#159 "Difficulty" should not apply to AI

Maybe_Later
open
nobody
None
5
2015-08-01
2012-04-28
No

I propose that various fields of "Difficulty" dialog, especially those affecting the economy, should not apply to AI colonies; they would always have the "Intermediate" difficulty for AI.

I played a game under 0.10.5 where I set all fields (except REF size) to the extreme - for example -999999 indian trade penalty, 100% max. tax, colonies with >0 colonists have -2 production bonus etc.

And the AI performed extremely poorly. Also MPope in IR 3520025 said: "The new AIColony code is
completely untested on that exotic setting. I would expect it to perform poorly."

I think that even if AI had human intelligence to choose the correct strategy (or if the player does not go to extreme difficulty where the AI code works poorly), hard economic settings for AI go exactly against the player's desire: to have a hard game.

The player increases various fields in Difficulty dlg if he wants a harder game. And yet it seems to me that he achieves the opposite: AI colonies will have economic problems, which makes the game easy.

I don't know enough about this (are all aspects of game for AI influenced by Difficulty dialog? Is the military strength of AI higher on high difficulty?) but it seems to me strange that AI is (at least in some aspects) weaker on higher difficulty settings....

Well, I realize that this proposal would be a radical change....

Discussion

  • Mike Pope

    Mike Pope - 2012-04-28

    You realize your proposal amounts to: ``Can you put the cheats that were recently taken out of the AI back in again'' :-)

     
  • Pavel Jelínek

    Pavel Jelínek - 2012-04-28

    I don't understand. I don't mind at all if AI always has the "very easy" setting. My point is that AI should not have harder economic conditions on harder levels - the contrary would make some sense, but I propose that AI has the same conditions on each level. (Then the human player can gain a feel, how strong AI will probably be in 1700 etc.)

    If my proposal is accepted, then AI will have more favourable economic conditions than now (at least on hard levels, or on almost any level if we put AI on "very easy"). This can potentially lead to switching cheats off, but I can't see how this could lead to switching cheats ON.

    Also, I know nothing about cheats. I am not in developer mailing list (if anything like that exists). Should I?

    Therefore I know nothing which is not in the manual or in release notes.

     
  • Mike Pope

    Mike Pope - 2012-04-28

    If you follow the forums there are numerous complaints about the outrageous cheating that the AI does, and the occasional developer or experienced player replying: `The AI is so dumb it needs to cheat to compete'. The developers have long agreed that the AI cheating is regrettable, and ideally it should play by the same rules as the other players. So far, you are the only person asking for the opposite:-).

    The recent rewrite of the AIColony code removed a significant cheat: AI colonies are now subject to the same production bonus rules as player colonies. This was accompanied by also making the AI colony code aware of the production bonus such that it tries to avoid driving it negative. I believe this made the game fairer, and the AI perform better overall in all cases. Unfortunately, you have found the case where it still performs badly (although I maintain it is still better than with the old code, which would most likely have just crashed under the conditions you have been trying). So now what needs to be done IMHO is to teach AI colonies to tolerate negative production bonus conditions better, while still trying to avoid them if possible. There is a *lot* that still needs to be done with the AI.

    What *might* be possible would be to allow setting the difficulty level for each player. Indeed this might be a useful
    thing in multi-(human)-player games to allow a coarse handicap system. I will look into this at some point.

    > Also, I know nothing about cheats. I am not in developer mailing list (if anything like that exists). Should I?

    We do not document the ways the AI cheats anywhere (sometimes not even in the code, much to my annoyance), but they are often pretty obvious. If you have ever seen an AI patrolling your coast with a frigate in the early 1500s, you have seen it cheating.

    All the freecol mailing lists are open, and accessible at https://sourceforge.net/mail/?group_id=43225. Archives are there too, so you can have a look at freecol-developers and see if you want to subscribe. Your bug reports are a significant contribution to the project, so you would be most welcome.

     
  • Pavel Jelínek

    Pavel Jelínek - 2012-04-30

    Do you agree that defeating (or protecting oneself against) the AI Europeans is the main part of the challenge? I strongly think so, because even a weak player can IMHO succesfully defend against natives and build prosperous colonies - and then nothing can stop him from building a big army and defeating the REF.

    If you agree with this, then the game will not become any harder, if I pick a higher diffuculty, because both I and my AI opponents will have the same (harder) economic (and other) conditions.

    Humans are usually more able to find effective strategies against hard conditions, so the game can probably become *easier*, but it will (almost?) certainly not become *harder*. Am I wrong here?

    This seems to me unacceptable, because it prevents the player from having a challenging game.

