James Lantz’ Challenge to the “Information Horizon”

This will be less of a formal “article” and more of a “blog post”. Tonight, following last night’s podcast with Frank Lantz, game designer James Lantz (yes relation) made the following tweets:

 

I thought that these were very interesting objections and I haven’t heard someone object in this specific way before. Essentially, everything James is saying above is completely correct. Does my “information horizon” concept still have any utility?

There are two major factors that feed into the information horizon concept:

  • Not Too Much Information: That the amount of data that players are capable of processing is carefully limited by the game designer so that we are in fact measuring decision-making ability and not just brute force calculation/look-ahead capacity
  • Not Too Much Output Randomness: We want the final outcome of the game to be as meaningful as possible, so as to give the player as much feedback as possible for his win or loss. If you’ve got to-hit rolls in your combat, that may be what determined a win or loss, and not your decision-making. In short, this means you have to play more games to get the same feedback you’d have to get if there weren’t random swings in the resources players have access to.

The “information horizon” phrase was my attempt to say, you don’t want your game to have no randomness (as in Chess or Go), but you also don’t want your game to have too much randomness (as in Hearthstone or Summoner Wars). There is some “point” at which you want the randomness – the unknowable stuff – to start entering into the game.

 

Just A Multiplier

James made the point that [output] randomness is “just a multiplier”. So, as he puts it, “imagine on the 60th move of chess, the king had a 50% chance to transform into a king w/ queenlike movement.” So let’s say you’re on the 59th move. By “multiplier”, James means that there are now two possible Chess futures – the one where the King turned into a Queen, and the one where it didn’t.

I agree! That is one thing that output randomness does to games: it multiplies their possible game states in this way. But that’s not all it does. Output randomness causes intense swings in the final outcome, which means that you now have to play more matches to get meaningful feedback. In other words, the game is less efficient than it otherwise would be.

(And it’s true that all randomness does this, by the way – input randomness does it as well, only less so, and to an extent that is worth the benefit.)

Also, James is talking about a weird thing: a single use of output randomness in an otherwise deterministic system. That almost never shows up in any games. The way it really tends to show up is more like every turn or every few turns. When you keep that in mind, we’re now talking about exponential kinds of multiplication of the game state over and over, which not only means that we have a massively large, but potentially calculable area – back to the original Chess problem where players will be accessing different amounts of data to process – but also we have a ton of interference on the final outcome. You need to play a lot of hands of a really random game to find out who is actually playing better.

What I advocate for with the “horizon” is a point where the designer makes every effort to minimize any further calculation. So, picture fog of war, and monsters/terrain or whatever are coming through the fog, and it’s just way too much stuff to possibly try and “predict”, and there’s really nothing to calculate, so the “calculation contest” thing is taken off the table.

 

Should it be called “information horizon”?

To respond directly to James’ third tweet, it’s true that one random event does not constitute an information horizon. Maybe that’s all I really needed to say in response to his tweets. But, I did think they were thought-provoking, and hopefully someone gets something out of this short article.

Unless someone convinces me that one or both of those above bullet points are incorrect things to want – which is possible – I need a term for this “balance point” of where the randomness will be coming into the game. I could see an argument for it being called the “calculation horizon” perhaps, but “information horizon” also seems fine to me.

  • Paul Spooner

    So, to add what you would consider an information horizon to chess, for example, the pawns could have two different states, “move” and “attack” determined by fair coin flip, with a queue of states maybe five turns long. That way you can look ahead a few moves to plan, but beyond that the possibilities would get out of hand.

  • Technically yeah, but I would advise against “you can look ahead *a few* moves” and “beyond that the possibilities would get out of hand”. The designer should have a “concrete-as-possible” number of moves ahead that a player can look. Also, the “there are just too many possibilities” type of horizon is fuzzier than “I just literally have no information to go on” type, which is why I’d generally advocate the latter.

  • Ben Costrell

    “it’s just way too much stuff to possibly try and “predict”, and there’s really nothing to calculate, so the “calculation contest” thing is taken off the table.”

    I think there is something to calculate, and that is exactly James’s point as I understand it. You can calculate how a move will play out for each of the possible random scenarios, in the same way you can calculate how a move will play out for each of an opponent’s possible moves in a deterministic game. In terms of calculation, what’s the difference between an outcome being decided by a random event and one decided by an opponent’s decision?

  • Thomas Bartscher

    You have to remember that calculation has time and space complexity. Both time and space are practically limited, so for a given player certain problems become unsolvable even if they are calculable in principle.

  • Ben Costrell

    Yeah, of course, but my point wasn’t that random scenarios are solvable, rather that you can apply the same kind of calculation-based decision making to random scenarios as you can to scenarios with an opponent’s input. As James says, “it’s just a multiplier” – “it becomes unreadable for a human not because it is random, but in the same way a giant possibility space is unreadable.”