Spring RTS Engine

McLoud wrote:Some AIs can learn and then with some statistic figure it out when its good or not (and at what times)

knorke wrote:I think Lua AIs made deliberately for one game are the way to go.
Games are using Lua for game logic more&more and if other AI interfaces have no way to read/controll that, they can not play the game. ie just taking zeroK: non-Lua AI could not know about mex overdrive, commander choice/upgrade, jumpset, energy grid,...

yuritch wrote: AI "learning" (as seen in a few Spring AIs) is about a completely different thing, they can't learn anything about commands. Basically learning just tells the AI which units to build, and NOT how to best use them.

• The lack of information arguement applies to Lua AIs too, your just pretending it doesnt because you have prior knowledge
  
  e.g.: Most native AIs specialising in 1 game, or configurations could have added morph support, this isnt because its knowable, heck most of the OTA stuff wasnt knowable to begin with, e.g. AIs and the Juno unit in BA, but if we know what to implement and how to do it, then we can get some of the way there.
  
  The same is true of Lua AIs, the main difference being that the creator of the game usually has a hand in them and make hardcoded modifications. This luxury is usually non-existant with native AI developers purely because game developers expect native AI devs to develop for, not with them.
• There are no learning AIs, and never have been. Submarine was hyping AAI when he said it 'learned'. What he ( and veylon and myself _ implemented was statistics based logic, based on statistics that got updated as games went on.
  
  The prime example of how silly it is to call it learning is KAI. KAI had an identical system designed into it, but instead of using dumb values and honing them in over time, it just calculated a good guess based on an equation that had been tested. It didn't start off dumb and improve over time, it was just as good first time as last time.

• Basically learning just tells the AI which units to build, and NOT how to best use them.

  This is correct, but not as extensive enough. It has to be realised, that even the basic TA commands sucha s capture etc have this issue. I had to code capture and cloak and dgun into NTai, else it would never use them. The same would be true of morph / teleport / jumpjet / research etc

• Native AI developers don't trust any content developers to work with them, so they develop for as many games as they can, or TA
• There are no dedicated AIs for games outside of TA.
• Popular anti native AI sentiment is leading to anti-competitive behaviour and a rise in the quick fastjob lua AI, and people are blindly seeing the quick lua job as superior to the 6 month C++ AI purely because its lua without even considering its flaws and merits.
• People are ignoring the fact that the same bus factor issue that affects native AIs also affects lua AIs. This is concealed behind a larger population of content developers, whereas its obvious for native AIs due to the collapse of the native AI community here.
• Zombie native AIs
  
  AAI/RAI/KAIK are dead, yet they are being extended beyond their natural lifespans, much to the detriment of AI and the engine as a whole. AAI alone is mentioned quite a lot, normally because it never works with anything. RAI works just enough at nonBA to tempt you then fail at playing a critical yet simple part of the game. These are limbo zombie projects
• Anticompetitive Developer policy that alienates the prime userbase for native AIs.
  
  e.g. "We don't support visual studio AIs" says Mr A to the AI crowd that almost exclusively uses visual studio. We can maintain our own projects thankyou, but the utter ignorance of this simple fact has alienated and crushed the motivation in a lot of people to continue or start AI projects.
  
  This has contributed heavily towards the dwindling in native AI developers over the last 3 years. Native AI developers cant fulfil content developers demands when they're near non-existent in presence.
• Content developers see a lack of progress in native AIs to support their games purely as a sign that native AIs will never suffice
  
  When in actuality, a lack of help from them, coupled with a non-presence in the AI community is actually to blame.
  
  No matter how hard you shout, you cant make factory workers in a factory work faster if nobodies there.

• Spring games change too much. By the time you've ran your experiments, your game is already too old to run under the latest version of spring, and the updated version has balance changes that ruin your controls. Games like starcraft broodwar are stable in terms of gameplay and little changes from week to week.
• The recent behaviour of Hoijui
• Lack of documentation
• Disregard of Visual Studio ( students cant always dual boot into a linux configuration, nevermind gain the privilledges to setup the spring build environment to build AIs )

There is interest in spring in the AI community at large but these issues prevent us from benefiting from a flood of academic and hobby interest: ...

people are blindly seeing the quick lua job as superior to the 6 month C++ AI purely because its lua without even considering its flaws and merits.

There are no dedicated AIs for games outside of TA.

So if Lua AIs suffer exactly the same problem native AIs do, then why are people advocating dedicated lua AIs?

A follow up release will be made at the weekend with extra API additions such as game:SendToContent(luatable) for gadget communication, and anything anyone puts forward between now and then.

Content developers see a lack of progress in native AIs to support their games purely as a sign that native AIs will never suffice.

When in actuality, a lack of help from them, coupled with a non-presence in the AI community is actually to blame.

Popular anti native AI sentiment is leading to anti-competitive behaviour and a rise in the quick fastjob lua AI

As for *A only, does nobody remember NTai running with Hubs in EE? NTai playing Nanoblobz?

attackerlist = {
	"bit",
	"bug",
	"bugold",
...