Science Thoughts on Artificial Intelligence

[cjp]

I have been thinking about whether an artificial intelligent computer program could become similar to a human being.

Suppose there is a program that can read pages on the internet, and 'understand' what is written in them. More specifically, it is capable of analyzing the contents of informative sites (such as Wikipedia) to such a level that they can be used to not only expand the knowledge database of the program, but also expand its capability of analyzing other pages. For instance, after reading about image file formats, and about computer vision algorithms, it is capable of developing its own extentions to deal with images on web sites.

Next, suppose the software allows its user to ask questions. Of course, the capability of the program to translate questions into queries for its knowledge database should also be built up by letting the program read the vast amount of texts on the internet.

Could such a program develop self-consciousness automatically?

I don't think so. I think it needs at least two more things.

How could its self-consciousness be tested? We could ask questions about itself, and see whether it is capable of answering such questions. Such questions would typically contain words like 'you'.

How would the program deal with the word 'you'? If it is not pre-programmed to handle them correctly, words like 'you' and 'I' can be very confusing to the program, as they continuously refer to other people. I heard very young children also have problems with that. But maybe, with the help of a dictionary, and some pattern recognition on dialogues, it might find out their function in dialogues.

I think, when a question about 'you' is asked, the software will not recognize it as an attempt to initialize a dialogue with the computer itself. It will more likely give someting like a 'syntax error', pointing out that no context has been given where 'you' has a meaning.

The problem is that the software is nothing more than an observer. It can learn about human beings and how they interact, but it will never recognize itself as an actor.

I think it needs to be capable of acting, and secondly, it needs to be capable of observing its own actions, and the effects of these actions. The fact that the original program can answer questions can count as the capability of acting, but the program did not look back at its own actions, and their effects.

It may also need to have an understanding of time, to correctly interpret the connection between actions and effects. It may need that anyway, because websites often change from time to time: the action of observing has a different effect, depending on time.

Anyway, I'm thinking about developing a self-expanding parsing engine, and I need some discussion to keep my ideas as sharp as possible.

To end this topic starter with a question: what do you think about this? And don't tell me you can't parse that question...

 

[Urwumpe]

Somebody said, the biggest achievement of human intelligence is the ability of being bored.

 

[Ghostrider]

Any program that is built for a specific task could not develop self-awareness as we know it - if we could ever understand what self-awareness is. It should be a general-purpose learning AI which should use a different set of data retrieval programs to enlarge its knowledge base and make its own connections, learn by trial and error and so on. Intelligence and self-awareness are still not very measurable anyway.

 

[cjp]

Somebody said, the biggest achievement of human intelligence is the ability of being bored.

There might actually be some truth in that. At least it is a way to avoid the halting problem. A human being will always halt at some point. On the other hand, a computer, writing its own algorithms and then starting them, might run into trouble. Accidentally starting a non-halting algorithm, without the ability to be bored, is a kind of a mental suicide.

 

[Hielor]

There might actually be some truth in that. At least it is a way to avoid the halting problem. A human being will always halt at some point. On the other hand, a computer, writing its own algorithms and then starting them, might run into trouble. Accidentally starting a non-halting algorithm, without the ability to be bored, is a kind of a mental suicide.

Introduces a whole new virus: the computer ADD virus. Makes the computer unable to stay on any one task for very long. :rofl:

 

[Max Pain]

I have very often thought of the same issue, if self-consciusness would just envolve when a programm achieves some sort of complexity.

I could recommend you the book "Goedel Escher Bach", because one of the main topics of this book is the difference between formal systems (like programms) and our brain, but I'm sure there are more specific books on this topic.

I am used to see consciusness like a spotlight (I have this from some book, but couldn't say which one anymore):
Imagine a big factoriy hall in the dark. There are many robots at work who don't have light to see what they are doing. And there is a certain robot who has the spotlight and he is pointing with the narrow beam in the factory hall, so that just always very few robots at once are in light.

Imho the point of this is, that consciusness is something outside of the system observing the system. No robot in the factory could evovle self awareness. Also the certain robot with the spotlight has no self awareness. He just has the awareness of what the other robots are doing. Of course, there has to be some complex interaction between simple factory workers and robot with spotlight.

Ähm, yeah I am pretty sure that the things I wrote don't make very much sense. However, how are you planning to programm your little project. Do you want to do a neurological network for the learning process?

Oh and always have the movie "Terminator" in mind when doing such stuff

"Of course we have a soul, its made of many little robots."

 

[tblaxland]

There might actually be some truth in that. At least it is a way to avoid the halting problem. A human being will always halt at some point. On the other hand, a computer, writing its own algorithms and then starting them, might run into trouble. Accidentally starting a non-halting algorithm, without the ability to be bored, is a kind of a mental suicide.

Humans can still suffer from a similar problem: OCD.

 

[eveningsky339]

Artificial intelligence can, essentially, only respond to stimuli. While humans are this way on a primal level, we are infinitely more complex emotionally and spiritually. Self-awareness is not a response to a stimulus.

