Get an Epic Experience with Premium
  • posted a message on AI & Balancing

    I watched your MineCon presentation and are truly amazed by Scrolls' potential.

    Regarding the issue of balancing which you briefly touched upon -- it looks totally daunting if you consider that there will be hundreds of cards plus new mechanics and tweaks added constantly -- I can offer a solution.

    It's probably best to start with some background information. I am in charge of an indie project with a focus on developing groundbreaking AI, our first application and showcase being Heroes of Might and Magic V. If you are not familiar with HoMM V, it does feature a good number of factions that each have more than a score of individual units with different stats and skills. In addition heroes (the army leaders) also can have a wide variety of skills and feats that alter gameplay considerably, plus an abundance of structures on the map and a powerful artifact system that again impacts what heroes and units can do. Everything is mathematically defined and turn-based. In short the complexity is comparable to what Scrolls might achieve one day.

    Balancing for Heroes so far is rather obscure. The number of possibilities in the game is mind-boggling, and balancing so far has been approached as an art performed by people with outstanding intuition, helped by mathematical evaluations, and refined by a rigor of playtesting and player feedback. This works to an extent, but even after years there is still discussion about inbalances and exploits.

    What I am doing is to approach the problem from a different angle. While AI so far has only been used to evaluate potential actions and arrive at a course of action, I have been working to develop an AI that understands the entire framework and how every element impacts each other. Technically an AI that recognizes strategic leverage in all its forms. This is a formidable mathematical undertaking, a goal that looked for a long time unsurmountable, but the solution I have now solves most of the inherent challenges, and the project is on track to release a first practical application soon.

    The one thing that is interesting about this, and of potential value for Scrolls, is that the most significant benefit is not stronger AI opponents but that this AI can evaluate every possible setup and identify to which extent and with which probability each element favours each party. As a result this AI can evaluate the entire range of possible options (i.e. the infinite set of possible games) and identify whether there are elements and combinations that will most likely win, and to which extent they outweigh other potential actions, in effect deciding whether these are redundant or not. Ideally the players desire a system in which each option on offer is valid and has merit. For this the AI can ensure that the entire system is balanced, i.e. that no options are overpowered, or if any that these are at least exceedingly rare. The same is true for any cards introduced later on, that the AI evaluates the full impact of its features and changes to the mechanics on the entire game, and either approves them or suggests modifications to maintain the balance.

    This is the solution I am suggesting to you for Scrolls, that you develop an AI that is capable of evaluating the entire range of options and probabilities and whether the presence of any possible combination will effect the outcome of the game. I am not sure to which extent you have mastered the necessary higher order functions of AI to make this happen, but I can offer advice and know-how if you are interested.

    Last but not least, I am very much interested to try my hands in Scrolls and would be very happy to participate in the Scrolls alpha. Maybe I can offer more insight if I have a good first impression of the current balance and features in Scrolls.

    Posted in: Suggestions
  • To post a comment, please or register a new account.