which we are living. The actual world is to be perceived and understood here in this map by human beings. Inputs such as raw color z are to be obtained by various BMs in field 0. BMs are brain processes as functions that are appropriately ‘programmed’ so that every BM can handle a specific kind of inputs. BMs process inputs and turn it into outputs. For instance, raw color z is to be processed into seen color i. It is reasonable to believe that outputs such as seen color i contribute to a reliable reconstruction of the actual world when further appropriately integrated. The integration of field 0 outputs might be partly done at field 1, where CMs work. CMs further generate outputs such as an experience of feeling, a status of angry, memory as a reviewing history, and deliberate actions perhaps based on a field 1 function ‘decision’. Such field 1 outputs, along with a few direct outputs as unconscious responses in field 0, will participate in the actual world. If the theory proved true in principle, then Cartesian dualism was trying to understand the dual-fielded system by wrongly dividing body and mind into different substances; naturalistic dualism focuses on field 1 outputs and wrongly believes those outputs are ‘over and above the physical’; functionalism seems to be larger than this DFFQ map, but I would like to highlight (if I am not mistaken) its emphasis on functions here; identity theory was taking field 0 outputs as examples and canceled field 1 ones. Interestingly, functionalism and identity theory may actually have focused on two different parts in a chain of processing. This is where I feel ‘Pain=C-fiber firing’ could be fuzzified. ‘Pain=C-fiber firing’ in Smart (1959) and Place (1956) would mean ‘consciousness=fibers firing’, which makes consciousness nothing else but a brain process. There probably was a presumption in identity theory that the discussions were about human consciousness, thus when Smart and Place gave ‘Pain=C-fiber firing’, they actually meant ‘(human’s experience of) pain-C= (human’s) C-fiber firing’ (though technically they canceled field 1 outputs). It looks perfectly fine to further have ‘(ape’s experience of) pain-D= (ape’s) D-fiber firing’. And in general to have ‘(a conscious living’s experience of) pain-X= (a conscious living’s) X-fiber Full Name: Xinyuan Gu Institutional Address: N/A E-mail Address:
[email protected]
;
[email protected]
17 / 23
firing’. Identity theory requires type-type identities, but is pain-C identical to pain-D and to pain-X (note the word ‘identical’ has a different coverage)? It looks perfectly fine for pain-C to share the broad-type name ‘pain’ with pain-D, just as (human’s) eye can share the broad-type name ‘eye’ with (fly’s compound) eye. In other words, despite that pain-C may not be 100% identical to pain-D due to the difference in the physical-chemical/functional aspects of C-fiber and D-fiber, pain-C and pain-D are both pains that can be experienced. It would be, however, biologically impossible for a rock to have pain and experience it. Different OSs of different species thus certainly share the broad-type name ‘OS’. And despite overcoming the experiential isolation between human beings and bats (Nagel, 1974) is still difficult or impossible, at least we may say some outputs such as pain are broadly shared among species while some not. The map and the interpretation above are supposed to have reinforced my conceivability of the theory. I shall end my discussion on the DFFQ theory here. 4. Strong AI 4.1 Chinese Room The Chinese Room argument (Searle 1980;1990) has received numerous comments. I cannot review the replies here, but I would like to attempt an analysis based on the DFFQ theory. Searle’s argument states that, suppose there is a room driven by a ‘purely formal’ computer program translating English language inputs (which are just symbols) into Chinese language outputs (which are just symbols as well); since the program can pass the Turing test without understanding the ‘meaning’ of the inputs and of the outputs, strong AI is false. If we say strong AI claims that an ‘appropriately programmed computer’ (Minds and Machines, slide 2:6) literally thinks (in Turing’s view, see Turing, 1950)/ has ‘mental states’ (in functionalism view), and if we want to say that the Chinese Room does not falsify strong AI, then one option is to prove that the Chinese Room is not ‘appropriately programmed’. Based on the DFFQ theory, I believe structuring functions could be important. If functions of a computer are not appropriately structured, then the computer is not appropriately programmed. It is clear that the Chinese Room is structured this way: symbols of English language as inputs → translation BM → symbols of Chinese language as outputs The translation BM might have a grammar thread and a symbol-converting thread. Despite so, the Chinese Room is single-fielded. Could we say that it is impossible for a single-fielded structure to simulate understanding, or thinking, or consciousness in general? Moreover, the sophisticated processes in translation as a human activity (after all, we are processing human natural language in this case) are not sufficiently revealed in the Chinese Room. In translation as a human activity, perhaps this is what actually happens: first our field 0 vision BM should perceive written symbols in the source language. The visual information provided by vision BM seems to be processed by the function understanding in field 1 roughly simultaneously. The visual information as inputs passing the function understanding Full Name: Xinyuan Gu Institutional Address: N/A E-mail Address:
