Spring RTS Engine

A goal that is already half designed for by Epic, you should really join the Epic think tank egarwaen.

We're nowhere close to that stage. And I don't think it's possible for an AI to only be good because it can pay attention to 'everything at once' - that's the case with all the current Spring AIs - none have a screen-sized 'attention-area' or anything similar.

These things must be kept inmind else when that stage does come it'll be near impossible todo.

Also simply programming for improvements is bad, architecture and internal arrangements must be taken into accout else the ai becomes complex unstable and eventually stops working and needs a rewrite.

Again we have common ground AF

Egarwaen wrote:AI-vs-AI play is pointless. You want an AI that gives an engaging game against a human, not one that plays a perfect game or that plays well against another AI.

A good example of this is Commander use. The Commander in a human-played game of TA is balanced by the fact that he requires a lot of micro. Like all micro-intensive units, if you're working with him, you can't be doing other things elsewhere. (Yes, this is a valid balance technique) The problem is that the AI doesn't have this limitation. They can go micro-mad with the Commander and do other things at the same time.

This is a good point.

I've been thinking about the whole issue of AI competitions over the weekend. I reckon that it would still be worth trying to design AIs to beat other AIs - after all, this is a much more easily defined goal than "fun to play" or "provides good practice". I would still like to try to set up an AI tournament at some point in the future, which any AI developer will be welcome to enter if he/she wants.

However, as you say, an AI does have quite an advantage over a human in that its attention can be focused in many places at once. So AIs that are developed to beat other AIs may not be fun to play against - they might be too good! This leads me to wonder whether we might want to improve the abilities of humans to play Spring with a greater level of AI assistance. TA Spring players don't seem to mind the use of AI assistants, as the game already comes with several AI assistants for specific tasks. Perhaps we will need more of these to beat the best AIs, allowing us to concentrate on the things that we are good at (macro-level strategy) while leaving the machine to do all the micro stuff. It's not cheating if all players agree on the rules :).

No AI can beat an experienced player, The average skill level of players is greater than the current AI's. The only exception seems to be KAI + EE at the moment.

And augmenting human players with AI is a bad diea. Why not go for difficulty levels instead?

NTAI URC beat KAI GD fairly easy I gotta say

really?

try KAI URC. Iv'e noticed most AI's break down and cry trying to start up a GD base. If you pit the same AI GdvsURC URC pretty much always wins.

AF wrote:No AI can beat an experienced player, The average skill level of players is greater than the current AI's. The only exception seems to be KAI + EE at the moment.

I think that our differences in this discussion come from a different idea of what an AI for an RTS should achieve. You see an AI as an enhancement to the game - something that a human can play against for fun or practice. As it is a part of the game, it has no purpose except to interact with humans in an entertaining or useful way. For this type of program, a competitive/collaborative development model is not very helpful, as there will always be debate about which features are the most entertaining or useful. Equally, it may be acceptable to allow the AI to cheat in some way, as its purpose is not to win, but to entertain. Your view of an AI is, of course, correct! But there is another way of looking at it.

When I talk about AIs, I am talking about 'programs that win the game without cheating'. An AI that fights other AIs has no other purpose: it's not for the enjoyment of a human, it exists to destroy whoever challenges it. It should be a merciless unstoppable killer that never gives you a break. For this type of AI, competitive/collaborative development is clearly helpful: the fittest AIs will be used to improve less fit AIs, resulting in an iterative improvement towards a more efficient killing machine. The result might provide useful practice for a human, but that would be incidental. Development would be a game in itself - a bit like Robocup, Core wars or T-Robots, but with a more interesting game world, a greater variety of units, and no restrictions on the languages to be used or the complexity of the AI programs.

I know that, btu what I'm saying is that that type of model killed off OTAI, ZcAIN, SAI, QAI, TAI, and 2 KAI rewrites (we're on the third KAI written from scratch).

That's at least 5 skirmish AI projects being knocked off. So out of 9 skirmish AI projects that have made it to the point of public source code or binary releases, 5 where killed off by the system in place, the one you're saying is the best option.

That's a 56% fatality rate of AI's

You argued that better AI's would force lesser AI's to adapt quickly to become better starting an AI arms race. In this case the difference in the AI's was too great to ensure their survival, aka survival of the fittest, species go exctinct, yet we can argue that they wont because they'll adapt, maybe they didnt adapt quickly enough or the change was just too great.

That and an AI arms race is demoralising, demotivating, and destroys any sense of community. Look at the AI community when JCAI ended and the second coming of KAI buggered up when the interface changed and wasnt recompiled, there literally was no AI community for a while, just a coldwar of veylon me and submarine with a krogothe boasting a KAI rewrite of extreme proportions.

JCAI faultered infront fo NTai because of the agressive extreme fast apced development I did in aim of competition at the time.The same competition that bloated the buildtree when KAI came making NTai actually worse

The same cold competitive methodology that drove away 10 or so very experienced people who knew a lot more in the field of AI than probably everyone in this forum put together, they came tried to help and where pressured into leaving by the cold atmosphere that existed as a result of competition.

This cold atmosphere split apart the AI devs too, in the early time there was me and a few people who wanted to start AI's but didnt, then there was Zaphod and a few people he pulled up, for example submarine. I've only come back onto speaking terms with submarine and zaphod/jelmer in the last couple of months.

Again I know what you're saying, but believe it or not, we used to operate like that and we did so for a long time, and it did a lot of damage to the Ai aspect of Spring.

If it werent for me, krogothe and submarine, the 3 AI's that there are today wouldnt exist, the AI forum would have half as many posts as it has now, there would be no Adv Metal AI or ETA AI, no AAI or NTai or KAI, no map exporter, no AI section in the wiki.

