What is Game AI?

Alex and Phil over at AIGameDev.net defined Game AI as “computational behavior” to separate it from academic AI (which is traditionally modelling humans).  I don’t think it’s going far enough, because it uses the word “behavior”.

Behavior is a pretty loaded term.  It implies agents, anthropomorphized agents.  That implies characters, NPCs, and significant choices of interaction.

(You can probably guess why I might be concerned about that)

To put a finer point on it, how many games that you’ve played recently meet that criteria?  With the exception of strategy games and to a limited extent shooters, agents tend to be simple at best. I claim that there is still lots of AI in those other games, and I’m going to back that up by… changing Alex’s definition.

Game AI is an algorithm that replaces randomness.

In plain English:

randomness-replacing computation


Good Game AI is an algorithm that provides superior designed results to randomness.

This covers a lot more then just behavior.

Note that this is very different from academic AI.  It is also different from computational behavior, both broader and more specific, and yet produces similar results in similar contexts, particularly behavioral contexts.

It is also (interestingly) creates a clean separation from Gameplay Programming, something that has historically been lacking. We can infer from the Game AI definition that Gameplay Programming is the creation of game elements. Gameplay programming is therefore correctly a required first step before AI programming. Cool! When correlaries naturally fall out of definitions that match reality, that’s a sign to me that we’re on the right track.

I want to get down and gritty in this definition, particularly from the point of view of AI Directors. I’ll do that in the next post, later today. Gotta go get our fish back from George.


GDC 2010: AI Summit: An AI Assist for Interactive Storytelling

Phew! GDC 2010 has come and gone, and I finally have a chance to breath.  The AI Summit in particular went fantastic.  A big thanks to Dave Mark and all for helping organize it.

Here are the slides from my talk on Tuesday on how you can use existing AI architecture you already have in your game to improve things you haven’t thought of before.  The game taxonomy has been particularly helpful in really understanding tricky game design problems.

AI Summit: An AI Assist to Interactive Storytelling

If you have any questions about the talk, feel free to post them in the comments and I’ll do my best to answer them.

I was deeply happy with how our Interactive Storytelling session came out, and a number of people told me it was one of their conference highlights.  Another big thanks to my co-presenters Michael Mateas and Emily Short.

There’s several posts that have come out of the conference, and I hope to get to them at some point.  But things are quite busy now and so they may have to wait on more exciting news.  They say action is better then words anyways.

AI IDE 2009

I’ve been invited to give a talk at AI IDE 2009 , titled “Bringing Interactive Storytelling to Industry”.  I’ll be discussing my experiences at several companies now implementing interactive storytelling systems, and both the playtesting/research results as well as the practical discoveries I’ve had.  Lot’s of work to do still, some new data that I’m analyzing now.  But I’m looking forward to it, and quite excited to get a chance to share my experiences with story-driven single player blockbuster action games.  Hope to see you there.

AI: Thinking for the Future

So I’ve been reflecting on Jane McGonigal’s talk.  I’ve been a fan of her work for a long time and I think she made some great insightful points about game design.  Not so much futurist but visionary and optimist.  I can get behind that.  But she poses a particular challenge to the likes of, well, me.  There’s not any room for AI in her vision!  Sure, we’ll always have single-player story games, people enjoy it, robots are cool, yada yada yada.  But, to paraphrase Raph Koster, what if multiplayer games are the future?  Where be some of that computer arti-fish-ial intelligence in this?

Well, what do designers like Jane need?  Boldly putting myself in their shoes, they need need ways to fight the fight.  They’re trying to make a different kind of game – massively multiplayer, socially changing, reality-based, impact-oriented.  Old techniques just aren’t going to be that useful to them.  That means:

  • supporting users,
  • maintaining the experience,
  • much shorter development times,
  • communication tools,
  • managing large numbers of users with a small dev team,
  • user training and guidance,
  • and real world integration

Hmmm…. sounds like web development.  What else do they need?  What’s game-y about them?  Something like

  • players assuming characters and roles,
  • goal-driven,
  • feedback,
  • and assisted team formation.

Those aren’t traditional AI problems.  If anything, they are closest to gameplay problems.  I recall Steve Rabin’s slide from the GDC AI Summit this year where he talked about how much of game programming was moving into AI.  But none of this helps these designers.  That big circle of “Game design”?  It’s splitting in two.  Some of those designers are betting on us.  But the rest have been heading in the opposite direction.  Fast.  And coming off of GDC, it feels like there’s a lot of them.  Designers weren’t talking about how to make, well, characters.  They were talking about how to make people interact with each other and the game better.  If we want to get in on that party, and I think we can, we should start moving soon.

I can see where user matching, player feedback and training, maybe even sinulation fit on the path.   I volunteer experience management as a second step.  What else can we bring with us?

GDC 2009: Friday

One last time, the usual applies – these are at-the-time stream of consciousness notes, unedited.  Pure live-blogging with the added disadvantage of not actually being live.  I have to write quickly to keep up, probably missed everything, apologies to speakers I just didn’t understand.  It was a fantastic conference, and you’ve all been a part of it.  My personal additions are in ().  This is nowhere near as good as having actually been here or having the actual audio, I’m afraid.  But here’s something to remind us all with.

Play in the Age of Social Software

Frank Lutz and Mark Pincus

Missed Mark but got in to see Frank talk about some of his recent gams. He makes the point that games before computers were all social. He’s doing games that use social mechanics to overcome game design problems. Pac-man in NYC. Using social mechanics to encourage players to work together in non-zero sums way. Or players working together to find stickers for Sopranos game, gratitude even though they would get a small advantage of getting $100K by working alone – the sharing really surprised them. And a lot of the players would gather online while the Sopranos were being broadcast, turning TV into a social space. It’s a single player game, not designed to be a social game – that emerged. This took them by surprised. So strong they wanted to encourage and support this community even in single player game.

Next he did Parking Wars for Facebook. Parking cars in your friends spaces to make money. Designed to be at Facebook rhythm. It’s a persistent game of low intermittent activity. Yet kind of a game you’re always playing. Also had an interesting visualization of the social graph – everyone’s game board is unique and totally different. Seeing friends of friends, picks up on that cocktail party vibe. And it can be played very cooperatively, but punctuated by these gotcha moments. They thought they were designing a little game for a little TV shows. You have to play for a very long time to get to the upper tiers of the game. But people hit that ceiling. “Worst game ever played” because people played for 2 months straight. Accidentally designed a casual MMO. Little tiny version of Blizzard. So had to make content for these players. So designed badges, to also try out a new idea about social badges, like achievements. Hidden info on what was required to get badges, but once a friend got it you could see what it took. There was a really tough one called the Flirt, and people would try and trick people into thinking they’d gotten it, it was so hard. One of things that’s interesting is the ethics/morality issues that emerge. One of them is collusion – symbiotic alliances between players. Public and knowable in Parking Wars, like between husbands and wives. People called this cheating. He thought it was just a non-optimal form of play. But it was kind of a Prisoner’s Dilemma variation within the game. Another form of cheating was fake facebook accounts. This is obviously cheating. Player community weren’t as much aware of this. Player’s also discovered a bug – if you have an ‘ in your name, you are unticketable, so a lot of player’s just changed their name.

Backchannel was another game made for The Hills. Basically competitive form of chat. Test a new social form of gameplay. Comment on the show with funny thing. And as people click on comments the the comments earn points, and if you’re fast the other clicker earns points. Similar to Acrophobia, Apples to Apples. Lot of math behind it to incentivize people to think about what other people like. MTV wanted traffic to their site, and so they did a lot to link players at the end of the game and using data analysis to forge connections between similar commented people.

Kelly’s Bags was a game to promote Electrolytes. They wanted to create a game for a particular audience, 30-40 yr women, people not already on Facebook, give them a social network game. Impromptu social network related to these bags. You have a collection of bags, scavenger hunt component, and every day there’s another fictional event, and you score points by bringing a bag. One of the ways you score points is by bringing a popular bag – popularity points. But don’t want to wait to see what’s possible, you want early time bonus points. Trying to model fashion and trends and a game dynamic. You can also create friendships with other players and get a bonus for having same/different bags from all your friends. Looking at game theory and the analysis of decision making and trying to make a complex payoff, a dilemma again. Whose going to take the unpopular bag bullet? Right away players figured out how to optimize the system, surprisingly – to change keep changing your friends. Thousands of posts at 11:30 of people trying to find friends to match their bags. Became a hilarious parody of fashion – caring more about your handbag then your friends. Game ran for 6 weeks. Built a tiny little intense group of players by having a prominent forum. Had all the things in real communities – fake bars, guess a word game. Ethics and morality of game actions also became intense in these communities. New kinds of social dynamics. Reality TV shows and EVE Online are doing it. Reflecting moral and ethical dilemmas reflect real world ones.

