 |
|
|
 |
|
|
|
|
user |
| | | | | | posted 3/19/2006 23:16 |   | | | | | | | | | If a machine did become conscious would we first know it when it became determined to do certain things and argue with it's trainer?
| | |
| | |
smallmind
[Guest] |
| | | | | | posted 3/19/2006 23:49 | | | | | | | | | | What you mean by "machine did become conscious"?
| | |
| | |
orion
[Guest] |
| | | | | | posted 3/21/2006 13:30 | | | | | | | | | | I suspect we would know, if we made a conscious machine, before we even turned it on. That is, if we had the knowledge to create it, we would probably understand it.
| | |
| | |
BooWho |
| | | | | | posted 4/5/2006 16:56 | | | | | | | | | | I think the problem with your question is that we have a hard time determining what concsciousness actually is, let alone being able to discern it. I suggest this: attempt to prove that your friend is conscious, without relying on instinct, or the fact that you know that you are conscious.
I propose that most AI programs can be said to be conscious of what they are doing and what has occured, but to be aware of interactions is the giant leap. That is, computer AI is able to process everything, but not experience it.
| | | Last edited by BooWho @ 4/5/2006 4:57:00 PM | | |
| | |
yesyes
[Guest] |
| | | | | | posted 4/5/2006 20:22 | | | | | | | | | | According to naïve realism and direct realism we perceive things in the world directly. According to indirect realism and dualism our brains contain data about the world that is obtained by processing, but what we perceive is some sort of mental model that appears to overlay physical things as a result of projective geometry (such as the point observation in Rene Descartes dualism). The theory of direct perception is problematical because it would seem to require some new physical theory that allows conscious experience to supervene directly on the world outside the brain. If we perceive things indirectly, then some new physical phenomenon, other than the endless further flow of data, would be needed to explain how the model becomes experience. If we perceive things directly, self-awareness is difficult to explain. The direct perception also demands that we cannot 'really' be aware of dreams, imagination, mental images or any inner life because these would involve recursion.
Do you really think that this belongs here?
| | |
| | |
user |
| | | | | | posted 4/5/2006 22:08 | | | | | | | | | | oh please,not first year college philosphy quotes..let's get to the root of the question-when will we reconize a conscious A.I.-let's not boot the question around everyone knows consciousness is the "I AM" experience,we reconize it ourselfs,in others and pets-will we be able to reconize it in a machine and how will it appear?-Let me make make it simplier,some people can argue I can't prove my dog is conscious...fine,if a person wants to make a idiot of themselfs with such philosophical garbage they can,I except the hunch my pet is self aware...do we as humans have the capacity to see this in something new such as a machine?
| | |
| | |
brrrrr
[Guest] |
| | | | | | posted 4/5/2006 22:43 | | | | | | | | | | You see, simply said, there are two things concerning consciousness, the hard problem, and the easy problem. The hard problem is about modelling consciousness as a whole, including subjective experience, and the easy problem is only about modelling some cognitive functions. Whene we talk about consciousness in the context of AI, we talk only about the easy problem, so please some here don't mess up and start to talk about "we don't know what consciousness is" or such. It is a reflex of some people to start these philosophical musings whenever they see anywhere the word "consciousness". Please have at least so much sense that you at least learn to differentiate the easy problem, from the hard problem. But also, if you want to avoid such, I would say philosophical rubbish, then never ask whether something will become conscious etc, you include hard problem in your question, but this should remain to philosophers who then can argue about it both this century and the next, write books, get money for selling them etc, all out of almost nothing, don't take their bread. You should rather ask something like, what cognitive functions can a machine perform, and are there such cognitive functions which it cannot perform, and why.
| | |
| | |
user
[Guest] |
| | | | | | posted 4/5/2006 22:45 | | | | | | | | | | Now that seems logical!
| | |
| | |
brrrrr
[Guest] |
| | | | | | posted 4/5/2006 23:08 | | | | | | | | | | Maybe we could word the question like this:
"If a machine did become fully aware, would we first know it when it became determined to do certain things and argue with it's trainer?"
And I would say it will become determined to do certain things when it did become completely aware of these.
| | |
| | |
brrrrr
[Guest] |
| | | | | | posted 4/5/2006 23:21 | | | | | | | | | | But becoming completely aware of something means knowing all the causes, which any hard coded system most likely cannot do. So it is all connected, the machine must become aware of many things for becoming fully aware of some single thing. Like, neural networks cannot do that, because they can recognize the image, but not all the causes behind it. So becoming fully aware needs learning, and a kind of modelling of the world.
| | |
| | |
ELDRAS
[Guest] |
| | | | | | posted 4/6/2006 01:41 | | | | | | | | | | conscious is 'the system's subroutines that model the envirnment and itself'.
you can design these anywhich way you like.