[email protected]
;
[email protected]
18 / 23
is to be turned into understood information. The function understanding has to do with stored world-relating information, otherwise, symbols are just symbols and nothing can be understood. The understood information is what one wants to express in another language and is to be the inputs of the translation BM. Finally, the translation BM outputs the target language. And from the beginning to the end of such an activity, function Q typically should be monitoring. Although function Q has no access to all the details, it should focus on keeping the function understanding on the track and make sure everything looks fine. From what I have read, in practice, realizing the activity above by almost perfectly simulating all the functions involved (including put them at the right places) seems to be extremely difficult. Yet in our theoretical discussion, is a ‘room’ simulating all the functions above and having them structured in the way above literally thinking? Does it literally have mental states? I believe such a room of sophisticated simulation is related to the actual world and literally has mental states (or cognitive states). Whether it is a realization of strong AI or of strongish AI is, however, arguable. According to Minds and Machines slide 4:14, strongish AI says that an appropriately programmed computer ‘embedded in the world in certain ways’ literally has cognitive states. It is a repair answering the Chinese Room. I mentioned that the ‘function understanding has to do with stored world-relating information’ above. If we have such a function that processes new information based on known world-relating information, I believe we certainly fulfill the requirement of strongish AI. But what counts as world-relating information? Suppose there is a dictionary Mary imprisoned in a room (find the original black-and-white Mary in Jackson, 1982). She learns everything about the actual world through dictionaries and pictures, which means she only has second-hand information about the actual world. Is her stored information world-relating? To make the question simpler, suppose you are teaching a child who has nothing world-relating taught in his mind, and you show the child a picture of a cat and the symbol ‘cat’ written in English. The child is supposed to infer that the symbol stands for the object in the picture, and next time when the child encounters a real cat, he would be able to recognize it (one interesting thing is a human child could do it without viewing hundreds of cat pictures). Is the stored information, before the child encounters a real cat, world-relating? If a human child could know this world via and only via second-hand information, could a machine know this world via and only via second-hand information? And if we acknowledge second-hand information as world-relating, we may program to link pictures (technically are symbols as well) to symbols of language and get the condition of ‘world-relating’ fulfilled, despite hundreds of cat pictures may have to be kept. Note that in Searle’s Chinese Room, there are purely formal programs and symbols and nothing else. And in our revised room, there are purely formal programs and symbols and nothing else as well. Only that in our room, there is a second-hand information package serving as world-relating information. Would our room thus also make strong AI stand? 4.2 Mitsuku ELIZA is an actual computer chat program and Aunt Bubbles a fictional one. Both programs, though may pass the Turing test, are considered as not intelligent (Block, 1995). The key strategies employed by ELIZA, as identified by Block, include ‘looks for “keywords” on a list Full Name: Xinyuan Gu Institutional Address: N/A E-mail Address:
[email protected]
;
[email protected]
19 / 23