Question: Community managers to keep people honest, and how do you build word of mouth? Mark does, Frank bootstrapping in, it’s time consuming. Porn spam is a full time job. Question: Used in companies or conferences? Relationship between training games and trust and corporate communications games. Serious games movement has done some work in this direction. Replacing some more traditional activities like sports. Question: Unwilling to give personal information? Facebook gave privacy control, was big deal for mainstream. Maybe not enough for everybody, but what makes it feel social is that you are real and you may see your friends and you have a reputation. The ethical questions make people behave better. Halo multiplayer is infamously hostile environment, and Facebook has solved this problem. Question: How important to do open platform, cross-platform, and what resources are out there for people to start out with? Don’t know yet. Spreading the Facebook plumbing is a revolutionary concept, that you have a persistent identity. Depends on your context, the size of your audience. Resources – great blogs, just google social gaming. Question: Game balance? Interesting balance between social fantasy and persistent world. Like a good cocktail party or tennis ladder, you only care about your friends and people around your level. So focus on making it not fun for griefers at your level. Depends on amount of persistence. Even if player’s aren’t playing optimally, player’s should feel they can play optimially, that the perception of you being able to compete if you start a week late is there.

Beyond Balancing: Using Five Elements of Failure Design to Enhance Player Experiences

Jesper Juul

People present designs after products, going up. Researchers start high and try and make it applicable. 2 studies of failure in games.

Failure is a situation presented by the game to the player and the player fails at the challenge and the game punishes the player for that. For a period of time we thought about balancing as tweaking the number of times a player fails at the game, but we should be tweaking the cost of failure to the player, particularly in turns of time.

Traditional view of Balancing – balance single-player games as an autonomous unit and place player in flow state, not too easy, not too hard. But this is harmful. There’s this myth of hardcore players like to fail, casual players don’t like to fail. Does the casual player dislike failure? Gamezebo survey – 183 responses, average age 42, 93% female – a game that is too easy is surveyed worse then a game that is too hard. Want to be challenged. Will work to overcome frustration. So in actuality, match failure design to the time constraints of players. And communicate failure appropriately. (How casual are these survey participants really – not clear how driven to website.) The Danger of failure – learned helplessness. Player’s ask – why did I fail? If you fail students the wrong way you can get them to never come back. Internal or external, global or specific, stable or unstable. It is only me, I am stupid, I will never solve this, as opposed to everybody has this problem, I’m bad at this one thing, I can learn to solve this.

Looks at a little game he tried and then surveyed. People enjoyed it most when they finished the game but lost some lives. Not completing or completing it without dying was markedly easier. So we should push people out of the flow channel a little bit. (I think this is misunderstanding the flow channel.  It’s supposed to push you.). Failure makes it obvious for players when they improve. “Too easy if not forcing me to use creative thinking or use my reflection.” “You never have to adapt your strategy.” Failure adds depth and content to your game.

Ways to tweak your failure design. 5 elements of failure design. Failure count – measure the number of times the player reaches the failure state. Failure Awareness – knowing that you might have lost makes it feel you good, knowing you could have failed. Try and make player aware of possibility of failure, even when unlikely. Benefits of failure without the cost. Failure Communication – failure can be communicated very differently – through game objects or directly. Objects are fairly neutral, direct is… stupid-feeling. Look at where the feedback is coming from. Consider Flywrench, where you fail an amazing amount of time, but only lose a very slight amount of time. Failure setbacks – players don’t have unlimited time, so snatches of time. Tweak the setback of each failure to match time constraints of players. Randomizing can make repetition interesting. It makes a level replay after failure more interesting .

Move from failure count to failure cost – failure count x failure communication x failure setback x failure repetition. High cost, like Mega Man vs. low cost like Peggle. Remember, the expanded audience is not averse to failure count, but to high failure cost, in emotion an time.

Question (Chaim): Research with children? Nope.

Arf! Arf arf arf: Talking to the player with barks

Patrick Redding

Barks refer to generic grunt noises from NPCs. He’s referring to AI Dialogues. Not conversational, scripted, not cinematics, not MP taunts. Barks.Why do we use barks? More immersive, verisimilitude, let’s AI express selves like real people. Makes the AI seem smarter, reveal intelligence, or mask stupidity. And reveal status of the game world to the player. Finally, it can support the game’s themes naturally.

So we take this numerical code set of systems and try and make them softer, more analog. Could be idle/ambiance fleshed out lives or letting player know their actions matter. Or interaction in combat or social through game systems. Or narrative content. Overhearing NPC conversations, for example. Good AI bark games – Halo, (5000+ lines of dialogue), Halo 2 is 15,000, Halo 3 is 35,000+. Fallout 3 had a robust conversational game, but needed to layer into that in combat, as well as attract player at long distance. GTA 4 is synonymous with this stuff, 80,000 lines of dialogue, 7,000 for Niko. Tons of voices for pedestrians in different languages. Babbling guys on sidewalk. DJs use AI systems to react to time of day, game pieces. Far Cry 2 had 4500 lines of dialogue, narrative content, and only 3000 lines of systematic combat and 1100 of buddy combat dialogue. 100 lines of zulu and afrikans lines were incredibly useful. F.E.A.R. put affordances in the world to make tactical decisions about the world and their dialogue tried to expose that. Fable 2 had 160,000 lines of dialogue to back up their world. Assassin’s Creed had 12,000 for crowds and Altair reactions to react to.

Considerations – systemic vs. Scripted? Primary function of AI in support of gameplay? Does AI have a life when player not around? How dynamic is combat and other actions? How open is the level design, how flexibility?

To consider scripting. A trigger hits script, defines event, and it goes through. Systemic has a source and a stimulus, that drives the event. The source states provide the frameworks for the barks. The systemic side of it has different inputs from different parts of the AI, is it larger oversight AI or on the character? Depends on whether it’s a deep or broad system. If it’s sensory, reciever looks at states and the source has game data states as well. These states are where the data is. If it’s dispatcher, the receivers are the same, but the source is a big babysitter and identifies global conditions become true. Maybe player is almost dead, maybe tension is high, maybe AI is told to use certain tactics. Tells receivers to act as if something was true.

Another thing we have is contextual/affordances. Used by Sims and Far Cry 2. In FC2, used for interesting idle behaviors. These needs drive their tasks. Interestingly, other Ais can be affordances which can generate conversations.

The scripted side is more mission driven. It pulls from gated story dialogue. Somewhat interactive story-esque. Could be quest or location pulled, but pre-set up.

The bark strategies. Generic vocalizations – terrible to write, direct, and perform. Zero subtext context that we hear over and over again. They add all this redundant data to the game. On FC2 they set aside a separate database for these, and had a lookup table – writer could just look up the tag without having to write it into the script. At runtime the dialogue manager detects the writer’s tag in the script and chooses one. Much easier to record because actors can improvise, and gets more different versions.

Can use heat maps to track where player is inputting damage, and use it as a game data state that the AI can track and react to. We can use this to react more appropriately, avoid bad stupid reactions through bark filtering. Also have ability to take RPG statistics to tweak reactions as well, to gate content. Also believes in different reactions from others, not just railing at the player. Chain dialogues. Before one would start a bark, check in radius around, and chain them through the Ais. If you have solo barks, you can also use different languages to cover for it not being chained or even it being repetitive – can get more from fewer.

Layers all these types of barks together to get best results.

Content management pipeline – spreadsheets are an ugly compromise due to so much content. Have dedicated tool, and you don’t want the designer doing it. Can also add meta-data to inform other systems of the game – animation, audio, at run-time, for things like gestures, facial animations, building an AI performance manager. What kind of prop is used? But hard-codes rules into other systems, which gets a bit messy. Also, use variation banks that can be swapped in and out through streams, and maybe use global AI conditions like alert state to swap out banks and get that variety. Use generic vocalizations to fill in for that. The Censor system regulates the sheer amplitude of barks, but also lets designers tune the mix, and avoid repetition. And finally in playtesting they had little bots that tracked frequency of playback of each bark and mined the database. Maybe some barks weren’t played – bug – or the player is getting spammed – get more variety, add new states. Really useful tool.