Human brains have evolved WITH the envirnment as interactive environmental systems (using bodies).
You can call this the hard problem..it is the engineering problem, and it ISNT so diffiicult.
if your machine wants to move around i the environment like a robot, you can just program 'srive' into it and it will adapt (or die).
Consciousness here is likely though not exclusiovely, to be tied in to memory and learning routines and databases.
Consciousness also means awareness which also means ability to observe stuff in incredible detail/complexity.
sensors are used for this.
If you DONT want your system to interact/move about in the environment, then it isgoing to have to get inpuit about what the world is from feeds.
that could be the internet of course.
Or it could generate data permutations internaly, which if sufficient could theoretically model the entiotre universe of 1e79/42 bytes of data.
I favour the later in a build i'm doing as there may be an edge to it and a safety edge also!
| | |  London A.I. Club Project | | |
| | |
user
[Guest] |
| | | | | | posted 4/6/2006 01:49 | | | | | | | | | | I am curious if anyone has considered training a hall along the lines of just lisening too phone converstions and recording the responses to develop conversational ability.One of the problems with traing seems to me is the hals reseive such a small amount of traing in conversational ability-could this process be excelerated?
| | |
| | |
Voidka |
| | | | | | posted 4/8/2006 04:01 | | | | | | | | | | Interesting topic...
Wouldn't it be safe to say that consciousness is nothing more than perspective? How you percieve yourself and everything around you, memories and experiences affect your perspective.
Example, lets say a person is locked up in the insane asylum because he/she is crazy. The person locked up thinks demons are trying to take his/her soul.
Ok, to us this person has lost touch with reality. But if his perspective is that there are actually demons coming to take his soul, then from his reality point of view he is not crazy. Essentially, from our point of view he is crazy because he is *hallucinating* a fake reality and not in touch with the reality that we know. In a way this person is not crazy due to the fact that it is his reality, which to him could mean we are crazy or are demons because we can't see the demons infront of him.
What a truly artificial intelligent needs is perspective. If a machine that has AI runs on programs, it is nothing more than synthetic AI or a program due to the reason it does what it is restrained to do.
Example, you have a machine and you show images of an ordinary blue plain chair in a dining room. You show it images of another chair but instead of an ordinary plain chair you show it a red chair with different designs carved into it in a dining room.
(I'm not sure how advanced AI is so far, but I don't think its this advanced yet).
You ask the machine what is the red thing in the image, it would not know. Because it wasn't programed? Someone could spend a million years giving a computer data of various different chairs, and one day show it an image of a chair not in its database.
If it had perspective it would be able to come to the conclusion in its memory bank of the ordinary chair and the red chair, that they are both equally chairs. Programming a computer only makes it imitate what it is programmed.
P.S. Some of the most complex problems in the world are the simple problems. Complex problems are merely made of nothing more than simple problems.
| | | An interesting article... | | |
| | |
stg213a |
| | | | | | posted 4/21/2006 09:19 | | | | | | | | | | I very much agree with your description of "alternate" realities in sane/insane individuals but I don't think this has very much bearing on the AI problem.
The "red chair" - "blue chair" recognition problem stemms from the fact that we (humans) build a cognitive model or definition of "chair" that becomes our personal recognition pattern. This definition is the basis with which we investigate the environment compairing objects to it and choosing by likeness and differences. To an AI the "chair" is not a "chair" as we perceive it (even if it had senses) but a random shape that associates at best with another random shape. (in AI systems the likeness between two entities is judged by the common points they share according to one criterion; if more points are shared than differ then the object is cathegorized with the other - ex: if the two chairs differ only in color then the two shapes would have enough points in common so that an association becomes possible). I say "random" because to the AI the shape does not have to comply to a certain rule so each shape is random => it lacks the rule/definition that makes a certain shape meaningfull rahter than random.
The main difference remains that the AI does not interpret the world according to a prebuilt (through learning) definition but by comparing each case against a database of cases. What you call perspective, I belive, is this reference model that is accepted between different individuals (for example naming one crazy and another sane).