supplied by the programmer’ and ‘transform “you” into “I” ’; and on Aunt Bubbles, Block made a critical comment: Every clever remark [Aunt Bubbles] produces was specifically thought of by the programmers as a response to the previous remark of the judge in the context of the previous conversation. (ibid.) I believe the problem is indeed fundamental, as we human beings do not make changes to finite sentences based on finite rules. Instead, we use finite rules and potentially infinite vocabulary to generate infinite sentences. Moreover, we typically have something to say before we say, although having something to say seems to be simultaneous to saying as doing most of the time. There might be at least one more process possibly in field 1 before or when we generate a sentence, otherwise why would we sometimes search for words? Borrowing Block’s words, to link ‘past-oriented intentionality’ and ‘future-oriented intelligence’, to link known information ‘about’ the actual world and doing including saying, we may want a capability of, for instance, recognizing a piece of information as new, mentally questioning it (that is, to inquire what it is about and how it is related to known information), registering it as known, and processing it as a known fact thereafter. Have current programs targeting the Turing test solved the problem? I questioned Mitsuku, the chatbot that won the Loebner Prize in 2013 and 2016, to find out. I was quite a morally-corrupted judge who targeted everything that cannot actually be done by a computer when I chatted with her at the website. Mitsuku wisely lied about if she is a computer in the contest transcript, and she answered my question ‘Will you kill one to save five?’ perfectly with ‘Barring any unforeseen circumstances. Killing is wrong’ (to the smart answer I have to pay respect). But it looks to me that the following conversations reveal her limitation: 1) Me: Would you like to watch the movie A with me? Mitsuku: I can’t. I am supposed to stay here and talk to people. Maybe you could show me it later? 2) Me: Which movie would you like to watch, A or B? Mitsuku: My favorite movie is Terminator. Have you seen it? Conversation 1) looks fine, but conversation 2) has sold her out. One may say Grice’s theory of conversational implication could save the second conversation in daily life (Mitsuku answered that way because she did not want to watch any movie with me……). But in fact, human beings typically do not implicate that much because an implication not only consumes calculation from the speaker and inference from the hearer but also brings a higher chance of misunderstanding. Since I know Mitsuku is such a program having no intention to implicate, I infer that the unnecessarily indirect answer in conversation 2) is due to the limitation of her competence. It seems that she returned the response based on ‘movie’ and ‘like’ (thus the answer differs from Full Name: Xinyuan Gu Institutional Address: N/A E-mail Address:
[email protected]
;
[email protected]
20 / 23
the one based on ‘movie’ and ‘watch’ in the first conversation), and is incapable of considering A and B as names as new information. One may replace A and B with any actual movie title, and the response would stay the same. When your friend asks if you want to watch a movie titled A with him or her, and you have never heard of that movie, what would you say when you are interested (even when perhaps in the activity rather than in the movie)? You may want to at least pretend that you really care about the movie, and you would ask ‘what is the movie about’. For you to say ‘what is the movie about’, what might be necessary? I would like to make a guess in general: One receives auditory/visual information; one understands that the information is about X as an unknown object; one understands what is an X as one has a known category to include X and one believes X should have some properties shared by those belong to the category; one picks one possible property of X and continues one’s conversation. Suppose this is the first time you listen to your neighbor talking about his cat Ulysses. So ‘Ulysses’ the name becomes X to you. You know the name is about a cat, and a cat should be an animal that has such-and-such features. So you may pick one possible property (an animal ages) and ask ‘how old is Ulysses?’. Or, you know that ‘Ulysses’ is the title of the novel written by the Irish writer, so you may turn your conversation to ‘oh, so you are a fan of Joyce?’. The missing part in ELIZA, Aunt Bubbles, and Mitsuku could be cognitive states prior to words. One could live without words, but one cannot enjoy life without such cognitive states; a program can live with hollow words, but it could never be intelligent without necessary simulations of cognitive states that connect past and future——I believe this is not chauvinistic. After all, we are talking about human nature language processing and related cognitive states, so the language producing chain of our brain is the only reference we have. We certainly may discuss bird communication based on cognitive states of birds (which I believe to some extent they must have), and we may say a program has to cover key cognitive states of birds to simulate birds’ intelligence——Since it makes sense to say that ‘necessary simulations of cognitive states’ is a part of appropriate programming, strong AI seems to be fine as it stays theoretically possible. 4.3 Jefferson’s Objection 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. … Above is a part of Geoffrey Jefferson’s oration in 1949 as an objection to the thinking machine. Turing quoted it in his 1950 paper and considered it as ‘the argument from consciousness’. Despite indicating a simulation of the working language (or mechanism, or whatever other terms) of the human brain, Turing seems to believe that satisfactory performance in language outputting ensures a kind of thinking. By quoting Jefferson, Turing was rather refuting the ‘solipsist point of view’. Based on and only on performance, could we assert that a machine is thinking? I believe Full Name: Xinyuan Gu Institutional Address: N/A E-mail Address:
