Experimental AI

[ilTallman]

I am seeking a game engine, or development platform that I can test some AI out with. Warzone 2100 might suite that purpose, but I have some questions.

1) Is this allowed, since this is open-source project and under a GNU, should be not a problem, but how many offshoots are out there at this juncture and how would this distract?

2) Is there anybody I can get on board with this, I would be looking for programmer/designers to help integrate the AI engine into this platform. If anybody is interested in assisting, let me know. I will require NDA's to be signed before they get involved, and I would like to keep this local to US citizens if I can.

[Chojun]

If the work requires modifying the game or any GPL'd parts, then you cannot require an NDA to be signed

However, you can do whatever you want that breaks the GPL, you just can't publicly release it (which sounds like it may suit your purpose better).

NIKER and Troman are the resident AI experts around here but as far as I know there aren't any who are US citizens.

Rman Virgil may know more.

[ilTallman]

The object of the fun is not to create a package to sell, maybe for free to demonstrate the AI, but I am not there yet.

Also, the NDA would be to protect our IP, not the changes made to the game. Does that help?

Thanks for the tip!

[Kamaze]

You can do whatever you want with the source code. Print it, burn it, wallpaper your apartment with it etc...
But as soon you redistribute or share it, you have to grant him/them the permissions as stated in the GPL.
That also includes your contributions into the source then. Since the source code is covered by the GPL.

NDAs are somewhat complicated...

However, IANAL

[ilTallman]

Well, maybe it would be best to utilize resources here to guide me to the points-of-interest (POCs) in the game so I can integrate the AI into it for my purposes. If I get something great, we can talk later.

Thanks for the comments!

[DevUrandom]

Well, maybe it would be best to utilize resources here to guide me to the points-of-interest (POCs) in the game so I can integrate the AI into it for my purposes. If I get something great, we can talk later.

Thanks for the comments!

Sourcecode changes would have to be GPLed. As do any changes to existing data (incl. scripts).
Of course you can develop own script code and publish as a mod under whatever license you want. (At least that's how I understand it.)

[Chojun]

As I said before, you cannot attach additional terms, restrictions, provisions, or stipulations to anything under the GPL without the original licensor's approval (the chances of this kind of an approval is remote in the case of Warzone).

The unique thing about the GPL is that it is not possible to retain I.P. rights if you are working with GPL software (that you are distributing publicly), unless you are the original licensor and you can multi-license your property.

In case you are new to or unfamiliar with the GPL, I'd strongly advise you to seek a test-bed/environment/project under a different license (such as a BSD-style license) that will allow you to retain control of your work, should you desire to ever make any more than just alms for it.

EDIT: Actually, DevUrandom makes a good point. If the only changes that you need to make, that you'd like to protect, falls under unlicensed game content (such as a mod, as mentioned), then you'd be safe.

[DevUrandom]

EDIT: Actually, DevUrandom makes a good point. If the only changes that you need to make, that you'd like to protect, falls under unlicensed game content (such as a mod, as mentioned), then you'd be safe.

If you create an own work (in contrast to "a work based on" another person's work), you are also free to license it as you like. As scripts do not link to the game, and thus do not form a "combined work", you can write scripts under different licenses than the GPL.

[Troman]

2) Is there anybody I can get on board with this, I would be looking for programmer/designers to help integrate the AI engine into this platform. If anybody is interested in assisting, let me know.

I'm familiar with Warzone AI and its inner working, but I can't tell if i'm up for a project as long as I don't know anything about it. Can you give out any other details: what kind of ai you are interested in, what techniques you are planning to use etc? Maybe you have some previous work that can give some hints about the direction you are heading?

I will require NDA's to be signed before they get involved, and I would like to keep this local to US citizens if I can.

There are indeed no US citizens in out community with Warzone AI being their main field of interest.

[ilTallman]

Hola All,

Thanks for the info. I will review the license, but again I am not interested in protecting any modifications I make to the game, just the IP I already have which would be incorporated into WarZone via remote (like TCP/IP or .Netting).

The "Big Idea" here is to elevate Game AI a bit and add in some Classical Conditioning (See Pavlov). I also want to create interfaces so we can teach the AI Bot using remote interfaces and have it learn from experience. Eliminating the need for costly scripts. I would like to finally see game AI work on several "Layers" following the Grand Strategy-Strategy-Grand Tactical-Tactical elements of true warfare.

Now, I am sure I will get lots of arguments about the cost savings, effectiveness of this, but thats why I would like to experiment first.

[Troman]

ilTallman, it seems we have similar interests and ideas concerning AI. I'm also using WZ engine as a platform for experiements with AI, while at the same time trying to keep it competetive and fun to play with for the game fans. Though I got carried away by another project at the moment and taking a break from Warzone AI scripting.

Hola All,
The "Big Idea" here is to elevate Game AI a bit and add in some Classical Conditioning (See Pavlov).

Although I didn't do Classical Conditioning, elevating game AI was also one of my primary goals, like implementing natural language processing, adapting as a form of stochastic optimization (in another thread I saw you have already discovered aiexperience.c/h files), cooperation - on a rather basic level, mind you.

I also want to create interfaces so we can teach the AI Bot using remote interfaces and have it learn from experience. Eliminating the need for costly scripts.

This is close to what I'm up to. My current Warzone AI is becoming bloated, apart from using experience, something like an automated planner might be a good tool for certain problems instead of hardcoding everything into scripts.

I would like to finally see game AI work on several "Layers" following the Grand Strategy-Strategy-Grand Tactical-Tactical elements of true warfare.

