Wednesday, July 21, 2004
Its possible
AI is possible. Its also easy.
The complexity if it actually exists is only in simulating or building a program that wants. Want drives us all. Idea of like versus dislike. If I can tell a program to like something, then I can teach it to want it.
Write a program that has 2 keyboard inputs. One of them is called joy (lets say the button 'j'), and one of them is called pain (lets say the button 'p'). If i push joy, then the program is happy (i'll define that later), and if I push pain, then the program is sad (and this). If I give the program the ability to input either of these keys (using any number of APIs), then I can let the program continuously be happy. This of course has one simple problem. AI is not just a program doing what makes it happy, but a program learning what makes it happy. How do you do that?? Well I can't necessarily say how from a program's point of view, but I like to try and look at a baby and figure out how a baby learns.
A baby is born (we are assumming no 9 months in the womb here...), and it doesn't know anything. Lets assume that it is hungry. It would now be feeling pain, so it would do whatever it can do to make that pain go away. It cries. The baby gets fed now, and finds that it is happy. How does it learn that it is happy when being fed? The baby has now learned that to feel happy, it needs to do whatever it did this first time, ie to cry.
How does this compare to a computer program? A few things were left out above, one of which is how does a baby know to cry? Interesting question... Well let me propose that a baby as an adult has certain involuntary reactions to different causes. Lets say that a computer has the same. I will tell the program that when it is in pain, it should cry (ie print the letter 'c'). When I the person writing the program sees the program print cry, I will soothe it by pushing joy. When it is no longer in pain, it stops printing cry, and I stop pusing joy.
What am I trying to do with this? I want to teach the program to crave joy. Perhaps this is too simple for AI, but what happens if we change the program's involuntary reaction to something different. Now when the program is in pain it prints either cry ('c'), sleep ('s'), or wait ('w'). Well what will the program learn now. It will learn that I will only push joy when the program prints cry.
Add to this an algorithm of simple proportion that sees what it printed, and sees what the programmer's reaction was, and therefore learns to print cry all the time, thereby keeping it from pain forever. If this can be achieved, then we have reached AI.
Of course the program would have to have a lot more involuntary reactions, allowing it to realize it has them, and allowing it to learn to control them. It would also need many more 'feelings' other than joy and pain.
Thats all for now.
Nachum Kanovsy
The complexity if it actually exists is only in simulating or building a program that wants. Want drives us all. Idea of like versus dislike. If I can tell a program to like something, then I can teach it to want it.
Write a program that has 2 keyboard inputs. One of them is called joy (lets say the button 'j'), and one of them is called pain (lets say the button 'p'). If i push joy, then the program is happy (i'll define that later), and if I push pain, then the program is sad (and this). If I give the program the ability to input either of these keys (using any number of APIs), then I can let the program continuously be happy. This of course has one simple problem. AI is not just a program doing what makes it happy, but a program learning what makes it happy. How do you do that?? Well I can't necessarily say how from a program's point of view, but I like to try and look at a baby and figure out how a baby learns.
A baby is born (we are assumming no 9 months in the womb here...), and it doesn't know anything. Lets assume that it is hungry. It would now be feeling pain, so it would do whatever it can do to make that pain go away. It cries. The baby gets fed now, and finds that it is happy. How does it learn that it is happy when being fed? The baby has now learned that to feel happy, it needs to do whatever it did this first time, ie to cry.
How does this compare to a computer program? A few things were left out above, one of which is how does a baby know to cry? Interesting question... Well let me propose that a baby as an adult has certain involuntary reactions to different causes. Lets say that a computer has the same. I will tell the program that when it is in pain, it should cry (ie print the letter 'c'). When I the person writing the program sees the program print cry, I will soothe it by pushing joy. When it is no longer in pain, it stops printing cry, and I stop pusing joy.
What am I trying to do with this? I want to teach the program to crave joy. Perhaps this is too simple for AI, but what happens if we change the program's involuntary reaction to something different. Now when the program is in pain it prints either cry ('c'), sleep ('s'), or wait ('w'). Well what will the program learn now. It will learn that I will only push joy when the program prints cry.
Add to this an algorithm of simple proportion that sees what it printed, and sees what the programmer's reaction was, and therefore learns to print cry all the time, thereby keeping it from pain forever. If this can be achieved, then we have reached AI.
Of course the program would have to have a lot more involuntary reactions, allowing it to realize it has them, and allowing it to learn to control them. It would also need many more 'feelings' other than joy and pain.
Thats all for now.
Nachum Kanovsy