 

[Andy44]

Consciousness, HAL-9000 style, as depicted in 2001, is probably not possible, unless HAL was just emulating human characteristics.

Human consciousness, the human mind, is integrated with the physical body. A disembodied spirit like HAL has no real body, no nerve endings, no taste buds, no eyes, no blood coursing through its brain, no feeling of sexual attraction or a need to reproduce, no primal fear of the dark, no relationship with its parents, and went through no childhood.

According to Clarke, HAL did have an approximation of some of these things, Dr. Chandra as "father", Discovery as a sort of "body", etc. But a machine like HAL will lack the richness of being a flesh-and-blood animal like humans or neanderthals, and is likely incapable of becoming such a high-level thinking creature. At best, HAL might be like a person with severe developmental problems, like an extreme idiot savant or something.

Recently there was an excellent article in Discover magazine about a scientist who believes that the idea of a mind as a spirit seperate from the body is likely in error and a dead-end for developing artificially conscious machines.

 

[Linguofreak]

I certainly think it is possible for a computer/program to achieve the complexity and computational ability of the human brain/mind. As a Christian, though, I do not believe that the brain/mind is responsible for the entirety of human conciousness/self-awareness/whatever you want to call it. (Nor do I believe that it plays absolutely no part). As to whether the existance of a sufficiently complex physical brain/mind is sufficient to somehow cause the existance of a spirit or soul, that is something the Bible says absolutely nothing about, so I really don't know, although my gut instinct is to say that true AI is impossible, but, once again, I don't know.

But I am somewhat puzzled at those who say that the human mind is entirely a matter of physical neurology and chemistry, and computational algorithms operating on that neurology and chemistry, but would nevertheless insist that true AI isn't possible, because I don't see why silicon can't compute anything that brain cells and hormones can.

 

[Andy44]

Well, this discussion gets difficult when you start speaking of the metaphysical aspects of it, but that's hard to avoid when you get right down to it.

I'm not sure if I believe in the supernatural or not, but when I talk science I simply disregard it, since it's not testable and therefore not science.

Nevertheless, I know that I am self-aware, and from the standpoint of me, I am something special in the universe. Consciousness means perception of your environment, and from the standpoint of my brain/mind/soul whatever, the universe only exists because I perceive it with my senses, and only as I perceive it.

It's really hard to imagine that there will be a universe at all after I cash in my chips, and thus lies the seed of the concept of a detatched soul and an afterlife etc.

But from an objective standpoint it's very easy to imagine, and in the case of the deaths of others, obviously the universe carried on without them, whatever they perceived in their lifetimes.

When I was 2 years old I was speaking my first words and learning to walk better. But I have no memory of it, and my current personality was not yet formed. To become the self-conscious being I am today, I had to go through the experience of growing up in a human body among other humans. My "soul" was being formed on the fly; it did not simply get downloaded from heaven to my fetus, and it will likely fade away into unconsciousness when I kick off. If I should contract alzheimers or some other mentally debilitating condition as I age, my "soul" will suffer damage and be altered along with the dying brain it resides in.

There is simply no way to detach who you are from what you are, not completely. Experiences, genetics, hormones, environment, all these things make you who you are, and if we want to build a self-conscious AI, or at least emulate one, we have to understand all this and provide our machine with similar ingredients.

 

[Linguofreak]

Well, this discussion gets difficult when you start speaking of the metaphysical aspects of it, but that's hard to avoid when you get right down to it.

Well, the metaphysical aspects are impossible to avoid, because they're the only thing that can possibly make prevent a machine from being intelligent, as I see it, and even there I have no idea whether they do or not.

I'm not sure if I believe in the supernatural or not, but when I talk science I simply disregard it, since it's not testable and therefore not science.

Nevertheless, I know that I am self-aware, and from the standpoint of me, I am something special in the universe. Consciousness means perception of your environment, and from the standpoint of my brain/mind/soul whatever, the universe only exists because I perceive it with my senses, and only as I perceive it.

It's really hard to imagine that there will be a universe at all after I cash in my chips, and thus lies the seed of the concept of a detatched soul and an afterlife etc.

I'm not sure I do find it so hard to imagine. And the soul is not entirely "detached" in Christian doctrine either. Disembodied spirits playing harps on clouds do not appear in any description of the afterlife in the Bible.

But from an objective standpoint it's very easy to imagine, and in the case of the deaths of others, obviously the universe carried on without them, whatever they perceived in their lifetimes.

When I was 2 years old I was speaking my first words and learning to walk better. But I have no memory of it, and my current personality was not yet formed. To become the self-conscious being I am today, I had to go through the experience of growing up in a human body among other humans. My "soul" was being formed on the fly; it did not simply get downloaded from heaven to my fetus, and it will likely fade away into unconsciousness when I kick off. If I should contract alzheimers or some other mentally debilitating condidtion as I age, my "soul" will suffer damage and be altered along with the dying brain it resides in.