Currently there are 2 layers in the stock Warzone AI: unit AI, that is hardcoded in the engine and game AI - scripts running for each AI player.

Now, I am sure I will get lots of arguments about the cost savings, effectiveness of this, but thats why I would like to experiment first.

Your ideas actually all make sense IMO.

[ilTallman]

Although I didn't do Classical Conditioning, elevating game AI was also one of my primary goals, like implementing natural language processing, adapting as a form of stochastic optimization (in another thread I saw you have already discovered aiexperience.c/h files), cooperation - on a rather basic level, mind you.

I will have to look at your natural language procecessing, how are you using it? The benefit using Classical Conditioning (CC) is that you get a system which builds actions & responses automatically and can adapt them during game play. It would have to be implemented on the 4 levels I suggested, but even then I think the lower Tactical levels should be hardcoded for better response times.

This is close to what I'm up to. My current Warzone AI is becoming bloated, apart from using experience, something like an automated planner might be a good tool for certain problems instead of hardcoding everything into scripts.

That is one of the goals. Planning here refers to meeting strategic goals instead of route planning, although I have some cool ideas on that as well.

Currently there are 2 layers in the stock Warzone AI: unit AI, that is hardcoded in the engine and game AI - scripts running for each AI player.

What I would like to see is the same unit AI staying the same, moving the scripting to another technology which manages the "player" at the top 3 levels of game play. Thinking even bigger - if this can be done remotely, then you create another component which could develop into a standard for all games. This component could be provided as a separate deliverable and have an instance running for each computer player (PlayBots). Now, think massive-cooporative games and having lots of PlayBots filling to the game play experience. OK, I'll stop now.

I would love the opportunity to work with a game like this to evolve the AI standard to something larger. Even if its loaded as an option.

[Rman Virgil]

------------------>

* ilTallman (& Troman).... you may find the following of interest and relevance:

* Progressive Strategy Project

* Which is directly linked to The Brain 4 Proggy

Cheers, Rman

-------------------->

* I may be mistaken but it would seem the modus operandi of using something like "The Brain 4" prog would be a useful precursor in creating an A.I. along the lines of Madden's "Playbook" in the NFL sim. Which is what I was thinking. The "Progressive Strategy Group" being an instance of Grand Strat > Grand Tac applicable to other domains like G4 Military Doctrine for example.

* BTW - I'm not seeing an EDIT Post function using the latest Firefox otherwise I would have done such with my last post.

- Regards, Rman

---->

* Here I was under the assumption that heuristic and stochastic could be applied within the same domain and that heuristic (even if only implicit) was yet the "bread and butter" in this context.

- Cheers, RV

EDIT:DevUrandom: Merged posts

[Troman]

I will have to look at your natural language procecessing, how are you using it?

My current Warzone AI - Aivolution - is capable of recognizing some commands (like "attack bob", "help me" etc) that can be given either by humans or other AIs. I use flex and bison for processing. First flex tokenizes the input and feeds a bison parser with the output. Bison has the nexessary grammar rules. Look at the chat_lexer.l (flex rules) and chat_parser.y (bison rules) files.

The benefit using Classical Conditioning (CC) is that you get a system which builds actions & responses automatically and can adapt them during game play. It would have to be implemented on the 4 levels I suggested, but even then I think the lower Tactical levels should be hardcoded for better response times.

To be honest at the moment I can't imagine how you plan to apply CC for the higher levels, since usually it is applied to individuums, as opposite to group of individuums, but I have no practical experience with CC so maybe I'm missing something.

This is close to what I'm up to. My current Warzone AI is becoming bloated, apart from using experience, something like an automated planner might be a good tool for certain problems instead of hardcoding everything into scripts.

That is one of the goals. Planning here refers to meeting strategic goals instead of route planning, although I have some cool ideas on that as well.

Yes, navigating through world states using actions that change states instead of navigating through map tiles by moving the units.

What I would like to see is the same unit AI staying the same, moving the scripting to another technology which manages the "player" at the top 3 levels of game play. Thinking even bigger - if this can be done remotely, then you create another component which could develop into a standard for all games. This component could be provided as a separate deliverable and have an instance running for each computer player (PlayBots). Now, think massive-cooporative games and having lots of PlayBots filling to the game play experience. OK, I'll stop now.

This is a huge task. But even if too huge it probably doesn't/shouldn't matter much to true AI geeks.

------------------>

* ilTallman (& Troman).... you may find the following of interest and relevance:

* Progressive Strategy Project

* Which is directly linked to The Brain 4 Proggy

Cheers, Rman

Ok, I had to skip the first one, because I had to learn way too many new concepts in the past days, there is just no room for more. As for the second link I did play with a similar program some time ago (if I guessed its purpose correctly), such programs can help you visualize your ideas. Sometimes I use programs like MS Visio or UML modelling tools for that. I might try that one out, looks interesting.

[Rman Virgil]

-------------->

ilTallman wrote: The benefit using Classical Conditioning (CC) is that you get a system which builds actions & responses automatically and can adapt them during game play. It would have to be implemented on the 4 levels I suggested, but even then I think the lower Tactical levels should be hardcoded for better response times.

Troman wrote: To be honest at the moment I can't imagine how you plan to apply CC for the higher levels, since usually it is applied to individuums, as opposite to group of individuums, but I have no practical experience with CC so maybe I'm missing something.

* I'll hazard a guess and say via Commander led Combat Groups. I'm also reminded of "Black and White".

- RV