Question (Adam): Facade conversation mixing? Hoped to do mission briefings in barks, that dynamic. Had big dynamic goals. Could probably strip out redundant systems using this model including player and mission dialogue. Certainly RPGs, thinks it’s a smart direction to go. Question: edge cases of trouble without enough variety, growth of variety? In FC2 didn’t have much of this, but could be a problem with it. Is doable, some succession of reactions for social mechanics. Question: Testing? Only highly successful system because done in early prototyping. Went through 3 system re-writes, iterations. Started with sensory and added more and more director stuff in the list. Depend on playtesting. No easy solution.

Vast Narratives and Open Worlds, Part Deux — Big Huge Problems

Ken Ralston and Mark Nelson

This talk is about blunder management. Going over there vision for their MMO and struggles. RPGs have more world interactions, so that requires special design. Scope is really difficult to manage, because expected to go large. Lots of different authors are difficult to manage – you can hammer them down to a single voice, or get all the different colors – but want both, so learn how to nuance it, you have to do 400% of everything and then through away things that don’t work. Moving targets and shifting focus – they triumphed by selling it to their publisher, but it’s a reality of making life crazy. Your vision needs to be able to adapt. Bethesda was brute force, more like jazz – so many notes, some may be bad, but there’s something for everyone. And fixing this up takes time and causes tragedy.

The classic blunders – scope and trying to do to much. We out think, outsmart ourselves and put ourselves in really risky points – at either complacent, huge ego or overwhelmed state. And then when you are on those extremes you do things that later you think weren’t that smart. Ken thinks in their IP the problem was they tried to do a mid-stream narrative IP revision when they should have stuck with it. There was a Plan A and Plan B, and when they fell back to Plan B, they didn’t work it through. Ken had these visionary world ideas, that’s his thing, but it wasn’t going to be fun – would have sucked. Mark thinks it was trying to do too much – RPG, strategy, big battle sim – all in one game. What they tried to do to fix it go back to the vision statement. Focus on an RPG, not a weird hybrid. Work on what they new, not be so far outside their zone. At the very outside edges of their blunder readiness. They were distracted from iterating on the vision. And didn’t want to lose company communication time on the vision rethink. Eats up all your leads.

Second blunder – the Emperor’s New Clothes. Ken thinks it was his silent complicity with the replacement IP vision. Ken wanted to give support, inspire people, keep their effective domain work. He didn’t savage Ken enough. Mark finds everything really fun, decided to redo the game and didn’t tell Ken. But Mark thinks he needed to tell Ken that big battles would never work – he should have said it up front. Oblivion‘s big battle was 8 guys with no AI turned on – they knew better. Ken thinks it was because the currency of victory is only testable in an implementation. But you can’t win that argument if it’s not on screen with the guy whose focused on something (the kinds of guys you need). They fixed it by re-opening the lines of communication. Take a step back and tell each other what they are doing wrong. Ken’s mantra is make decisions are always good – right or wrong. He was banging back between complacent and overwhelmed, and blunder ready. If you see someone in this, help them out and get them help. Also, better document the vision and communicate it – team synchs, blogging, small Powerpoint presentations of condensed knowledge. Blogging was a great tool, just a couple lines at the end of the day, but it kept everyone updated through their selected RSS feeds. When people are coasting and going to the internet, it gives people a casual place to watch the soap opera of development. To see the emotional color. Find new ways at work, new trick, to get people to communicate. Stay on message. Be conservative early – you cannot just rewrite the IP on the fly during production. Lot of detail work on the IP, and it’s tragically difficult to change the high level stuff, particularly when you hide it in the original themes. They had a core objectives bullet lists, 10 of them, but they let it get away from it. Forgot about it. Use it as reference for their decisions. But they had to diverge from their list. It’s really hard to rework your top 10 list. Plan for the future and scout the terrain. Mediate the tragedies for your scopes. Don’t always be let people get away with not delivering and be supportive.

The team – it was difficult because of the need for experience. Transitioning from RTS to RPG. Want people in genre who can say their experience with authority. And get an important need hire every 3-6 months for someone who sets the team on fire. Staying hot for 3 years is the big challenge – managing your own enthusiasm. So look for hires who can help. And be aware vast scope gets people isolated and makes people feel separated. If don’t know what art or features are coming, they won’t use them.

Classic blunder 3 – Counting chickens pre-hatched. Mark thinks it was not playing to the strengths and experience of their designers. Weren’t paying attention to who their designers were and what they were good at, and play to their successes. Once they did, jacked up the quality of the game. Somehow lost good people because didn’t make him feel important enough and didn’t listen to him enough. And putting good people on another team because they were complacent about what they had, when they were really in risk of failure. Ken doesn’t think he knows anything unless he can explain it to a bright 10 year old. He wasn’t energetic and brutal and autocratic enough to the innocent.

They fixed this by giving out domain ownership to people, small areas to people to own. They owned it 100% – made people’s become experts quickly. Chosen by who was asking questions in the meeting. And it created less work for them. Gave them more time to look at the game as a whole. Distributing burden, sometimes to people who were better at it. Also, Mentoring, there’s no universal theory. Every individual needs something different – some every day, some only once every few weeks. And cross-discipline seating. It’s not comfortable, people want offices. Bioware has folding tables and moves people around on the fly. It sounds horrible, but it was so success for them. More review, shorter cycles. Getting into young people’s stuff early, eyes on it, eyes on it, eyes on it allowed them to grow quickly and learn what they were looking for. Go 2 days instead of 2 weeks.

So how do you strategize hiring? You need to hire just in time. Maybe that means contract work for a very short time. Not just which people to hire, but when they’re hired. Look for the TRP – the tragically responsible person. Giving people ownership makes them feel good about what they’re doing. And manage the enthusiasm burden – know where the individuals are and who the individual leaders are. Mentoring, promoting learning readiness. And use blogs and wikis. Gives you revision history and quality control. But not Sharepoint. And knew better – unwilling to spend the time to stop producing and make things actually work.

The Narrative – why it’s difficult. The scope – content is king, have too have too much because it’s the bar. Be vast, modular, and extensible. When writing, stop making art and stop making fun. To the moment to moment gameplay of the player. And dialogue – nothing is more poorly simulated. Ken hates it, but knows it’s essential. They got a big win by getting a “least harm” dialogue system.

Their Classic Blunder – started writing narrative way too early, before systems were nailed down. And write your narrative to bring your player to the fun, which means waiting for that to be ready. The Cart Before the Horse. Ken thinks it was late and poorly integrated faction design, as well as him using celebrity before it was actually ready. His big focuses in writing are Faction, Setting, and Theme. And faction has a little chunk of story in it because it gets all the associations and relationships with the factions. And he forgot that.

To fix it, wrote and rewrote. The quests were overcomplicated and boring. Designers were trying to tell stories rather then writing gameplay. Simpler is better. The Lord of the Rings is a FedEx quest. The Quest of the Holy Grail. These things are fine. It’s the moment to moment stuff that makes it fun. Oh, and they put factions back into the game. They were trying to do a “blue ocean” RPG to do something new, a main narrative. So they stripped it out, when they were good at them. And they brought back “Tuxedo content”, simple but elegant content.

The Classic Blunder: Less is More. Ken thinks it was building an on a Narrative designed for a Defined Protagonist. You lose lots of protagonists to meet and the customization character. Mark thinks it was trying to make “RPG Lite”. RPG players like all those stats and numbers, it makes the game fun for them. So they remembered what mad games fun, played to their strengths, and iterated to see where they were right. They sprinkled loose threads and narrative hooks. Objects in the environment, everything you can do. Explore fast-paced dialogue menu system vs. the lingering conversation experience. And concentrate on core RPG Experience. Designed the systems to support the narrative. Design the narrative to focus on the gameplay.

To make all this easier. Design 400%, and cut 350%. They don’t do math, but bring it back down to a manageable amount. Repetition, clarity, focus. Communicate vision consistently and often – posters, shirts, anything. And make decisions. Any decision is better then no decisions. The team – give people domain ownership where they can succeed, and stay on top of them. Mix the seats up, and use effective cross-team documentation. And for the narrative, figure out what you’re making first. And remember you can change the story a lot easier then you can change the gameplay. Finally, seed the world with setting, theme, faction with loose hooks and develop into those.

