Search Unity

  1. Megacity Metro Demo now available. Download now.
    Dismiss Notice
  2. Unity support for visionOS is now available. Learn more in our blog post.
    Dismiss Notice

Predicting the future - relevancy to the user

Discussion in 'Game Design' started by imaginaryhuman, Jan 31, 2015.

  1. imaginaryhuman

    imaginaryhuman

    Joined:
    Mar 21, 2010
    Posts:
    5,834
    Ok so each player is different and each player has different interests, needs, requirements, learning curve, level of excitement, attitude, patience, values, etc. An ideal game would be very sensitive to the user, understand the user, adapt the experience to the user, and create the best experience for THIS user, with no regard for any others. This is all about relevancy.

    A few parallels....

    In the world of e-commerce, it used to be you could throw one website experience at everyone and treat them all the same, and let the user be the one to adjust. Over time this was proven to not be an ideal fit, and so we started to see personalization, websites showing particular products based on your behavior, your history of browsing, what you showed an interest in. Maybe sophisticated sites even profile your personality and watch your performance to adjust what content is shown to you and when. Email marketing started to emphasize segmentation - recognizing that different audiences have a different set of interests, and some things that are relevant to them are not relevant to others. In essence, web designers are trying to read the minds of the users and to understand the user very closer, on a one-to-one basis, and provide each with a customized, ideal experience. Ultimately this kind of experience is more interesting, compelling, engaging, acceptable, and ultimately leads to more sales and retention.

    Similarly, think about data compression. You try to squeeze a file down into fewer bytes to save space. Maybe it's an image with pixels in it. In compression algorithms, it's all about trying to most accurately predict what the next piece of data `might` be, because the closer your prediction is to the actual value, the more actual data you can omit. e.g. you can just keep a record of the difference between the real and predicted values, which are generally smaller numbers. The better you can get at predicting the future, the more adaptive the algorithm is, and the more effective the compressor is.

    So think about this in terms of designing a game. How can you create a game which can SENSE the specific user, and adapt itself to this specific user so closely that this user has the most ideal experience that THEY want to have. I think we're already doing this in some ways... e.g. offering difficulty levels, unlocking certain features or content if the player shows enough skill, providing continued new interesting things to do for those that play longer, that kind of thing. Also analytics about how people play the game can be a good source of feedback which can let the design be iterated on.... perhaps though still targeting that `same size fits all` most ideal single product, unless the analysis can show you some segmented demographics - different personalities or profiles or types of player that you can learn to target.

    I guess my point is, shouldn't we be trying to make games adaptable, morphable, sensitive to the player, much more dynamic in the experience it generates based on the individual player, rather than trying to create a single-design-fits-all kind of game? I suppose one argument against this is knowing and picking a specific audience and targetting them... that's a good start... being sensitive to a certain set of people... but what about making games more dynamic at runtime so that they can adjust much more to the player, whoever they happen to be? Dynamic AI? Procedural realtime level modification? Making a unique game so crafted to the individual that it's practically as though you made the game JUST for that person? Really it's about predicting the future - predicting what the player is most likely to respond to next, and presenting it to them, and the closer you can get to an accurate prediction, the more you can provide exactly the most optimal experience that this player will enjoy.

    Your thoughts?
     
  2. JoeStrout

    JoeStrout

    Joined:
    Jan 14, 2011
    Posts:
    9,859
    Yes, this is something I've thought about too. I think an important start is a machine learning algorithm that models the user — predicting where the user is going to have difficulty, where they choose to spend more time, etc. Then you can dynamically adapt the game to enhance the player's experience.

    What would make this much better would be watching the user. Lots of players these days are sitting in front of a front-facing camera, either on their tablet or phone, or on their MacBook, etc. Facial expression recognition is a pretty well-developed field. With this you could get a pretty good idea of where the user is getting frustrated or otherwise upset, and where they're having fun. I haven't seen any games attempt this yet, but it seems inevitable to me.

    Even without going that far, though, it might be worth trying to code various "frustration detectors" based on the inputs you have. Button mashing, increased pressure on pressure-sensitive buttons, etc. might be indicators that you need to dial down the difficulty a bit. On the other hand, it could just mean the user's really getting into it. It's a tricky problem... but I agree with you that we need to try.
     
