Could we handle an AI?

[nummi]

Hey guys,

The hardest part about this would be defining what the purpose of the ai would be. Once that is accomplished one could simply write a evaluation function of each possible state. Then one could use a Local Search Algorithm or Adversarial Search Algorithm such as Alpha-Beta pruning to achieve that goal. Consider it like playing a game of chess. I don't have time to work on this but I would be interested if someone else did. I would recommend using the CPU and writing it in ZASM.

Nummi

[Schilcote]

Let's just say that not even the CPU works quickly enough to facilitate an entity with the intelligence of an unborn fetus. Regardless of the combined coding finesse of the entire Wiremod community.

That's not really accurate, as an unborn fetus has no intelligence to speak of untill the brain develops, and even then it dosn't do much. I'd say it can handle something about the intelligence of a rather dumb dog.

[nescalona]

Computational power != AI strength != intelligence

[Echo51]

Well, not a true AI can be made, even HL2's aint true AI, yet still does somewhat well.

True AI should be able to find its way though a maze, only by itself, no computer feeding info on the maze, or saving stuff.

and to mention, HP have failbot, which i belive is around 63 lines of lua including all breaks and that stuff its not too good at talking, but its funny at times :P

[BlackholeWM]

Just my thoughts here. Can we create an artificial intelligence that can learn?
tl;dr: let's make robots that make themselves better!

There are many crude ways to do a lot of what AI should do, basically currently the best AI that humanity can create is the one that has the fastest computer attached to it, since we have no idea if one way is better than the other... etc.

So, what should our AI do, if we try for one?
Say we wanted a creative robot that made contraptions.

- observe as players create things, how those things are made and what they do.
- make general rules about how contraptions for various tasks should be made.
- find out what contraptions players like and what contraptions they don't.
- try to create some contraption and see if it works well, then make it better.
- show it to the players, watch their reaction.
- try to make some other contraption.
- decide, a device of what purpose would most benefit the players, and build it.

It should also think in a certain way:
- make random assumptions.
- make assumptions based on observations.
- measure how true are certain assumptions.
- use them later to gather information.
- create rewritable algorithms based on previous information and measure their effectiveness.

hierarchy of motivation:
- want to learn, i.e. create universal knowledge base that is sufficient to solve any next problem.
- want to make players happy.
- randomly choose what to do.

What else should it do:
- try not to lag the server.
- try to use it's memory efficiently.



Contraption itself I suppose could consist of a moongate chip that provides an extended set of world manipulation functions such as spawning any props, and discovering constraints, along with what ranger and target finder and all other things could do.


A radically different way to do this:
Create a hundred E2 gates with "go forward", "turn", "set self color", and "get average surroundings color" physical functions, and persistant "Battery charge". Then create a colored zone that charges batteries, "food zone", and the opposite "poison zone", that discharges them. Every E2 should have random coding. At equal periods of time, All of them are shut down and the ones with highest battery level make children (one with no changes and two with random coding constructs added in random place of their code).
Thus, we have evolution. I basically stole the idea of those some university guys.


What's the point of this thread? Nothing, I'm just sharing my thoughts, maybe someone will have a better idea how to build one.
Don't come here just to say it's impossible.

I was going to do the last suggestion but the adv dupe teleporter has trouble with E2's. If that could be fixed I could do it easily.

[Nikita]

I was going to do the last suggestion but the adv dupe teleporter has trouble with E2's. If that could be fixed I could do it easily.

Have you tried Adv. Prop spawner? It had some trouble but I don't remember if it's good with E2's. If it's compatible then I made a plan for automated contraption duplicator bot in some other thread. Stool turret and prop spawner should be enough to copy just about anything

Computational power != AI strength != intelligence

True.

I think if we made a learning machine, eventually it could do learn how to do tasks instead of us hard coding them in. This could cost alot of processing power. But we would still have enough left to put it on a server with alot of players building and let it observe for a few hours, and then again the next day, and we'd already have some material to work on. It doesn't have to be a neural network to learn. There are many other approaches, for example, rule making.

But doesn't anyone really look forward to seing his computer get better when they are afk?

[Schilcote]

I was going to do the last suggestion but the adv dupe teleporter has trouble with E2's. If that could be fixed I could do it easily.

I can never even spawn it, which is a major setback for anything that uses it. What's the point of an entity that can't be spawned?

[BlackholeWM]

Have you tried Adv. Prop spawner? It had some trouble but I don't remember if it's good with E2's. If it's compatible then I made a plan for automated contraption duplicator bot in some other thread. Stool turret and prop spawner should be enough to copy just about anything

Yes but nothing else has a changable string output of dupe information. I could spawn them but they would all be the same.

[Dragonheart]

IF someone is even remotely willing to try it wouldn't be to hard to make the E2 or whatever ran all of those actions, just make a bunch of them and link them together, like different parts of a brain.

[Tasuit]

I have yet to see any progress on this, the field of A.I is still pretty new, and the only way to suceed with this kind of thing is:

A ) Learn Neurotics, Math, Alot of different theroies on randomness and probally alot of other stuff.

B ) Try.

-
I've been trying for long, making random thingies, etc.
But never have i suceeded, maybe im doing it wrong, maybe not, who knows.