Question: How did prototype feed in? In a bad way, to try to sell it to publishers. They were making it up, not using their tools, it was false fun. It was not genuine system fun, producable fun. Ken Ralston can wave hands and make a pitch and sell bad stuff. Some of the good early experiments were good. Question: Tech? Because we’re not tech people, when they failed we couldn’t tell whether it was really hard or whether they needed help. They needed an experience tech guy. Question: How many new people every 6 months? Need to be more sophisticated about what phases of production they are in. Bioware really knows how to do this stuff, have multiple teams going on, really helps. They were too arrogant about how far they were trying to go. Should have tried to talk to them. It’s a goal – just want someone still stupid. Question: Meaningful choice? Everyone says they want it. But when you get it, someone says they didn’t get to do some other thing. Morrowind had meaningful factions, Oblivion did it the other way, People complained about both. Ken: But main quests need to become less important. The most important story is “I’m not dead yet”. Ending the main quest is like being dead. Give your world a narrative structure, but less there and more finesse, the users focus on the other thing. Main quests suck. They just serve to lead the player through the game. Question: Factions first? Yes, narrative later, yes sometimes both ways, but structure around faction decisions. Ken’s rule is 3 different kinds of factions. (Rock-Paper-Scissors. Breaks the symmetry, 4 stabilizes too much.) Question: main quests? Still important.

Lionhead Experiments Revealed

Peter Molyneux

Why Experiment? Lionhead tries to innovate, innovation is inherently risky. It was a drunk night. Innovation helps with the success of a game because its what consumers want. Back in the Black&White days of 30 person teams innovation was easy and cheap. Lots of people at Lionhead have great ideas, a forum was needed to explore these. Peter Molyneux comes in with lots of dumb and stupid ideas, and he needed an outlet for his chaos. A lot of the good things come out of this group – one button combat in Fable 2, dog in Fable 2. Yes it’s costly, but it doesn’t really matter, because consumers really want that innovation. Be innovation without that huge risky. You can do it at 30 people easily, but when you’ve got 100 people, the last thing you want is a designer tromping in with some new cool idea. And those ideas everyone else has, they can’t express it in “the game” otherwise. So they’ll leave your company, unless you give them that forum to explore and innovate.

Plus, experiments avoid the nasty habit of innovating late. They only added a number of ideas because they experimented with those ideas first. The bread crumbs, the ambient orbs, all of it. It does not cost the earth and you can do more and do it quickly. And it doesn’t upset your production.

Anyone at Lionhead can start up an experiment team – a short term projects that have 1-5 people on them, short term projects that tend to take place at the end of a project when team members are free. Usually lasts 1 week – 12 weeks with the average being 4 weeks. The concepts can be proposed by anyone, they just need a sponser from a senior member of staff. The sponsser sells the idea to the Lionhead creative board, and it gets scheduled like any other project. About 6 people, most senior about 1-2 junior. Meet once every few months. There’s some structure to it. There’s milestones and checkpoints in this mini-production, and have a producer. Milestones can be a day or a week or in an extreme as a month. Coming back to the board, they ask why this experiment, what does it add, what does it cost, maybe even what do you see? Also, how much will it cost to make, how often will it be seen in the game, and if applicable what do you say to teach it to the player?

They have tools specifically for this – a very simple prototyping engine which strips away all unnecessary code and allows laymen like scripters to play around. Graphically the prototypes don’t look amazing but get to the core of the idea. They use existing game engines as well to experiment in, where it’s most appropriate. Also creating “concrete” internally – a way to share assets like graphics, animation, even a piece of code – tags to allow it to talk to the system and understand what the asset is and move it from game to game.

Afterward, they evaluate. They create a risk assessment, they work with the team until the tech is implemented, designers are briefed so it is fully used, and get any available patents. It’s very important to let design know, otherwise it will never be seen.

Question: Only Lionhead assets? No, anything you can. Avoid the excuse that you aren’t using something because of X, Y, or Z. (Second time he’s said that.) Question: Yes, between games, to use the fallow time. Downloadable content is eating into this, yes. But these experiments are so valuable they are so worth it. Question: Money? It is essentially attributed against a group. There’s a Shared Tech group, and if it’s for a game then it goes to the game, and if it’s central technology, then it goes to Shared Tech, which has a separate budget. But it’s rare to not be able to justify it financially, since it’s such a small amount of money.

OK, on to his examples. Protodog – dog for Fable 2. Tried punishing, rewarding, a bunch of animations and looks. Realized they wanted an experience of bonding, and so they dropped punishing or rewarding, and that saved them from waiting a year to figure that out. But they tried again, putting the dog in front of the play, and found the dog was leading you rather then you leading him around, and really cemented him in the game. Also departmental research – 1000s of creatures on screen using GPU, raytracing, water and fluid of mechanics, deformable skin, soft shadows. Most successful experiment was one-button combat to get combat available to players, picking different animations based on distance, positions, and objects/walls around the environment.

Shows an experiment that didn’t quite work. The Room – which he’s shown before as GDC – with the malleable clay and portals. It had great character, a lot of detail, but didn’t seem to go any further.

Question: hit rate? Around 70%, quite a lot of experiments fail at the start.

Player’s Expression: Far Cry 2 Level Design: Structure & Beyond

Johnathan Morin

Are we respecting our player’s enough? Not talking about respecting everyone, talking about the first time someone tries something. And then those who explore it deeper, and then have a quest for excellence and have an urge to create. That through the invention players can get greater at it then the creator can. Cultivating character – use emotion as strategy. “We are what we repeatedly do.” – Aristotle. Goes into a story of how his kid was exploring a toy, discovered something, interpreted it, and it influenced him, which changed exploration. We think we know shit and we base our life around it. Playing Street Fighter 2, there’s emotional buildup. The kind of guy you are is important. Anyone can become angry, but at the right time and in the right way, that’s difficult. The player is trying to exhibit self-control, direct himself. If it’s that hard to control players, how can we control them?

In Far Cry 2, they tried to do it with World navigation, Cause & Effect Loop, and Micro Decisions build-up, and then opportunities to breathe. In world navigation, they had quest choice, then the condition of the player, and the risk and uncertainty, all of that affects the player’s decision. Based on his character. But you’re going to screw it up sometime. We always screw up our plan. And so we get angry, for getting pushed out. So what do you do when things don’t go according to plan?

As designers, we need to think about there character to understand how people react. In Far Cry 2, there are Strategists, Rambos, and Fugitives (avoid confrontation). They filled this with Planning & Approach gameplay, Combat gameplay, and Escape & Evasion gameplay. Stealth approach, scouting, planning, AI Alert = Combat for strategists. AI Alert = Combat, Mercs Dead = Win, Player Dead = Player Win, or Player Flee = E&E, for Rambos. AI Own Player, player flee AI, AI Follows, Vehicle/Foot chase for the fugitive. Different mechanics for very different kind of players. These 3 players form a loop where each type overrides the other type. The pacing is the player. The designer doesn’t have control.

Micro decisions build-ups – people are making decisions based on their characters, how do we do emotional buildup? Hmmm… density. Layering density, in this case cover, to build up pacing. Looking at the 3 different modes, Understand how they react to the different densities, and map those onto your level designs. Look at the flow of gameplay, where each mode pushes the player. And it can get really complicated, so we’ll need to think a bit more about it.

The players also need opportunity to breathe, which is why there are diamonds in Far Cry 2.

Questions (Chris Hecker): analogue between guitar makers and developers as opposed to composers? Question: Tempo inside your game? Didn’t have time. Tried to control it, but in the end shouldn’t try, should be inherent in the place. Does say he wants to do it in a systemic way though. Question: Expected player to feel anger? Hard to please everyone. But also, when you get pissed off, you actually like it, when you succeed. He wants player to have emotions and reactions to the game, to be destabilized. It’s a thin line. Not a perfect job on Far Cry 2 – but the ones who liked it really liked it.

GDC 2009: Thursday

Again, the usual applies – these are at-the-time stream of consciousness notes, unedited.  Pure live-blogging with the added disadvantage of not actually being live.  I have to write quickly to keep up, probably missed everything, apologies to speakers I just didn’t understand.  It was a fantastic conference, and you’ve all been a part of it.  My personal additions are in ().  This is nowhere near as good as having actually been here or having the actual audio, I’m afraid.  But here’s something to remind us all with.

(A great quote from casual conversation:  Jonathan Blow – “Process is what everyone talks about in post-mortems, but the learning and creativity happens outside of it. Minimizing process is the goal.”

What does that mean for us? Can we minimize process away and have managers serve to cover the edges?)

