AboutChuck Cosby Expertise I can answer questions about speech recognition and natural language understanding. I am particulary strong in knowedge based natural langauge techniques. I cannot answer questions about robotics, nueral nets, prolog, or vision recognition - just speech and natural language.
Experience I have spent 25 years developing natural language software products. I have never developed speech systems, but I have developed sophisticated interfaces from natural language to speech. I have been working with speech recognition systems also for 25 years.
This question seems to be an absolutely essential, and core question, whose answer difinitively separates humans, from what would seem to be all future digital machines with AI. An answer that would explain why a human has a "spirit of something", while a machine with AI does not have a "spirit of something". Or, personal-empirically, it would at least separate me from future digital machines.
The core "spirit" question being, why do colors appear as colors? It is not the mere question, "does red look like red to another person or machine?", but specifically, "why does red look like red?".
If you connect a camera to a digital AI in the future, it can only interpret the sensor data as an array of points, with each point containing a set of numbers. A set of three numbers, when modeling for most humans. Three, for red, green, and blue reception, or three for intensity, hue, and saturation, as you prefer. A machine can tell you that 0 degrees hue, is the string "RED", and 180 degrees hue, is the string "CYAN", but to the machine, it is only a 3 dimensional number, that can be categorized, and it is without the sensory qualia that humans experience as "red-ness" or "cyan-ness" in "color", with no regard to numbers, as the qualities are innate to vision, in biological flesh. So how can a machine truly "see" color, when it is merely a numerical representation? Does the electronic-phosphor of the camera, send a secondary message through the elecctronics of it's systems, showing the very "color-ness" nature of the molecules, interacting with photons, on the sensor, as "seen" by the computer ... I say it doesn't, as only a triad of digitized numbers, in an arrayed stream of data, is sent through the channel, to the digital processor. A machine may, at a higher level of abstraction, have shape data with color attributes like "RED" or "RED-ORANGE" attatched, but these are just strings, and moreso, just a color ID number in the table of color strings. Now a human can look at a shape with the word "RED" typed all across it's surface, but it doesn't make the color of the shape red. Only the photons of light that are red, that strike the retina, and are interpreted by the brain, will do that for a human.
Is there *really* a certain level of number-variable interpretations in heirarchy, that will make a machine actually sense colors, only when processed enough, and brought to some colorful level of interpretation? Or if you add a color monitor, and camera pointed at the monitor, to a machine, and have it sealed in it's core, and then have the monitor display what a camera *outside* of the machine is "looking at", does the color monitor inside the machine, and the color sensors in the sealed camera, at a quantum physical level, give the digitizing camera *inside* the machine, a sense of colors, or is it just nesting the same problem, in a potentially infinite regression, of cameras and monitors leading to the real world colors sensed by the outermost camera, that is outside of the machine?
I believe this is a very important basic question, perhaps with no answers, without a "higher power" of analysis, as flesh with "something-ghosts", or "something-spirit" sense one thing, whilest all future digital machines of finite complexity, will only see numbers, or color ID's, without there being some blurring of the lines, between the digital machine, and the analog flesh. And does this perhaps mean, that to create a thinking feeling machine, we need to throw in an amorphous lump of unprocessed silicon connected to all of the digital processes, in order to blur the machine's sense, and make colors appear to the machine's stone "spirit", at that point?
I can't quite answer this one, but your commentary answer would be appreciated. It's simple, but it's deep.
ANSWER: I generally agree with your conclusions. I have never believed that AI can re-produce human consciousness (which is what you are really talking about). It can only process information in a very sophisticated and tricky way. There may be a deeper level to the universe where our 'life energies' exist. perhaps at some quantum interface level. and it is from here that we sense and feel color and all the other things that make us human. here is a general explaination of AI I like to use:
AI is based on the ability to communicate with a computer. The obvious best way is through speech. It would also be good for computers to be to 'see' as well. The problem with true conversational speech interaction - human to computer - is enormous. This problem has not been solved and won't be for at least 5 - 10 years. What about all the telephone based speech systems that United Airlines and Amtrak and others use? These are extremely primitive systems that can't really converse with the user even in the most basic sense. They ask simple questions and expect simple answers. What is the definition of AI as it relates to conversational speech based interaction? Of course you can see in the movies and on TV what the expectation is. The most extreme was the movie AI where the small boy in the film was an 'android' and could be 'loved' by humans. This kind of thing is centuries in the future if ever. What we would like to see as a first approximation of AI is a functional conversional system for purchasing products (like airline tickets) in a completely natural conversational fashion as if you were talking to a human. This may occur in the next 10 years or so. You might also be able to manipulate other software programs with natural conversation such as Microsoft Office. This will be the primary practical advantage of AI, the ability to control computers and machines.