I half agree and half disagree here. Definitely the experiences of early life shape both mind and soul. But I would say that it is the mind that decays in old age, not the soul, because I don't believe the soul is stored in the brain, though it does have a connection to it.

There is simply no way to detach who you are from what you are, not completely. Experiences, genetics, hormones, environment, all these things make you who you are,

Quite agreed.

and if we want to build a self-conscious AI, or at least emulate one, we have to understand all this and provide our machine with similar ingredients.

But here I have to disagree. We don't have to provide anything. It will have experiences and an environment by virtue of existing and having a means to interact with the world. It won't have genetics or hormones, but it's hardware and software, and their structure, will influence how it thinks and who it is. It won't be human, because all the background elements you mentioned won't at all be what they would be for a human, but not being human doesn't preclude it from being self aware. (Assuming that the metaphysical issues I mentioned don't preclude its constructibility by human beings entirely).

 

[Andy44]

I need to find that Discovery article. The scientist in it was building various "brain-based devices". These were machines with sensors and actuators and so forth, but he distinguished them from "artificial intelligence". They weren't very intelligent, but lots of intelligence is not required for self-awareness; think of lizards and frogs, which have a very primitive form of self-awareness, but are not intelligent at all by the standards of a human, or even a parrot.

So this guy wasn't providing all the required elements I named above, he was allowing the machine to learn them itself, the way natural creatures do.

EDIT: Found it! http://discovermagazine.com/2009/feb/16-what-makes-you-uniquely-you

 

[Quick_Nick]

Some of these have the ability to 'learn'.

 

[jedidia]

Only one thing to say on the Topic:

Roger Penroses' book "the emperors new mind" is a rather thoughtfull, if somewhat long, analysis of this topic, in the technical as well as the philosophical aspect.

 

[cjp]

There is simply no way to detach who you are from what you are, not completely. Experiences, genetics, hormones, environment, all these things make you who you are, and if we want to build a self-conscious AI, or at least emulate one, we have to understand all this and provide our machine with similar ingredients.

I used to think this too.
A computer program without a 'body' like ours will never be similar to us. When provided with a different 'body' (e.g. one that only exists in cyberspace, by only allowing it to interact through web interfaces), it might theoretically be intelligent and maybe even self-conscious, but it will be completely different from us.

That does not mean it is necessary for the developer to understand all aspects of human beings, to make a similar-to-human machine. An intelligent non-similar machine can observe human beings, automatically gather the skills to understand them, and then emulate them. With current processing power, I'm afraid it will still be an idiot savant at its best, though.

About the 'terminator' scenario: I intend to keep a tight control on its acting capabilities. For now, no physical actuators, and only communication through HTTP interfaces. In the worst case, it will be like an evil and intelligent human being, being put inside a jail.

And I already finished reading 'Goedel Escher Bach'.

 

[Andy44]

About the 'terminator' scenario: I intend to keep a tight control on its acting capabilities. For now, no physical actuators, and only communication through HTTP interfaces. In the worst case, it will be like an evil and intelligent human being, being put inside a jail.

There's always the "Bladerunner Option": make it so it automatically dies after a predetermined time...you can always recycle the parts. Mwah ha ha ha ha....

 

[jedidia]

Or do it like good ol' Asimov...

 

[Artlav]

The main problem with it is that there is no algorithm to construct algorithms.
And, a digital computer follows algorithms, and is incapable of anything beyond that.

Now, an important part of human intelligence is an ability to make a solution to a given problem, by finding structure in observed things and making the solution's algorithm.
Even more, an ability to understand that there is a problem, and a solution is needed.

Then, think about processing power.
Image analysing is a good example - a modern computer have only one way to process images - by reading each pixel, computing something, moving to some next pixel, etc.
What a neural net does is take the whole array, pass it thru itself, probably with a lot of feedback, and get the result.
Emulating a neural net on a computer doesn't solve the problem - we still do parallel computations in sequence. A parallel machine is needed.

Then, there is another large performance gap - between the analog and digital processes.Think about a radio receiver. When it's near a given frequency, it snaps to the strong transmission on that frequency. A digital machine analysing a signal will have to process it bit by bit, and demodulate precisely the signal on the given frequency.

Better example - filling a random shape with water under gravity.
Digital computer will have to do billions of computations on a grid or something like that, while water will flow into minimum-energy state, doing no computations.

A digital operation is a composite of many fundamental operations, capable of simulating anything mathematically describable, but exponentially slower than these operations, because real things don't compute their states - they just have them.

As you can easily observe yourself, a human intelligence have a large analog part - observe what you do when solving some problem.

So, making a strong AI on a digital instruction grinder might be possible, but making one with any kind of usable performance is the closest thing to impossible.
A human is an entirely different kind of a machine.


Damn, how did this post ended up so big?

 

[cinder1992]

i'm not touching this topic with a barge pole...

DO'H SH--