  3. RockoDyne

    RockoDyne

    Joined:
    Apr 10, 2014
    Posts:
    2,234
    First off, isn't this kind of the whole point of sandbox/open world/non-linear games? It's not actively tailoring the game to the player, but it's still providing a "custom experience" in a sense.

    On to another topic; hidden, dynamically balancing difficulty isn't unheard of, but it's not typically easy to spot though. The Ratchet and Clank games apparently did some of this based on how often you are dying, but it's effects weren't readily visible for the most part. Occasionally you will see games that are explicit about shifting difficulties and are more likely to mean it as a taunt.

    In general, I don't think this is something that is necessary for games. The point of this for commerce is to placate customers through your channel. You want to be one step ahead of the customer so that you can be the one to guide them to what they purchase. This model doesn't have much of a point in games though. You could hold the player's hand and try to lead them through their perfect experience, but why? There's no reason to care if the person has already bought the game. It makes some sense with freemium, although most of the conflict there is probably to keep people from realizing they are just being strung along.

    This is a part of the whole personal experience/shared experience debate which doesn't have any real right or wrong answer. There is as much merit to discussing a shared experience as there is to your own unique story.
     
  4. imaginaryhuman

    imaginaryhuman

    Joined:
    Mar 21, 2010
    Posts:
    5,834
    I don't think this is solved with a sandbox.. this does permit user freedom, but it doesn't anticipate what will make the user happy at all... they're left to passively stumble upon their own fun. The idea I'm aiming at is to `meet` the player with an experience that they are able to like and enjoy, not just leave them to their own devices.
     
    JoeStrout likes this.
  5. Kiwasi

    Kiwasi

    Joined:
    Dec 5, 2013
    Posts:
    16,860
    This is where I see future AI work going. Not so much in the agent space, but in the GameMaster space.
     
  6. RockoDyne

    RockoDyne

    Joined:
    Apr 10, 2014
    Posts:
    2,234
    And the issue with this is...? There's nothing wrong with leaving people to their own devices. If anything, it's preferable to trying to suss out the player's current state of mind. Try to tell the difference between someone lost and someone trying to explore everything, or between someone having fun with vehicular manslaughter to someone who sucks at driving.

    I'll reiterate the most important issue here. Why? To what ends does this actually serve? It is more important to deliver people an experience than it is to deliver their "perfect" experience, however misguided that would probably end up being.
     
  7. imaginaryhuman

    imaginaryhuman

    Joined:
    Mar 21, 2010
    Posts:
    5,834
    I don't know what your point is, either. This is the way things WILL go. Games WILL get more dynamic and sensitive to the user and they WILL continue to adapt to the player, so as to keep their interest. It's already happening. There is always room for the open games with less control etc, but so long as there is money involved there's going to be an increasing trend toward increased relevancy to the player and adjustment of the experience to elicit the most engagement.
     
  8. RockoDyne

    RockoDyne

    Joined:
    Apr 10, 2014
    Posts:
    2,234
    If people lose interest, it's probably because the game sucks and they have better things to do, or they have exhausted the content they wish to explore. If you make a crappy game, I highly doubt these systems will save it (never mind that the development time on these systems is probably the reason it sucks anyway).

    As for the latter issue, the only instance in which these systems help is to get people to content that they can't find. This does assume that the content is interesting to the player and that the player is uninterested in searching for content that they find interesting (which at least implies they don't like the game, otherwise they would be perfectly capable of finding it themselves). Using these systems for this reason means the content was produced in bulk, either procedurally generated or done by hand, so it's quality is likely to be sub-par, and you've then sacrificed that quality to keep people busy.


    At the end of the day, you can't be all things to all people. It's just not economically viable to try and adapt for the players when people are much more capable of adapting themselves to the game. If they can't adapt then they probably weren't in your target audience anyway.
     
  9. TheSniperFan

    TheSniperFan

    Joined:
    Jul 18, 2013
    Posts:
    712


    While I'd certainly love to see gaming go towards this (which would also be logical by the way), I don't see this trend.
    The games industry seems to "evolve" in the opposite direction by 'streamlining experiences' or 'appealing to a broader audience'.
    As a shooter fan, I know what lead the person to making this comparison. Even though I haven't played The Elder Scrolls, I haven't missed how long time fans point out how the newer installments are loosing complexitiy to be more 'mainstream-friendly'.