GDC Microtalks – One Hour, Ten Speakers, Unlimited Ideas

Clint Hocking, Eric Zimmerman, Frank Lantz, Jane McGonigal, Jenova Chen, John, N’Gai Croai, Richard Lemarchand, Robin Hunicke, Tracy Fullerton

Robin Hunicke on @Home and the 4 Cs – Creativity, Collection, Community, and Competition. Climbing, Jamming, Where’s Waldo, Graffitti.  Was booting up laptop, couldn’t get most of it down.

Eric Zimmerman has a game called colors, depending on adjacent colors in the seats. He calls out how the index cards immediately got meaning, transformed the room with what would normally be trash, made a messy system with a new social system, speaking to the power of play.

Clint Hocking on Five Stars and the Truth – discussing Kent Hudson’s deconstructing lockipicking systems. It was a good talk last year, and he judged them on 5 star ranking system. Why 5 stars instead of 100? Well, 100 is too fine a granularity, yet we could get that level of specifity. Compare to reviews – where it’s even more abstract then design. Goes to the Wine newsletters, and how there’s formed the Cult of 90+. “90 is not just one more than 89.” Actually 90 is just one more than 89. Attributing mystically significant system. 100 points should remove dependence on thresholds. The wine industry has incentivized critics to give higher review scores. Inflation. If you look at colleges, colleges have to keep raising thresholds for admissions, teachers inflate grades, colleges inflate thresholds, etc. Except it comprimises the whole system. So we can’t compare games across years. And we as designers need to be able to do that. Are we getting inflation? Over 85% of the top 41 games have been released since 2001. College pressure grades, students pressure teachers, teachers relent, and inflate grades. Market puts pressure on developer, pressure on reviewers, reviewers make comprimises, which leads to inflation. The problem is the valve. Review are too analog, the pressure release is too granular – but it just slowly increases over time. Which would game reviews a great tool for everyone.

Jenova Chen. Evolution of Media – technology and user experience are evolving at different rates, and we’ve reached a threshold where user experience has exceeded technology. Our fun right now is too limited. It’s natural for a new medium – think early cinema. But as audience gets older people want more variety. The Video game audience is growing every year, and there’s more desire every year. It’s driving all our new genres. He sees games stimulating everyone. How? Socially, Intellectually, Emotional. You can map all our entertainment mediums to somewhere on this pie chart. But what is in the middle? First he thought of Disneyland. Games are good at Intellectual and Emotional, but we haven’t got Social. We kep trying to ducktape it in. Social is so subtly, and we are bad at subtle. We use 10 year old technology, IRC and VOIP. Because gameplay is very primal, all we are expressing is very primal. The gameplay itsself has to push social ideas. Think reality shows and their communities, or Disneyland. (Reminds me of an idea to push people to talk and build communities, and try and push the game in to that, rather then the other way around).

Frank Lantz – Games are not media. How do games mean? Media delivers information to our hungry brains. Why does it matter if games are media? Assumptions: Games are brand new – media is new. Not hundreds of years old. Oh wait. Games are not brand new – adding a computer doesn’t make games work. Assumption: Games go in computers, oh wait. Games Don’t go into computers. Computers will ultimately go away, we are really just interested in computation. Computers will go into our games. Assumption: Games are Content. Oh wait, games are not content, that’s just market pressures. There are more opportunities out there, games as community, club, service. Assumption: The message model of meaning. Games make messages that translate to the player. Games are somehow a statement from sender to receiver. But games are not a message, players are not an audience. Games are like meaning machines or networks. Players and designers are participants, and systems are participants. Suggests a brand new way to think about communication and knowledge in general. Stop trying to fit games into existing categories. Rethink our categories in light of games. Media is an artifact of old historical assumptions. Stop and think about how games are not media. Games can lead us out of media, and to a place of networks built around conversation.

Jane McGonigal: My Idea of Fun. Remember CZADOF -when she says fun, think C – Confuscious had a theory about Jen – what seperates us from zombies. Little acts of kindness. There is a concept of Jen ratios that helps measure the health of the space. Burning Man has an extrordinarily high ratio, youtube comments are awful, and it relates to how you feel. The higher the happier you will be. So L4D has the highest Jen ratio on Earth, which makes it the Happiest Place on Earth. And it’s the Zombie Apocalypse! Some cool science to support this. Eric Weiner and Bhutan, no room for assholes. Left 4 Dead is the Bhutan of the videogame planet. Fun Checklist – mathematically optimized, lots of fear, ancient chinese philosophy. Humilation = Happiness. When we’re embarrassed we have a physiological response to happiness. So Dance-offs – shows her dance-off Top Secret Dance-Offs game and the level-ups, stats, quests, as in the crosswalk quest where people break into dance outside. Which creates happiness for both sides. Use masks to hide identity, embarrass themselves, and make others happy, which makes them happy

Keynote: Solid Game Design: Making the ‘Impossible’ Possible

Hideo Kojima

Making the impossible possible. Focusing on avatars as players. Too easy to steal from others, to use others to define what’s possible, but instead need to think laterally and come up with your own solution.ereatively. Avoid stereotypes, think around problems. Don’t always go through, don’t think you have to go over, go around or go under or do something creative. So what barriers of impossibility do we encounter when we develop games? In games, the floor that pushes you up to the top of the wall rises with technology – you can sometimes get over walls with technology. Hardware and software. But game design and game creation is the ladder that can get you to the top. Goes into the history of his games and Metal Gear. Designing a combat game in 1986. 2D, more then 4 enemies, a player, bullets. And remember the hardware – could only draw 8 sprites at a time on the background. So creating combat games for this hardware – totally impossible. So tried to work around it. What about no fighting? What about just running around? Or a combat game about hiding? But not heroic enough. So he turned it into an infiltration game. Where when spotted lots of enemies can come to train you. In sequels added wider vision and a larger range of dangers outside the screen with PiP radar. Noted that he hit some boundaries he couldn’t cross – 3D for MSX2. But when Playstation came out he could do 3D, hardware helped him get over the wall.

From Counter-Strike to Left 4 Dead: Creating Replayable Cooperative Experiences

Michael Booth

Describes goals for why make L4D – co-op, blend single player/multiplayer Valve experience. Required cooperation, core feature, wanted crisp focus and to mitigate risk. Also replayability, encourage community and long term development.

How did they encourage cooperation? Structure game so players want to do the right thing – require cooperation. From the ground up, ensure the only way the players can win is together, not at the expense as their friends. They treated the whole survivor team as “the player” They also penalized non-cooperative behavior harshly and simple – abandoning the team = death. The trick was to avoid player rebellion,, natural design solutions in the environment. So no invisible leashes, teleporting, that sot of thing.

Survival Horror genre was the perfect fit, because well established and everyone knew “The Rules” – it matched their cooperation goals. For Enemies, they pushed on cooperation. Right at the offset there are thousands of zombies. No way a player can survive on their own – clear and obvious. Totally outnumbered, lays frame work for implicit cooperation without the leash. Also, infected stop you when you get hit, so you can’t just duck and weave away. In addition, there are special tougher infected for spice on this baseline. Oh – and playtest, playtest, playtest. Started as world full of grey boxes and got pretty fun, and they knew it was good. But the spice – each has a special ability that incapacity’s or overwhelms at least one character that forces interesting team choices.

First, The Hunter – nimble and can get anywhere. Exists to outrun and kill stragglers and “lone wolf” players. Completely incapacitating pounce attack. The Smoker pulls apart tightly coordinated teams – pulling survivors out of position. Survivor teams that were too disciplined would damp down the drama, and the Smoker breaks that out. Force fun moment players were denying themselves through being too good (hmm… interesting depth here). The Boomer – whose purpose was to break the rule of shoot everything that moves, to break open the mechanicalness. Hesitate on your shots. And of course the vomit explosion creates excellent moments of Dramatic Tension.

Incapacitating attacks, which prevent a player’s input, make players fear being separated, which builds team cohesion. If you aren’t the victim, then you also get to be the hero. Player’s really like to help each other – huge driving force for the survivor team, which brings in lots of different kinds of players.

The Boss Infected – specifically the Tank. It forces the OH SNAP moments, to change tactics and react. Breaks up the pacing and forces the players to immediately talk to each other. Puts survivors in defensive position, requires full attention of entire team. Also forces reconsideration and reevaluation of their environment. Tanks can throw cars, for example. Oh, and more Dramatic Anticipation.

The witch – breaks the rule of shoot everything that moves, but with more contrast then the Boomer. If you make a mistake someone is probably going to die. Just leave her alone. She gives you a lot of warning – crying, music. She changes up the pacing again, becomes stealth game – turn off flashlight, watch every bullet.