---------- FOLLOW-UP ----------
QUESTION: -------------------------------------------------------
Likewise, I generally agree with your conclusions, on digital speech centered language processing, as you have expressed. Even a project as vast as Doug Lenant's Cyc, of building and linking the world of words and meanings and relations, in a digital database, would not engender an image of, say, a chair in its database processing "thoughts", but only produce the spartan digital syntactic reltions of it's own database, like that of a chair often having legs on the floor, connected to a seat, and often cconnected to a backrest, but in digital form. A human has actual deep analog visual imagery, in imagination, to internally flesh out the spartan relations recorded by something like the Cyc database, up to a level enabling the seeing of exmples of virtual chairs, in one's human thoughts, which is a feat that a basic Cyc will not have, without being connected to an extended analog databse and internal 3-D modeling program, to give the Cyc database "physical" imagination, beyond the bare skeleton of syntactic relations. I believe this is much of what makes text speech processing so difficult, even if provided perfect digital conversation data, to bypass sound recognition, because so much syntactic relations fall below the threshold of machine "consciousness" today, such that only constrained problems like ticket purchase worlds are *encodeably* feasable today.
As dire against machine consciousness, as both of our statements, and my question, has been, to even the digital no-spirit, versus, human spirit, regarding seeing colors;; I'm compelled to ask a counter question, alluded to in the quantum physics posits, before.
Question being, in the same way as a human connected to a virtual digital world with goggles, can forget that they are in a room, but think they are "centered" in the computer's digital space, as any marathon gamer might attest to;; couldn't a digital computer be given in it's internal systems, a camera pointed at an analog "homunculus or spirit or consciousness center" made of an analog material block for 3-D display, that would make it's own processing existence analog?
The thought being, that the material block can display holographic images under it's control, to help display coherent 3-D color images of its internal logical-digital syntax relations, much like a human can dream, and uses that matter block analog field, to later on, make decisions, and gain a physically expressed understanding of those essentially abstract digital syntactic models, stored in it's digital memory.
Moreover, it appears that some of the machine's processing is literally external from it's digital logical side, and distributed into this block of analog material. This is because the analog material's noisy nature, would alter decisions the machine will make in the future, as it first displays an imagined or actual observed thing, and then later it examines the representation at lesiure in this material display block. Moreso, the machine could even put some of it's own digital variable numbers being analyzed, onto the block, to later get an impression of what it should do. The machine would become a core of digital data and processing, with sections diffusing into an analog block of coherently displayed information, and even processes. Part of it's processing would be in the digital processor, but part would be stored in the slightly noisy analog material. It would be an extended existence, and for extreme cases, much of its fuzzy processing self, could exist in the noisy analog display block, while the digital side keeps or maintains the internal analog activities in digital-logical check, rather than being the sole center of processing.
It is much like an experiment I did pointing a fixed laser at a piece of fixed frosted glass, styrofoam, or sponge, and I saw that their interference patterns were static because the materials were cast and inactive. But when I pointed the laser at a solid sphere of fused bundles of fiber optics, with a crisp optical flow along the fibers, some amount of crosstalk between all of the fibers, and chemical impurities between the fibers, that the light scintillated in active changing patterns, along the core optical axis ... an eerie sight indeed. The scintillation was random and incoherent, but it did carry a "live" random process, within the sphere of fused fibers. A coherent display material block could carry all the more coherent processing than this random scintilating bundle.
I know, I know ! ... I argue against machine spirit in one, and argue for it in the other, but does this slightly-noisy coherent-analog-digital feedback-processing mix sound like a plausible systems description for a future machine with a potential "analog consciousness spirit" that could learn a sense of color in images, or dare say fear in things that might damage the block, achieved through empirically collected data that is processed internally, crosstalked, and fedback;; and all of this, without the problems of infinite regression of internal digital displays and digital cameras, or even fuzzifying digital calculation with random numbers in a disjointed digitally clocked system of existence?
Answer The devil is in the details. You may talk about the fuzzy logic all day long - but what is the exact allogorith? That is what i have to deal with. I need to write exact computer code that creates a fuzzy logic. It has to produce a meaningful result - a human to machine conversation that impresses the human with its flexibiltiy and responsiviness. We can't even do that, not even close. Some people talk about neural networks. I think this is the direction your logic would have to go in. But what is the exact construction of a neural network? Again we have to actually invent it. This reminds me of books by Ray Kurtzwell - he is a very good engineer. But in his books he speculates about things he has no idea how to build. Keep in mind the feedback loop of a human includes (and this is key) the body and all its movements and interactions with the real world. The baby puts everything in its mouth. Why? becuase it a very good way to build up a model of what the thing is and how it behaves. Can your 'analog material' put stuff in its mouth? The path to true AI (which my gut instick says is nearly impossible) has got to be with the neural net (ala 'Data' on Startrek)