    But let me get back to your original post.​
    The way I see it, the best thing to do is to take subset of the gamers and really focus your development on them.
    Do not go down the website route, trying to create a website that works for everybody, because games aren't websites. Given that they're entertainment, you need to take into account how hugely different the tastes of gamers are.
    People were quick to point out sandboxes, but my point is: Not everyone likes sandboxes. When you try to alter your sandbox in a way that makes those people start liking it, you're running risk of losing the people who liked it before.


    Let's take a step back and look at it from a higher-level perspective.​
    In order for a game to predict the player, the developer must predict them fist in order to write the algorithms accordingly. For the developer to be able to predict the players, the developer has to really know his target audience. On top of that, the target audience must share very specific traits. Those traits they share are exactly what the developer has to be working with.
    The more diverse your target group is, the harder it is to find those traits, because they have to be specific. The very generic traits they share won't help you, as they will be covered by the basics of game design.

    The more focused your target group is, the easier it is for you to understand and predict them.
    "Gamers?" Nope
    "Core gamers?" Nope
    "Core gamers who like zombie games?" Nope
    "Core gamers who like horror games?" Slowly getting there...
    "Core gamers who like slow-paced horror games that rely on psychological warfare against the player, rather than jump-scares and action?" Bingo! That's something you can analyze and work with.


    Furthermore you need to take into account that there a huge problem you'll run into when creating 'adaptive games'.
    For a game to be truly adaptive, it needs to alter its content according to the player to offer the required amount of variety. Such a game relies on procedural generation.
    For a developer to effectively to use his predictions to adapt the game for the player, he needs a large amount of control.
    The huge problem I was talking about is that this is a contradiction.

    Procedural generation ALWAYS trades control for variety.
    The more your game can change, the less control the developer has over the effect. And while this may sound weird at first, think about this:
    A game in which every last thing is subject to change, becomes unpredictable for the developer.
    Less is more; Be subtle. If you pull it off correctly, the player will never know that something changed to specifically cater towards him.



    So what games have attempted or even succeeded at this?​
    Well, the first thing would be good horror games like Amnesia. The developers knew that the player will be walking down that specific corridor, facing that specific direction and open that specific door because of their curiosity. And they used - or rather abused - that knowledge to provide some impressively effective moments.
    But these aren't dynamic, merely scripted.

    The only truly dynamic system I know of happens to be one that does its job really, really good. I'm referring to the 'AI Director' Valve has developed for the Left 4 Dead series.
    In case you haven't played it, let me tell you a bit about the more advanced version in L4D2.
    Based on the behavior of the survivor team, the AI Director makes subtle changes to the amount, type and location of items, enemies, special infected and bosses they encounter and how they behave. There's also at least one occasion where the map changes randomly, but it's hardly worth noting.
    If the survivors do well, the AI Director throws more, stronger enemies at them and gives them less items. If they take a shorter route, the difficulty is raised too. Survivors who let their team behind are actively punished for this behavior by becoming high priority targets for the infected.
    The AI Directory uses an intelligent spawn system that relies on occlusion to determine what the players cannot see, to potentially spawn enemies just there. So instead of having hard-coded spawn points for enemies, they can spawn literally everywhere, given that nobody looks and the level-designers haven't specifically blocked the location.

    While this system only makes superficial changes to the experience and thus is far away from what you envision, it is incredibly effective. It is because, instead of focusing on making many, huge, obvious changes, it makes few, small, subtle but important changes and does them right.
    I think what Valve did with the AI Director is the way to go when it comes to adaptive games. Do little, but do it right.

    To me, the next logical step for a potential L4D3 would be to take the changing of the environment to the next level.

    If you're interested, here's a presentation of the AI Director as used in Left 4 Dead 1:
    The AI Systems of Left 4 Dead



    Just one more thing...

    While I see what you mean, that's a bad analogy. Mostly because your understanding of how compression algorithms work is wrong.
    Compression algorithms don't work by predicting the future. While the design of certain algorithms, like the lossy video streaming algorithms you mentioned, were designed with some predictions in mind, they do not predict the future themselves; They don't adapt.
     
  10. imaginaryhuman

    imaginaryhuman

    Joined:
    Mar 21, 2010
    Posts:
    5,834
    Actually with regards to compression I am well versed in this field and I do understand a lot about the subject. I know there are other factors involved but the `holy grail` if you will is the ability to predict 'what comes next' without having to store 'what comes next', which metaphorically speaking, is like predicting the future. The more the algorithm is aligned to producing the output that the data is supposed to contain, the better, which means the compressor has to adjust and change in the same way as the data, which is similar to how games need to adjust to match the user. ... but maybe its a bad analogy in that the data itself doesn't change whereas people do.