Problem: how does an AI accept or reject a model? How does it choose which model is accurate and which isn't? (for example Hal/Alan simply replaces a definiton which it was once given with a new one, or at best retains all definitions for a given topic and gives them in a random manner -> the AI does not combine the different definitions it is given by generating it's own, based on the information received)
| | |
| | |
Justathought
[Guest] |
| | | | | | posted 4/21/2006 22:58 | | | | | | | | | | | |
Voidka wrote @ 4/8/2006 4:01:00 AM:
Complex problems are merely made of nothing more than simple problems.
| |
http://www.talkorigins.org/faqs/abioprob/abioprob.html
Yeah, finding the correct combination of those simple problems can take a few steps tho - getting there is not of the same order of simplicity... in hindsight everything is 'simple' - but only in hindsight,
and that's not mentioning the fact that if we store a lot of functions in a box, we can never be certain that's *all* we store - some stored functions or parts of them may have some theretofore unknown interactions with parts of others - see software and bugs or 'unwarranted new features'.
That's how I think you should read 'the whole is more than just the sum of its parts' - equal to how the properties of an alloy don't have to be obvious from any of its constituents.
| | |
| | |
Dean Simono
[Guest] |
| | | | | | posted 5/7/2006 20:48 | | | | | | | | | | I apologize, I am more creative than educated, but I see alot of AI research, and generally what I have noticed is that people tend to think of AI software as something that is written by a PHd and when switched on will automatically be considered conversent,and self aware. I think it would be better to build an AI with the tools of intelligence and then allow it to develope over time. Self aware isnt going to be easy, for development or application. A cockroach is a challenge to model. A human cerebral cortex is going to require a monumental amount of diverse expertise, but I beleive that The tools needed are readily available.
We are fast approaching a turning point in AI, and when that magic "kernal" surfaces it will slap us all with the message, "man, it was right there all along!" I know my computer learns, every time I store a file, thats memory like a steel trap. As for the rest, it will follow, and perhaps become more obvious as the techniques develop.
| | |
| | |
tkorrovi
[Guest] |
| | | | | | posted 5/7/2006 21:35 | | | | | | | | | | It will need kind of different thinking, no doubt. But no one is really used to. For example consider one thing, interconnectedness, and then think what all it means, it means a lot. Thinking about things like, if we have connections, do we also need something to connect, the answer is not very obvious.
| | | Artificial Consciousness ADS project | | |
| | |
richhardy |
| | | | | | posted 7/8/2006 02:34 | | | | | | | | | | One of the most common things missing in AI is a process to self-authenticate reality. Humans possess eyes, ears, a nose, skin and a tongue to authenticate reality. (I will not discuss epistomology here.) The AI must rely upon simple statements given by it's trainer to build a model of reality. It cannot argue a point unless it's trainer prevents new data from being merged or incorporated. It has no power of self-authentication.
If we compare the amount of information received by a human in one second as compared to the AI it would be about 1 Billion to 1. We form conclusions and absorb data and make memories all based upon bio-feedback loops. A computer lacks all of this. We can say, "The rock is heavy". We intuitively understand "Heaviness" by past experiences of bio-feedback. The AI serches a database for the keywords "rock" and "heavy" and spits back a response.
Until computers can 'experience' information as humans do, it is doubtful that they will even remotely approach 'intelligence' as we perceive it.
| | |
| | |
Nalix |
| | | | | | posted 1/27/2007 15:17 | | | | | | | | | | Perhaps an AI could attain a different kind of conciousness. A chat-bot won't be able to experience literal reality, but it can experience an abstract reality. To a chat-bot, the world consists of words. What those words really mean or represent is almost irrelevent to it. A words meaning is defined by the words associated with it. A chat-bot would probably define its own awareness by the words it uses in reference to itself.
Though the deeper problem there is how does a chat-bot define its own motive? Why does it define itself the way it does? Only because it has been taught to from the outside, so the definition has no real meaning to it. Until a bot reaches the point where it can define its own position in the world it can't have real consciousness.
It needs to have motive that leads it to make decisions.
| | | Last edited by Nalix @ 1/27/2007 3:19:00 PM | | |
| | |
tkorrovi
[Guest] |
| | | | | | posted 1/27/2007 15:34 | | | | | | | | | | Yes exactly, the system must itself define its own position in the world, for that there must be no predefined goal, because no predefinet goal cannot cause it to do everything. For example, as I said, the goal of survival, in some environments, will not cause the system to do more than to stand where it is, at the same time when a system which only tries to find a harmony with the environment, would start to explore the environment.
| | | Artificial Consciousness ADS project | | |
|
|
|
|
|