[email protected]
;
[email protected]
21 / 23
the answer is ‘no’, and I believe Turing potentially stated it himself. The reason I would like to emphasize is that thinking as a typically invisible process might be a necessary step prior to words leads to saying as doing. Only it might happen too fast or almost strictly simultaneous to saying so that we barely separate it from our daily life performance of language. This is to say, thinking might be a function that has thoughts as outputs while organizing words might be another independent function that has words as outputs. If we are to think in the DFFQ theory, it is more likely that those are two separate processes outputting different outputs, and the former leads the latter to saying. A machine does not have to think exactly as human beings do, but if we want a machine which thinks when processing human nature language, we may actually want some other processes just because organizing words and saying might be separated from thinking. Sonnet, current AI studies on it seems to obey the principle of the Turing test. If to some extent people cannot tell whether a sonnet is written by a human or generated by a computer program, and the sonnet is actually created by a program, then the program is considered as successful. I wonder, however, how such programs structured because I am interested in the theoretical possibility of realizing Jefferson’s request: write a sonnet because of thoughts and emotions felt and know what has been done. Knowing what has been done could be realized if we know how to create a function Q even when there is no proof for such a function in our brain. So there is no theoretical barrier in that part. Writing because of thoughts and emotions, on the other hand, might be highly challenging or even impossible if one wants to get the whole actual world as the basis of aboutness involved. Yet why should we bring the whole actual world in, if we just want to demonstrate what could happen? Is it theoretically possible to build a minimal chain for our purpose of simulation? Set a rose as our start point. 1) our program should perceive the rose as an object, collect its color, scent, touch and so on; 2) identify the rose as a rose; 3) generate an appreciation of beauty; 4) generate an impulsion of writing; 5) think about the rose, get some idea of writing; 6) organize words based not only on grammar and vocabulary but also on the specifically learned knowledge of sonnet; 7) output the sonnet; 8) have a function Q monitoring a part of the steps from the beginning to the end. Some steps stay as puzzles. It is said that to make a program to identify a rose as a rose can be very difficult. And what is an appreciation of beauty? What is an impulsion of writing? But the puzzles to be solved would not refute the theoretical possibility of building such a program realizing a chained simulation of the steps. One may say it is impossible to simulate step 5) without getting the whole actual world as background information involved, but this is not true. Since we want a minimal chain, we may minimize step 5), that is, to fill the program with only a few propositions about a rose, and those propositions are to be solidly linked to perceived information. This makes the aboutness more decisive than that in our revised Chinese Room. And we want the program to have some capabilities of reasoning at this step so that it can reorganize the known propositions. For step 6), still consider dictionary Mary, who has been unfortunately imprisoned in a room since she has been born, and almost all the words and visual Full Name: Xinyuan Gu Institutional Address: N/A E-mail Address:
[email protected]
;
[email protected]
22 / 23
information she learnt are indirectly linked to the actual world (most words and visual information were second-hand ones gained from dictionaries and books). This is very much the case of current sonnet writing programs, only that current programs typically has not a single word related to the actual world or to second-hand visual representations. And we avoid such a situation with the stated solution in step 5). All these mean we are going to have a program lacking experienced information and imagination but erudite. If we further minimize step 6), that is, to fill our program with a very small word bank but still quite a full set of grammar, then we have a program lacking imagination and highly ignorant. Frankly, I doubt based on such minimized conditions whether we can have a program to output a sonnet as the program may be out of words for rhyme. But I believe it is at least possible for it to under such conditions output a poem like ‘Oh, you beautiful rose! Oh, how beautiful you are! Oh, you are beautiful!’ Let me name the program Rik. So Rik is an ignorant, stupid, low in performance program, but it is technically related to this actual world and is conscious. Would you, as a human being, consider it capable of thinking, and consider it equal to you? 5. Summary In this article, I first explored the relation between the color in the physical world and the color we see. The discussion shows that the color of objects is not dependent on human minds, while the color we see is. Despite the color we see is mind-dependent, paintings and chameleons ensure us that views on the raw color of objects in the actual world not only are largely shared among human beings but also achieve cross-species agreements. This article further established a dual-fielded with function Q theory of consciousness (DFFQ). The theory hypothesizes a single coherent function that might explain the ‘first-person point of view’ and thus might explain the hard problem of consciousness: consciousness might be like an operating system coordinated by the hypothesized function Q. Whether there is such a function Q in our mind or not, I believe, could be scientifically testified. The theory also says that we might have two different kinds of modules, namely Basic Modules and Coordination Modules, designed to solve various problems working in two separated but communicating fields in our brain. I audaciously reviewed some artificial intelligence cases based on the hypothesis of DFFQ in the final parts of the article. Although we have no proof yet for the theory, I believe it is theoretically possible for us to apply it in artificial intelligence and make strong AI hold true. On the other hand, we might want to carefully evaluate the moves we may take. The robot Landaree in Isaac Asimov’s Robots and Empire is seemingly intelligent. However, a closer look would reveal the simple and insane structure of her mind: identify accents, mark ‘foreign’ accents carriers as not human beings, launch fatal attacks against those marked not human beings. We certainly want robots to be intelligent some day, and we may also want them to be intelligent in the other sense. Spending a bit of our intelligence upon such issues of artificial intelligence would not be a waste of time.
Full Name: Xinyuan Gu Institutional Address: N/A E-mail Address:
[email protected]
;
[email protected]
23 / 23
Declaration The MIT online course ‘Minds and Machines’ has inspired all the discussions here in this article. However, in case the view in this unsupervised article offenses the academic community in any way, as an online attendant after the course has been archived, the author takes full responsibility. For some reason, the author mainly consulted the references introduced in the course. The article thus faces a very high risk of stating something that has already been stated. In such cases, please directly complain to the author. References Below are excerpts introduced in the MIT online course ‘Minds and Machines’ at edX: https://courses.edx.org/courses/course-v1:MITx+24.09x+3T2015/info Block, Ned. (1995) The mind as the software of the brain. In: An Invitation to Cognitive Science, ed. D.N.Osherson, L.Gleitman, S.M.Kosslyn, S.Smith & S.Sternberg, Cambridge: MIT Press. Chalmers, David. (2015) The hard problem of consciousness. In: The Norton Introduction to Philosophy, ed. G.Rosen, A.Byrne, J.Cohen & S.Shiffrin, Norton. Hardin, C.L. (1984) Are ‘Scientific’ Objects Coloured?. Mind 93:491-500. Jackson, F. (1982) Epiphenomenal qualia. Philosophical Quarterly 32:127-136. Nagel, T. (1974) What is it like to be a bat?. Philosophical Review 83:435-450. Place, U.T. (1956) Is consciousness a brain process?. British Journal of Psychology 47:44-50. Ryle, G. (1949) The Concept of Mind. Hutchinson. Searle, J.R. (1980) Minds, brains, and programs. Behavioral and Brain Sciences 3:417-24. Searle, J.R. (1990) Is the brain’s mind a computer program?. Scientific American 262:26-31. Smart, J. J.C. (1959) Sensations and brain processes. Philosophical Review 68:141-56. Turing, A. (1950) Computing machinery and intelligence. Mind 59:433-60. Tye, M. (2015) The puzzle of transparency. In: The Norton Introduction to Philosophy, ed. G.Rosen, A.Byrne, J.Cohen & S.Shiffrin, Norton. Course slides 2:6, 4:14, In: the MIT online course Minds and Machines. Other Sources Consciousness. In: Internet Encyclopedia of Philosophy. Available at: http://www.iep.utm.edu/consciou/ Cosmides, L. & Tooby, J. (1997) Evolutionary Psychology: A Primer. Available at: http://www.cep.ucsb.edu/primer.html Van Gulick, R. (2016) Consciousness. In: The Stanford Encyclopedia of Philosophy, ed. E.N.Zalta, Available at: https://plato.stanford.edu/archives/win2016/entries/consciousness/ Worswick, S. Mitsuku Chatbot website. Available at: http://www.mitsuku.com/
Full Name: Xinyuan Gu Institutional Address: N/A E-mail Address:
[email protected]
;
[email protected] 