AI, Where it's at!

[HagNasty]

awwww crap now I have to learn Genetic Algorithms.

[dalyk]

AI is something that overtime is getting more and more complicated, and it is predicted that eventually computers will think for themselves, however is this actually entirely true.

At the end of the day, a computer can only follow a set of instructions, so in essence, it will never be able to communicate for itself. What is your theory on this?

[HagNasty]

There are a lot of ways that allow a computer to learn. The real question is if you have a program that puts together code and compiles it into it's memory, then it is possible for the computer to randomly come up with a piece of code that will let it communicate with itself. Thus giving birth to skynet

[Adrir]

Originally Posted by dalyk

At the end of the day, a computer can only follow a set of instructions, so in essence, it will never be able to communicate for itself. What is your theory on this?

Computers have potential...at the moment they have a limited capacity to learn, adapt and write their own instructions. How many advanced in technology will be required until they appear to think for themselves? Even if it is just simulation?

However, I guess when they get far enough, they would cease to be computers computers and become electronic sentients. Assuming, such an advanced creation wouldn't be used to "compute" stuff ;-)

[dalyk]

Originally Posted by HagNasty

There are a lot of ways that allow a computer to learn. The real question is if you have a program that puts together code and compiles it into it's memory, then it is possible for the computer to randomly come up with a piece of code that will let it communicate with itself. Thus giving birth to skynet

Thankyou for your view, much appreciated, the comment is validly accepted, and very interesting.

So what you are trying to tell is that computers need to know the fundamentals first and then develop from these in order to achieve advanced things, rather than just be programmed to do it.

For example, if a robot needs to cross a bridge it could easily cross i by just moving forward, however if the robot crossed the bridge, and feel down a hole in the centre of the bridge, it should be able to minipulate the code, so next time it crosses the bridge (if not destroyed from falling down the first time), it will have experience from crossing a bridge, and know to analyse the bridge for holes next time it tries it.

Bassically, are you saying that computers need to have the sophistication and the basic tools, to be able to minipulate its fundamental code to compensate for the situation?

[HagNasty]

Originally Posted by dalyk

Thankyou for your view, much appreciated, the comment is validly accepted, and very interesting.

So what you are trying to tell is that computers need to know the fundamentals first and then develop from these in order to achieve advanced things, rather than just be programmed to do it.

For example, if a robot needs to cross a bridge it could easily cross i by just moving forward, however if the robot crossed the bridge, and feel down a hole in the centre of the bridge, it should be able to minipulate the code, so next time it crosses the bridge (if not destroyed from falling down the first time), it will have experience from crossing a bridge, and know to analyse the bridge for holes next time it tries it.

Bassically, are you saying that computers need to have the sophistication and the basic tools, to be able to minipulate its fundamental code to compensate for the situation?

Well a good way to think of it is as a command module controlling the robots remotely. It has the capability to create commands for the robots over a network. That way the first robot falls and is destroyed and the command unit learns from it (command unit is local it would be destroyed as well). This is following the idea of a action / reaction learning process.

Say robot dies crossing bridge. the Command Unit (CU) now write an improved core based off of a simple check

if(!success)
{
improve();
}

the improve function is where all the hard work come into place. It would have to identify areas of the robots code that are downfalls and rewrite the code for them, run them through a compiler and download it onto a new robot.

Then attempt again and check and repeat until it has become better. This adds the idea of physical limitations to the robot. There are some things that are impossible. Like a brick flying under it's own power. If the robot was capable of creating a piece of software that let the robot use tools and expand on it's physical self then the possibilities are endless.

That would be interesting to try and develop. A AI that focuses on self improvement as a goal. The only thing holding it back is the fact that it can't randomly pick options like we can.

We can say lets try adding legs but the computer has no idea about legs. What we really need is a way for a computer to learn practical knowledge and theorize on how to use that information.

[HagNasty]

If anyone wants to brain storm on practical ways to have a computer artificially assimilate information (most likely via the internet) I would love to work on a mini project to create it with you.

Just gathering the information is one thing after we get that down, we can try and figure out a way to use it.

Remember to check you sanity at the door.

[Adrir]

That sounds incredibly challenging. Although I would guess it isn't domain independent, otherwise that would be truely hardcore!