    - - - - - - - - - - - -

    So this seems to me as a very convincing reason why I think that some conditions for the AI *must* be different than for human player. Even if the only difference would be initial money and number of units.

    But if this will be the only difference, it will IMHO have unpleasant consequences: The player will either have extremely difficult start, or the later stage of the game will be extremely easy. (I think so because it would IMHO take an enormous labour to make AI as clever as an average FreeCol player - or even the smartest strategist who will ever play it. And so the human colonies will IMHO grow faster.) To avoid this, the programmer would have to program the AI to avoid unprovoked attacking in early stages - but then we will have a paradox like Warcraft I : the computer will be labelled very stupid for not throwing the giant initial army immediatelly against the human player.

    This is why I think that the difficulty settings should contain:
    - initial units count for AI; I don't care if it will be complex - or if the player just says which multiple (of the initial unit set) AI will get - for example 5 caravels, 5 pioneers and 5 soldiers.....)
    - some settings that affect the growth rate of AI colonies.

    As for the later, there are many possibilities, but a simple one would require just one field: "Population growth factor". If this factor is 130%, then whenever the AI should gain a new colonist (in any other way than capturing an enemy), it has a 30% chance that it will acquire an additional one. Of course, this number can be well above 200%, if someone wants a really quick growth of AI.

    Of course, this can lead to a stupid situation where the AI player is very populous, conquers a very large part of the map, but his army is ill-equipped and he has almost no buildings and FF. If someone dislikes this, then we can add fields "AI buys in Europe for 70%" (70% is the example value that the player can choose) and "All indoor and outdoor professions of AI have +30% bonus). I think that this would make the advantage for AI equally distributed into all game aspects (army size, buildings, FF's etc.).

    - - - - - - - - - - - - - - - - - - -

    Well, I am tired tonight, so I may have overlooked a disadvantage of my proposal. But I absolutely can't understand why any difference in conditions (between humans and AI) is automatically called "regrettable cheating". I think that I have proven above that some kind of difference is neccessary - and if it is clearly defined in "Difficulty" dialog, then noone can blame the developes for "secret cheating" (and noone is forced to switch it on....)

    Pavel

     
  • Pavel Jelínek

    Pavel Jelínek - 2012-05-01

    I think I was wrong about the field "AI buys in Europe for XX % of price", because:

    1. It is needless (it was intended to let AI buy more tools and muskets, but this will be accomplished by the field "All indoor and outdoor professions of AI have +YY % bonus", which will let AI produce more furs or coats etc. and therefore earn more money).

    2. It would lead to some paradoxes (for example: if we don't want to let AI buy cheaper than sell, then we would have to decrease both prices, but then AI would have difficulty to earn money).

    So I renounce the proposal of this field. And I hesitate if "Population growth factor" is neccessary. Maybe the field "All indoor and outdoor professions of AI have +YY % bonus" would be sufficient and would distribute the advantage evenly among various game aspects.

    (A small step for the programmer, a big leap in the game challenge...... :-) )
    Pavel

     
  • Mike Pope

    Mike Pope - 2012-05-04

    > Do you agree that defeating (or protecting oneself against) the AI
    > Europeans is the main part of the challenge? I strongly think so, because
    > even a weak player can IMHO succesfully defend against natives and build
    > prosperous colonies - and then nothing can stop him from building a big
    > army and defeating the REF.

    I only mostly agree here. I think the native AIs need improving to
    the point that they are a much more serious threat.

    > If you agree with this, then the game will not become any harder, if I pick
    > a higher diffuculty, because both I and my AI opponents will have the same
    > (harder) economic (and other) conditions.
    >
    > Humans are usually more able to find effective strategies against hard
    > conditions, so the game can probably become *easier*, but it will (almost?)
    > certainly not become *harder*. Am I wrong here?

    ATM no. With better European AIs, who can say?

    > But if this will be the only difference, it will IMHO have unpleasant
    > consequences: The player will either have extremely difficult start, or the
    > later stage of the game will be extremely easy. (I think so because it
    > would IMHO take an enormous labour to make AI as clever as an average
    > FreeCol player - or even the smartest strategist who will ever play it. And
    > so the human colonies will IMHO grow faster.)

    It will require a lot of work but I do not share your doubts that it
    is possible. Several people have mused that if FreeCol had a stable
    AI API that it would not be unreasonable to use machine learning
    techniques to automatically create a powerful AI. There is a fair way
    to go before we are at that point however.

     
  • Mike Pope

    Mike Pope - 2015-08-01
    • Group: --> Maybe_Later
     

Log in to post a comment.

Get latest updates about Open Source Projects, Conferences and News.

Sign up for the SourceForge newsletter:





No, thanks