Now 2 of those 3 AI devs leading the AI community have turned around and said they dont want the sort of competition. Lots of people experienced in the AI field saw this community when it worked that way and ran for the hills. A request for someone to take over NTai garnered zero applications despite people clamouring to test or help with other non-AI projects. The majority of new AI's released so far are groupAI's not skirmish AI's, maybe because there's no compeition in a groupAI? There's a 5/9 chance an AI will get killed off within the first 3 months of its existence. And finally, after 2 years there are only 3 skirmish AI's left under active development and their development is slowing.

All out competition is bad. Try to emphasise other aspects of the AI's rather than just their difficulty.

For example AAI has a system of sending waves of attack units in plans and patterns, but nobody mentions that? NTai can use nukes and anitnukes but nobody has talked about that.Most people think only AAI uses learning stats (NTai does too and KAI will have it if it doesnt already), or that no AI's do ranged attacks (KAI+NTai), just to name a few.

Jack wrote:When I talk about AIs, I am talking about 'programs that win the game without cheating'.

Win against who? Other AIs? Human opponents?

If you're hoping for an AI that can beat someone as good as one of the WarC or XHC members at AA without cheating, you're in for a long wait.

And finally, after 2 years there are only 3 skirmish AI's left under active development and their development is slowing.

IIRC it was about one year ago that I released the first JCAI wasn't it?
So just one year of actual code development.

jcnossen wrote:

And finally, after 2 years there are only 3 skirmish AI's left under active development and their development is slowing.

IIRC it was about one year ago that I released the first JCAI wasn't it?
So just one year of actual code development.

And did you stop due to the terrible and suffocating competition atmosphere apparently caused by me?

Well back then I was at the anti AF camp remember
And you were only doing metal classes at that time anyway...

But mainly I stopped because I didn't like AI development: debugging takes much longer because you have to run the simulation for a while. I never got JCAI completely bug free, probably due to myself making things simply too complicated, and therefore also too complicated to debug.
And mostly I wanted to go back to my main area of programming expertise/interest: 3D game graphics.

besides: we all should be very happy that jelmer stopped jcai and works on the main engine now since we need more spring devs and he has done a really great job so far :)

well and i disagree with af: i still think competition is useful and kai (and otai in former times) made me think about a lot of minor but fairly effective tweaks and improvements of aai

I agree with Submarine that competition has advantages but I think it should be competition like in a match and not as survival.

AFAIK: the only one whining about competition here is AF, who is clearly not capable of handling that other AI's are getting better than his AI. THe only actual demotivating etc is on his side. Other AI devs go along perfectly well and learn from each other etc. They seem to be actually enjoying this AI making! Now this might say something..

Point here: the discussion should not be about the level of competition shown here, as the only problem here is AF himself. Don't shift focus.

Don't mean to flame, it's just that this shifting part isn't ok. AF; you've done great work! I personally really think you've contributed a lot to this great game and community! It's just that this personal attacks really aren't neccessary.

Well, the only ai devs against competition are AF and probably the devs of most of the AI's that have died off...

Also, it's hardly whining when the man brings up some valid points.

I think we have historical precidence for AI death... now if competition was really the cause or not, I guess we can't be sure, but it seems to me that's the crux of the arguement.

BvDorp wrote:AFAIK: the only one whining about competition here is AF, who is clearly not capable of handling that other AI's are getting better than his AI. THe only actual demotivating etc is on his side. Other AI devs go along perfectly well and learn from each other etc. They seem to be actually enjoying this AI making! Now this might say something..

Point here: the discussion should not be about the level of competition shown here, as the only problem here is AF himself. Don't shift focus.

Don't mean to flame, it's just that this shifting part isn't ok. AF; you've done great work! I personally really think you've contributed a lot to this great game and community! It's just that this personal attacks really aren't neccessary.

Me doing personal attacks? I'd say that post was more of a personal atatck than any other post in this thread......

We've all had our time where we dont feel too motivated. i could go through these forums and lobby logs and find at least 5 instances of jelmer, me, veylon, krogothe, or submarine saying something along those lines. It happens, but it tends to happen more readily here.

The current AI setup with the 3 major AI's is a delicate balance, if one of them faultered and dissapeared then the results would get very veyr interesting.

But really NTai lost its top position quite a while ago. Did you not wonder why I started branching out from NTai? Cookiebot? AFlobby? Darkstars? Epic? all those groupAI? With those ti was nice to implement stuff because I wanted to not because I needed to compete. The need to compete can restrict freedom, as I've noticed many times in the past developing NTai, only to see it then come back to haunt me much later in the project in the form of rewritting entire sections of code.

Yes I agree that some competition is good, friendly competition. That sort of competition happens naturally, encouraging competition past that is bad though.

btw anyone know where NeuronExMachina and greenail went?

I haven't read all the posts.
There are issues with having some competitive metric to measure the "best AI". Many would only design to be the best in that metric and not on providing a good gameplay and user experience. Or others would lose motivation alltogether, since their different purpose AI would be left with little publicity because of the uninformed hordes just downloading "the best AI".
There is a lot more to design to than just "beating other AI:s", and it would be unappreciative and misrepresentative of the AI coders' efforts if such a stomp was held as the _only_ way of comparing AI:s.

The competition is interesting though and it's always fascinating to watch the computers do their stuff. :) Just one has to keep in mind that there are lots of other important aspects. Indeed, computing resource use, crashing, interesting and varied gameplay, mod support. One could of course have some kind of general "AI review" article , but that'd piss of the AI makers too. :) Like having some matrix about features and the AI:s and which is good at which. That'd be more useful than just deciding the best AI in a competition. But still, competitions can be useful and motivate coders. Just if they're not held as end-all be-all.

I have lots of ideas for AI:s but I'd like there to be more than one open-source one so I could contribute.