Furthermore, they did quite a bit with vocalizations. For example, how much stuff Francis hates. This built community unexpectedly, people exploring game. But original intention was improves situational awareness – AI for every player that communicates the situation. It’s automatic, builds team communication in an obvious way. It also communicates short term goals – letting players know what they should do next. And lastly it encourages a baseline of camaraderie – sets a foundation of civility. Discourages jerks in a subtle way.

They also limited critical resources, which they thought was counter-intuitive. First aid kits are rare, and using them is critical for success, but players have to work together to use them wisely. You could be a jerk and just grab them. Because of the structure of the game though, people share their resources readily, because of the other mechanics which encourage cooperation – everyone wants an extra gun. But when someone else heals you, it helps break the ice as well, makes everyone feel like friends.

Helplessness also demands cooperation. It’s clear to the team that the other player needs help, and is effective for a lot of the same reasons health kits are.

Moving on to Replayability. How do we do that to keep it fresh? One was Dramatic Anticipation. If event X implies event Y after a short delay, adds anticipation of reward or punishment is very strong. For example, the old Boomer vs. the new Boomer. Originally he was a bomb, but this was really a problem for new players. But exploding as a zombie attractor was very different – it avoided the bomb problem, but it also created anticipation for all the players. Lots of examples everywhere – breaking through doors, tank incoming, witch nearby, interactive music anticipations, finales/crescendo events, finale escape vehicle, incoming Mobs, Ledge hanging save, car alarms, incapacitation, “Third strike”/last leg. Car alarms are like double anticipation because there’s first you see the car and then you set it off.

What about their Structured Unpredictability – collection of interesting possibilities selected at runtume using intentionally designed randomized constraints. (Interactive Storytelling!) Low probability + High drama = Memorable. Designers often want players to see everything – in L4D they deliberately avoid that. By being rare, when they juxtapose or sit on top of each other they create the crazy stories. Even the worst possible game is pretty compelling, this greatlly enhances replayability and drama, creates memorable stories.

Part of this is Adaptive Dramatic Pacing. Algorithmically adjusting game pacing on the fly to maximize “drama”. It was inspired by observations from Counter-Strike – the natural pace of CS is spiky, big unpredictable fighting with periods of quiet tension. Peaks and valleys of drama, and constant combat is fatiguing, and long periods of inactivity are boring. Key thing is peaks and valleys need to be unpredictable. Same scenario can be used, but as long as different/compelling experience each round it doesn’t matter. de_dust has been played millions of times in CS.

The AI Director creates these peaks and valleys of intensities. Estimate’s the emotional intensity of each survivor. Track max value, if it’s too high, remove threats. If too low, generate more. How do you estimate the intensity? Use a value. Easiest was that, if damage is taken, raise proportionally to damage taken, when incapacitated, when zombie killed, etc. Decay over time but not if active infected in the area. Use Survivor intensity to modulate the infected population – build up to peak survivor intensity threshold. Sustain the peak for 3-5 seconds at peak, and then go to Peak Fade where you decay out of peak ranges while they mop up what they are fighting. Then settle into relax mode with a minimal threat population for 30-45 seconds, or until survivors have traveled far enough towards the next safe room. So Build up is full population, Relax is minimal threat population – no mobs, no wanderers, no special infected. Boss Encounters are not affected by this – they tried doing it, but the worst case games were always bad because there would never be a boss battle. The best thing about the system is that because it’s influenced by the player’s action, each game is totally different.

He shows a graph of the procedurally generated population chart. Sort of a debug mode for the AI Director. Then he shows the tracked survivor intensity. It’s using a pretty stark too high intensity threshold. In Relax mode wanderers appear again after the intensity has decayed all the way, but they are still in the relaxed time. Some nice structural randomness. Shows what it would have been like if the AI Director hadn’t happened, and you can see the biggest change is how the Relax mode really limits the population.

How do they fill the environment wit interesting threats? Layers of structured unpredictability. And only populate around the survivors. Akin to Perlin Noise, having independent layers that stack on top of each other creates an interesting, robust population that can be modulated effectively by the AI Director. They use the Nav Mesh to populate – reason about the space your in and put attributes in the area. Has an area been seen by an actor? Is area X visible from area Y? Have notion of flow distance, 0 at start safe room and maximum at end distance. This defines the direction the level is going in. There’s an active area set of nav mesh that follows the survivor. As it cleans up, they delete things and repopulate in front of survivors. Populating is done at randomized intervals (90-180 seconds on Normal diff). Usually create Mobs behind the survivors, to create a pinch behind wanderers threats. Special infected are populated at randomized intervals except during Relax areas, and they can spawn in areas you’ve already cleared. Boss population created every N units along escape path at random amount, shuffled with Tank, Witch, and Blank. Only can’t have same card next to each other. Avoid manually placed script/triggers – they started with that, they had randomized sets of scripts, but player’s figured it out, and player’s learned all scripts and locations and player’s lost all suspense. And then it kills cooperation because player’s expect everyone to have memorized all encounters, and it becomes a race.

Slides on Weapon placements as well, skipped. He mentions other supporting technologies – VOIP for communication, Game Instructor for training help to encourage cooperation, In-game voting to deal with jerks, split screen helped train buddies, achievements that encourage co-op actions and replayability, Stream/Matchmaking for groups that want to play together, and Robust AI Actors/Bots for all the enemies as well as survivor bots, which allowed them to assume the survivor team was full – reduced the number of variables to balance. Assumed the baseline. And gave them drop-in/drop-out for free, which was critical given how long campaigns were.. Another really cool thing was they could do automated testing, and the AI Director needed Ais to direct.

Summary – players really do like to help each other – if your game is structured to facilitate it. Procedural Content is an excellent solution for a replayable multiplayer experience – avoids scripting becoming really complicated. Allows smaller teams. Dramatic anticipation and structured unpredictability are very powerful, and simple algorithm can generate excellent pacing schedules. The sales seem to bare it out – 2.5 million retail sold, #1 new IP on it’s platforms.

Question (Borut): Do player’s notice the Relax timeframe? Really got that number from playtesting. Lots of tweaking. Effects of too long was boring, too short was boring and people burned out. Question: Why not limited ammo? Militantly simple. Only handful of weapons, wanted to focus on teamwork – game is not about zombies but teammates. Experimented with different ammo situations early on, but just not fun to run out of ammo. Question: why are sounds for the Special Infected not structured unpredictable? Really important not different, just where you encounter is different. Want to encourage recognition of Special Infected as a skill. Question: Non-combat co-op mechanics? Did combat specifically because known quantity. The sharing or the help mechanics were shockingly compelling. To the point that there seems to be a class of players who play just to help others. Question: How do Dramatic Anticipation with turn-based? Depends how you quantify time. Give notification and then time to plan to deal with it. Just enhances the contrast within your game. He argues it’s the structured unpredictability that keeps people coming back. Question: Player classes? Yes, went around with it. But again, wanted to be as simple as possible. Barebones, period. No arguments about who is who within the team, who fulfills what role. They didn’t need to add it, when they playtested it themselves it was fun. Question: Random constraints for spawning, how was it authored? Mike in C++, not ideal. Became a constraint when team got really large. Question: Why does versus seem to be more popular? For long time was only versus. But split into too modes, and in versus now no AI Director. Let the player’s sort it out. Ongoing area of research. Question: Is it survival horror? No! Not survival horror, gave up on creepy early on, wasn’t sustainable. Not replayable. Just were able to do excitement, thrill. It’s a survival thriller. Question: Did AI Director mess with difficulty? It doesn’t, because it’s reactive, only backing things off once hammered. Either everyone makes it, or they all die, or it’s a great story. You have to be able to survive the first hit, so it’s maximally difficult. Question: Weapons/Health factor into enemy spawning? Disconnected, almost entirely. There’s some designer hand in it. One exception is AI Director can change pain pills into health kits.

Master Metrics: The Science Behind the Art of Game Design

Dan Arey and Chris Swain

Survey studies of game design measurements. First study – Measuring Noobs vs. hardcore – Hit ratios – figured out Army of Two there was a 4 frame firing delay that experts took into account by aiming and new players didn’t, so in the next title expect that to be 0 frames like in Gears of War.

Or controller retention test: quizzed them on certain moves between Fight Night 3 and 4. And then kept tweaking FN4 controls until retention score percent for each move was as high as they wanted. They used this to make sure everybody could play through measuring. And also found out that players thought it was more fun, as a bonus.