    I do see your point about not being able to please everyone all the time, and that's valid. But I guess my main point is that games in general a very rigid, trying present a given fixed story or a fixed set of levels or a premeditated experience which may or may not fit the user. Just as I see procedural content being used more and more I see games becoming smarter at detecting the player and adjusting themselves, even if that means adjusting based on certain `premeditated ways to adjust` by the developer.
     
  11. imaginaryhuman

    imaginaryhuman

    Joined:
    Mar 21, 2010
    Posts:
    5,834
    I guess this also might spiral off into the idea that game development is going to become more real-time, and less off-line? ... ie more decisions will be made at runtime and less at design time?
     
  12. JoeStrout

    JoeStrout

    Joined:
    Jan 14, 2011
    Posts:
    9,859
    That's a reasonable way to look at it.

    I think an analogy might be drawn with story-telling. Way back in the days of oral tradition, story-tellers would tell the same story many times as they traveled from place to place... but they wouldn't tell it quite the same way every time. They would react to their audience, tailoring it to how the audience was responding. They might skip lightly over some bits, dwell on others, use more or less flowery language, etc., to make sure their listeners stayed engaged.

    Then along came the printing press, and this became impractical; a story had to be carefully crafted exactly once to appeal to some "target audience" as well as possible, and everybody experienced it exactly the same way.

    Movies and TV shows are a modern form of story-telling, which apart from the medium, is not that different from printed books.

    Games are much like movies and TV... but different in this important way: not everyone has to experience the story the same way. Already this is (obviously) true, but it's often true in very crude ways: you experienced a story that followed an epic character arc all the way to a triumphant victory, and I experienced a tragedy where the hero died on level 3.

    What I think our imaginary friend is proposing is that games could adapt themselves to their audience (the player) in more subtle and interesting ways, emphasizing some aspects, de-emphasizing others, and generally making sure the audience is as fully engaged in the story as possible. Like him, I see this already happening (e.g. the AI Director), and think it obvious that games will continue to get better and better at this.
     
  13. TheSniperFan

    TheSniperFan

    Joined:
    Jul 18, 2013
    Posts:
    712
    Okay, I see where you're coming from now.
    But saying that it's the 'holy grail' doesn't tell us much, now does it? I mean, p=np essentially being the holy grail of computer science also doesn't change much about our current reality.
    Are there real-time compression algorithms? It's a genuine question, because I don't know of any. Even compression used for streaming works with buffers iirc. That's not dynamic/real-time/predicting at all.

    As for games being rigid: I fully agree.
    I also don't think that the obvious answer - going all sandbox - is the right one. Rather than that, Valve's approach in L4D seems the way to go in my opinion. Only make stuff dynamic that matters and then work from there on, improving and extending your algorithms.
     
  14. Kiwasi

    Kiwasi

    Joined:
    Dec 5, 2013
    Posts:
    16,860
    The other thing that could make this all possible is cloud gaming. Your game master AI does not have to adapt to just the players data. You can crunch huge amounts of data from all players, do some data crunching to sort out your various target groups. Then adjust the gameplay for each group.
     
  15. RockoDyne

    RockoDyne

    Joined:
    Apr 10, 2014
    Posts:
    2,234
    The thing about the AI director is that it serves a very specific purpose, to control the pacing of action. It's reason to be there is to provide a source of randomness and unpredictability in play, so the system is in place to find a balancing point for difficulty. It's a solid reason for investment, and it's more than capable of using easily accessible metrics to provide feedback on it's performance.

    As far as trying to extrapolate these systems to the story, there are a couple of issues that have to be addressed. First is where does the feedback come from? If feedback does not come internally, it doesn't exist. External factors like biometrics are useless to account for if it's only applicable to ten percent of the playerbase or ten percent of the time.
    Second, videogame stories are usually terrible, so how could this possibly improve that? I would be willing to bet it won't. Trying to actively manipulate a story requires far more mastery over storytelling than just telling a good story.

    The fact that most of this discussion seems to be about choosing options for the player instead of letting them have an actual say in the matter seems off kilter.