One more example, DICE and Battlefield asked can metrics help validate their own design? They built a matrix with a target goal and percent, a current player percent, and a why asking. The lead designer writes up the stuff he wanted for competence in the game, and the team agreed and ranked the importance of each thing. And it gave them targets to iterate for. Players would get surveys in game after they learned a new move and they would rate it right there. Junior people would monitor how players behaved and when they would fight when they weren’t supposed to. In the example 40% were doing it right, but wanted 80%, so kept iterating and retesting. They were finding that they were doing it at the end. They made game out of polishing their game – gave each department levels playtest has to pass. Best practices for persuading developers with data. 1) Developers do not read whitepapers. 2) Video record and show video of cameras hands and faces where the issues occurs, and then show it back. Measure like titles to get apples to apples comparisons. Relative accessibilitiy useful for many different types of tests.

The more data you look at, the more patterns you see in human response. Halo heat maps show organization of play in the levels, and could watch it move around as they changed and iterated. They’re talking about getting some much data you can see subtle patterns designers can’t predict. Godel, Escher, Bach – book you had to read if you were going to work at Naughty Dog – discusses the underlying structure of math in science and art. Bach would have been a programmer in the 1970s. A Fugue is a very specific structure, beautiful and flowing to hear, but highly mathematical at its core. Can be built by software. Levels are very similar to music in structure – there’s a creative heartbeat. As you play a game, you still have a heartbeat players are responding too. For early Naughty Games, they would compose the notes of gameplay. They’d count the challenges like musical notes. 2 or 3 in a row would be double/triple notes. Levels were very structured in Crash Bandicoot – interesting flow process players would detect in the back of their head. Jak & Daxter had 20-25, much higher density. Had dynamic continues that determined where in the level players would respawn. They were mathematically figuring out time between challenges, the distance each challenge covered, the length of each challenge Added timing to levels, total gameplay per area. Very much like music. Did all the way through Uncharted. If you want to ramp your challenges, they would ramp the challenges to climax successfully. A bit of linear pacing control. Consistency is the point – to create that structure in pace. They in the design phase had a wheel that would predict where players would go within open areas. And they’d space challenges out to match where players were going. And then when you get actual numbers you can react your plan to where players are actually going. A ramping heartbeat. Count the challenges a player will face. There’s a harmony like a fugue. Compile the numbers in a mathematical way. Not formula but form. Place patterns in the gameplay, it’s a manageable structure.

Another study – Immersyve’s work – 15,000 test subjects, 20 studies to date, getting at hy/psychology of what the player is doing. They measure Autonomy, competence, and Relatedness for each game using surveys and other techniques. Model has been predictive of sustained engagement. Research of motivation in games is forming around self-determination as the big theory. What motivates people is consistent acrosss different areas of life – work, play, relationships, all of it. Humans value the activities that satisfies these specific motivational needs, which are well-defined and can be measured and predictably create motivation. So Immersyve builds a system that measures those 3 things. Developers can iterate until metrics goals are met. Needs satisfaction predicts sustained engagement over 9 months – fun does not. (May be why fun is not a good fun metric). Needs satisfaction gives survey results that lead to buying sequels and playing longer.

Measuring Autonomy is the need to experience personal agency and feeling of freedom of choice. The principle is games that have more choice will satisfy gamers need for Autonomy more than games that have less. And Autonomy is the most important need player’s have. Are you giving that feeling of Autonomy? Denser spaces with less empty space is better autonomy then larger with more stuff. (My my my!) Measuring Autonomy – 1) Designer designs level, 2) Immersyve tests with real people, 3) if Autonomy satisfaction < target then go to Step 1, else Done. Example of this data from 5 different builds on scale from one to 5 among multiple types of gamers. High Autonomy satisfiers include Sims 2, WoW, GTA 4, and Little Big Planet. Why is WoW so sustainable? Maybe because it is hitting all 3 levels? If you look at the activities in the game – leveled up, items acquired, area reached, quest finished, crafted, skill raised, etc. On almost all of these levels Autonomy is exceedingly fulfilled. The Immersyve guys can actually explain why something isn’t satisfied, because they understand the game.

Measuring Competence – the need to experience of efficacy, growth, and mastery. The principle is games that allow player to develop tangible skill and express mastery of that skill deliver Competence satisfaction. Competence satisfiers like Guitar Hero, God of War, and WoW again. WoW satisfies competences a lot at exploring new zones/area, getting epic items, and a little bit in rare items and group quests. Competence is a little less important then autonomy.

Measuring Relatedness – need for meaningful connection to others, defined by others support for our competence and autonomy. Humans are intrinsically motivated to connect with others, so design for that. Wii Sports, Magic, Pogo, and WoW are all Relatedness games. WoW is particularly good with Group quests and epic items, and Relatedness.

OK, next topic. If counting the fun is like creating a Bach Fugue, thn Combinatorial Play Design is like composing a Mozart Symphony. The Art of being playful with the gameplay. The patterns and harmonies are more playful . By listing out and adding up the combinations of enemies, objects, environments, and challenges, the designer will not only Document their available opportunities, they will also inevitably Discover much more than they were looking for. Surprising patterns and combinations. The player will subconsciously Feel the breadth of possible experiences and interactions in the game. Junior designers make lots of VOIDs – varying only in degrees. Only varying enemies by adding a few more. Reality is still the same gameplay, just changing rates. Good gameplay is to avoid the VOIDs – change up the gameplay. Looking at the enemy list of Mario 3 – Miyamoto understood human interactions in games – each enemy in the game elicits a different player response, every enemy forces the player to do something. Koopas you hit twice, Baboombs you hit once and run away. Paratroopers zigzag and require different movement. BulletBill was fast moving, straight, required jump attack. And Miyamoto would then combine them in different ways. Half the guys look like what they do and then others do what they look like. So put together game design documents paying attention to attacks and defense but movement and also what makes them different to the player and what creates their unique play, something widely different. Go back to the mathematical concepts and put them on a map and comparable. Really think about your combos before you put them together. Sony, when testing to see what mechanics loved, got really good bang for the buck. Portal is based on only a few objects and obstacles, maybe 16-17 moving parts, but combinatorially can mix and match to create their puzzles and gameplay. Plan it out, graph it, and force yourself to take that analytical step to set yourself up for the Symphony.

1 more: Reasons to use User Researchers who are independent from the Devs. 1) URs are passionate about user experience, not about design specifics. Microsoft Game Studios puts about 3 URs per project. 2) URs practiced at translating user experience into something measurable. Eg Measuring chaos in a level of Halo 3. Can target towards abstract feelings developers are struggling to achieve. 3) URs specialized in research methodology. Ex. From Bill Fulton on survey design. If you do a forced choice it’s more actionable for game developers, as opposed to gradation scale. Gradation scales can hide the interpretation. User actually has to think harder and define their own criteria in the decision model. Last, URs help figure out what to measure and how to translate that into something useful. They can look at metactritic for example and evaluate whether it’s valuable or not. Eg metacritic only measures reviewers, who have the most experience. A UR measures the average gamer, a much different bell curve. Reviewers are not a representative sample. They can skew the measurements about your game. A UR will go direct to gamers and get source data.

Costs? Microsoft spends about ~3-8% of production budget on UR. They’ve found this to be the most cost effective way to use resources for delivering top products.

Question: Related music to gameplay a lot – can you manipulate/transform game design in the same way? Note Guitar Hero‘s instant fusion of music and game design. Yes, USC is looking at cultural translations.

Read Me: Closing the Readability Gap in Immersive Games

Patrick Redding

On measuring immersion. Not about accessibility or UI or uncanny valley, although consistency and interface feel ultimately trump graphical feel. Leading player through all their senses through the world, even though we’re putting all this work into these other systems. Representational depth in games is going up exponentially compared to dynamic complexity. Games now benefit from this giant bandwidth they are able to toss at the player in sharp contrast to the player’s thin-pipe input. Meager channels to input choices, and the contrast leads to a weird disparity. And almost a sense of overcompensation. So the games systems as a result a lot of times become unrecognizable and disconnected from what the player is actually doing. Shows a chart of games on representation vs. readability, and he asks how well can we make simulation/emulation to low/high readability. And there’s a clustering of games in each grid.

Elements of readability – he starts with game cues going through, to give hints to what players can do, and then feedback when the player’s take that action. Simple games use flashes, particles, UI for game cues. More immersive games will use camera, body languages, animation anticipation. And on the feedback side, people do away with the HUD now by using softer collision model yet still using level cues, animation, particles of blood, animations conveying enemy state and next action.

There’s also an element of coherence. How does the player perceive the information is actually being delivered to them – easy to recognize, easy to interpret, and prioritized. Players know when they are being told something. As games feel this kind of pressure to push info out of the HUD and onto the game world, they still need an obvious way to flag importance.

Also important is consistency – the representation should match natural expectations and respect conventions, as well as group meanings and behaviors of similar forms. Don’t violate the logic of form. The simple example is if forward it walk, then think very carefully about changing what left thumbstick does.

Finally, there’s the notion of context. The idea that there’s the game space or a possibility space which is something the player has a mental model of.

So, working through MDA (mechanics, dynamics, aesthetics – LeBlanc, Hunicke, Zubek), looking at readability you can look at even simple games and you can change one simple thing and have long far reaching consequences for the game’s experience. He describes an A-designer and A`-player, where designer passes authorship to the player after ship to play. He defines a line to the left of that A1-systemic, and a line of simulation that is A2-representational depth. Usually we add representational depth to build up the production values.

MDA represents a sort of plural vision, but we have a multiple component Aesthetic experience, an awful lot of which is not coming from MDA. Let’s not confusing aesthetic experience for aesthetic content, be careful. Why does this happen? Usually for production reasons. As we try and innovate, things get riskier, and someone decides at some point that it’s enough – that we should stop designing the depth and just throw representational aesthetics at the problem directly. Part of this is the content creation pipeline hiring and kicking into gears. Those artists should be doing something, right? It’s not difficult to justify that, but it is a willful act of self-deception. There’s no relationship between the procedural and non-procedural aspects of those aesthetics.

Jonathan Blow at the MIGS 2008 keynote articulated this succinctly – games are becoming a place for meaning, but there’s this overwhelming urge to put all this dramatic tension in the game. But they are antithetical and it creates procedural fiction. Jon doesn’t think we can solve that problem, and Patrick’s not sure he agrees, but while traditional authorship might feel safer, doing it can be antagonistic to meaningful play.

One of Far Cry 2‘s goals was immersion. Wanted the player patching themselves up, and they found that required a huge amount of data – graphics, animation, audio, physics. This is an action that happens from pushing one button on the controller. We’re adding a huge amount of representational content to the gameworld, but it’s only one action, so we’re creating a procedural disconnect from the simulation. Not to say that it was bad, but there was a readability impact. Game information is now both game system state and ‘verisimilitude’ and player needs to be able to distinguishes these two.

Frank Lantz talks about how we fetishize immersion. He says immersions probably a good thing, but all games are procedural simulations. Players get satisfaction from trawling the boundary conditions. We should be careful not to get seduced and cleaning up these rough edges and preventing their satisfaction.

There’s this unreliable witness problem. Groups that get debriefed do markably worse. Discussing results corrupts your memory of what you already saw. As players move back and forth between representations and understanding systems, we may be degrading their understanding of the game. Might be the root of the readability problems?

Dynamic narrative systems had lofty goals. Wanted emerging story from the player’s regular gameplay. Wanted to have designers help craft the player’s input for them. There’s some set of local minima. We want the progression of local minima in order to yield a narrative dramatic arc. One of their failures is they have this story output to the player and the player’s game input, which are very much divorced and causally disconnected – sheer timescale of hours. Doesn’t support meaningful play. Presenting buddys and unique characters but not linking them readably to the game’s system, so the player’s investment is in the representational level and not in the deeper systems themselves. We’re reinforcing the wrong part here, and pulling players away from the meaning, which is in the systems themselves.

His point is we might have dug ourselves in a hole and gone done the wrong road. Recent art games are simple maybe because they are trying to get out of the hole. Let’s not get preoccupied with representational, let’s roll back a bit and get simpler and find things people are familiar with and find meaning in that. Down the road we’re going to do a better job of balancing the two here.

Interlude – parts that need to be fixed. Redoing production. Fight design precision in decision. Fight lack of incrementation in abstraction phrase. Actual design is happening during implementation, by programmers who know about design then they want to admit. Industry now dominated by game designers without programminng backgrounds. And all this iteration and playtesting is more then finding the fun, but players understanding their choices. And we’re fire-proofing our interface schema.

Proposes an actionable readability model. Redefines giving player’s visibility of choice. Not giving away the best choice. Take a look at Portal. Portal is linear, constrained, FPS, and subversive. It subverts every aspect of the game. The game environment is highly linear, and the game is about non-linearity, projected topology. Subverted. Challenge design is intended to give the player a huge range of options. Implication is that the player is reading the game and figuring out the biases of the environment, that favor and contract a set of choices. Looks at Portal the Flash game – very similar on paper. But it has complete overview of gamespace, perfect information. The real portal differs because you can look through portals, a level of spatial information. But the subtle information comes from parallax, moving back and forth through portals, giving you additional spatial information. It’s a unique instrument of readability in Portal.

There is a number of scales on how player’s make choices, on importance of each choice, and measuring upside, down side. Wants a more readability-centric scale. With threats, progress, conditions of locality, agency/boundary/constraints, and state of the avatar/self. Player’s read these things, and are presented with opportunity to map a number of choices onto a number of different axes. The player evaluates whether the choice is obvious, immediate effects, and whether information to base it on is complete or incomplete, at the very least.

For readability, player is looking at individual game ingredients and the game as a whole. There’s one channel but multiple dimensions. Different players see more in the information presented as they get more system literacy, learning from the A all the way down to the M of the game (and back up). Also, if you are simulating it, there is less readability in it . Far Cry 2 has way more “noise” then Doom, and we shouldn’t be trying to simulate everything. Systems get fuzzy and ambiguous. Special case representations fail though because they can be simplistic and misleading, so they aren’t perfect either. Designers favor systemic over the special case because it’s harder to make special case more granular, systematic easier to fix. So usually combine, and have to look at readability of both representations. In Thief, torches are a dynamic system of light. Player can try and snuff these out with arrows, which have systematic physics, but the rule that snuffs the torch out is an emulated rule. Point is there should be some consistency between systems.

Last principle is game systems are only readable if players can perceive the relationship between game cues or feedback, and the player internalizes the game’s metaphor. A working model.

We need to solve these challenges – breadth of simulation, depth of abstraction, and how we integrate them together. We need to introduce these ingredients and keep them consistent, represent them procedurally and affordably. He presents a challenge management pipeline for readability we can use throughout production. We need core activities at every given moment understood, with motivating goals. We need accurate actions devolved into player verbs. What is your gameplay based on in most simplistic terms of player verbs? We need low-level controls from that and understand the range of meaningful input and whether we’ve communicated that to the player. Do the player’s understand what the verbs do? Can it be tuned in response to pre-release player feedback. Then, we need this palette used by level designers – player’s using them in the context of the larger game. Beyond a bag of parts, but a tested and known readable set. And then can we look and see this parts in playtesting?

It’s just one possible tool – a lot of different ways we can do this. But we can’t preempt the problem through documentation – we have to get it on screen.

Question: Longer game, change readability to reflect trawling? Trawling is a willingness to step outside fiction, so not sure we want to break metaphorical cushion in response, likely to obscure metaphorical consistencies. Question: Is readability always good? Can’t think of alternates. Even if a player is willingly suspending disbelief with imperfect information, there’s a contract with the designer that defies designer abuse. And obscuration is a form of abuse. Question? For social play (implication about hidden information on systems being useful)? With a social motive, people are interested in gesture, we don’t gain a lot by obscuring how to gesture. So don’t interfere with those affordances of the game? Question: Eternal Darkness insanity mechanic? Exception that proves the rule. Question: Mercs in Far Cry 2 bad representational examples? Not as bad as the narrative problems. People though hated the respawning checkpoints. One of the problems is that respawned checkpoints aren’t always readable because the AI is off doing something to the side, and lead the player into making mistakes. Or situations in Far Cry 2 required preparation work, but there were systems like malaria had a disruptive effect. They wanted to have more analog failure, but readability was hard, and it hurt accessibility. Question: Geoffrey Kaplan on WoW quests – improved readability through exclamation marks – improved readability? Used to think it was a joke, but can get away with a lot more then you think you can. There’s this escalating ramp in the dev community, and we’re probably a bit too hard core about it. There’s always ways to do it that are subtle and well-executed. The HUD can be just good graphic design, and fun to look at. WoW sacrificed immersion there for readability. Question: Augmented readability through scans, but restricted at baseline play? Is that really a degradation of readability if it’s not concealing information about the game’s systems itself. This is almost adding more info, above and beyond.