The toy room

(Back to TOC)

9 December 2004

by Mike Rozak

One of my cousins had a huge toy room with an order of magnitude more toys that my brother or I had. Whenever I visited his house, I saw the toy room as nirvana, and thought someone with that many toys would never get bored.

As an adult, I know that conclusion to be false; kids can be just as bored with 100 toys as they are with 10 toys. In fact, I've come up with some fundamental observations about toys:

1. A toy room with more toys will provide more entertainment, but the amount of entertainment is not a linear relationship with the number of toys. Thus, twice as many toys almost always results is less than twice as much entertainment.
   
   This law is not always true:
   
   
   • 12 toys received all at once for Christmas will result in 4-8 weeks of entertainment for all the toys. 12 toys received once a month over the year will produce 1-2 weeks of entertainment per toy, or 12-24 weeks of entertainment.
     
     
   • Some toys are synergistic. 1 Lego set provides 1 hour of entertainment. 2 Lego sets provide more than 2 hours of entertainment because the pieces from each set can be combined in new ways.
     
     
2. A toy room is more fun when enjoyed with a bunch of friends. Old toys are rejuvenated when friends stop by to play.
   
   
3. If an adult joins in the play, toys can be kept fun longer, perhaps an infinite amount of time longer. Unfortunately, adults cannot spend all day playing (or they get bored of it), so adult-mediated play is a rarity.
   
   The reason why adults greatly improve a toy's fun is because:
   
   
   1. They have more experience, and are able to come up with new ways of playing with the toy that children haven't thought of.
      
      
   2. They're an authority figure, and can direct the imagination a bunch of chaotic kids into more sustainable activities.
      
      
   3. Their play is selfless. Adults make play decisions designed to make play more enjoyable for the kids, not necessarily themselves.
      
      
4. Because adults cannot spend all their time playing, they sometimes create scavenger hunts, which a pre-programmed play sessions. A scavenger hunt has the kids play with a series of toys in context to provide more meaning. Scavenger hunts aren't as good as a real adult, but they're usually better than a free-form toy room.
   
   
5. When I was 12, I began programming games on the Apple ][ after I read an issue of Byte magazine that had an article about Zork, along with the Basic program for a simple adventure game. Ever since, computers have become an everlasting toy for me. I could be locked up in small room, and as long as I had a computer, I would keep myself entertained.
   
   I don't know if a programmable computer is the ultimate toy (for me), or if at 12 years of age my brain matured to the point of being able to entertain itself.

Virtual worlds and toy rooms

As I have stated in "Virtual world as platform", I often think of virtual worlds as being collections of sub-games. Or in other words:

a sub-game in a virtual world = a toy

a virtual world = a toy room

If this is the case, then my observations about toys infer the following observations about virtual worlds:

1. Offering more sub-games will allow a virtual world to hold a player's interest for longer, but not linearly. Twice as many sub-games do not provide twice as much entertainment, unless:
   
   
   1. The sub-games are rationed out to players over the course of time, such as quarterly updates with new features, or limiting sub-games to quests.
      
      
   2. The sub-games are synergistic: Combat is fun. An economics game is fun. If the two are combined they're even more fun.
      
      
2. A virtual world is more fun when one's friends are around. To a lesser extent, having unfamiliar players as well as enemies also makes them more fun.
   
   
3. A virtual world with a game master (filling the role of adult) is more fun than free-form play. Game masters work well in tabletop role-playing games, but aren't economically viable in most online virtual worlds. A few worlds have extensive game mastering, and many have occasional game mastering. Most have employees that they call "game masters", but who are really in-world product support staff.
   
   
4. A quest (or adventure game, or offline CRPG) is the same as a toy-based scavenger hunt. Quests are not as good as having a real game master around, but they're usually better than free-form play.
   
   
5. What is the everlasting sub-game for a virtual world? Is it the act of creating a world? I don't know.

Toy categories

Over the thousands of years that people have been inventing toys for children, they have only created a handful of toy categories, between 10 and 20 depending upon how you count. Some of the toy categories, as listed by E-toys, are: Action figures, animals and stuffed toys, arts and crafts, blocks and building sets,  dolls, DVD and VHS, etc. Each category has thousands of variations. For example: Action figures can be GI Joe, superheros, Bionicles, transformers, etc.

Virtual world sub-games likewise fall into categories such as combat, economics games, chat, puzzles, etc. These two can be subdivided into thousands of variations.

I perceive several weaknesses in contemporary virtual worlds:

1. They don't provide enough sub-games, and put almost their effort into combat. It's like a toy room with only a handful of toys. See Virtual world as platform.
   
   
2. In general, contemporary virtual worlds all provide the same small set of sub-games. Therefore, they cannot differentiate themselves by their choice of sub-games. Instead, they differentiate themselves with eye candy, genre, and by the details of the sub-games. One virtual world will have GI Joe action figures, while another one superheros, and a third will be transformers. Yes, there's a difference, but not much of one.
   
   
3. Most contemporary virtual worlds subscribe to the "sandbox" theory of design, which postulates that if players are provided with enough toys they will make their own fun.

Because virtual worlds don't have many sub-games and they are just sandboxes, then their only differentiator is item #2, the specifics of each sub-game.

I am always looking for ways to differentiate, so I advocate more sub-games (#1) and a break from the sandbox theory (#3), as well as unique implementations of sub-games (#2). More sub-games and different implementations are easy enough. However, to escape the sandbox, I need to have a game master, but since an army of game masters is too expensive, I need one based upon artificial intelligence.

Replacing the game master

Being a computer programmer, I try to solve every problem I come across by writing a program. Obviously, I want to write a program to replace a game master...

At its most basic, a program to replace a game master (or adult) would pull out a sub-game (toy) for the player to enjoy. After a fixed amount of time, the computerised game master would distract the player and pull out a new toy. Repeat the process. It's a simple idea that would only work if the players were utterly stupid, which they're not.

I can think of a few improvements though:

1. Have the AI game master try to determine how much the players are enjoying the sub-game. If they're still having fun, stay with the sub-game a bit longer. If they're "looking" bored, either try a variation of the game that the players haven't seen (such as having them kill giant rats when they get bored of killing ordinary rats), or abort the sub-game ahead of schedule and try different sub-game.
   
   
2. Intelligently determine which sub-game would be good to use next. If kids are getting bored with GI Joe action figures then pulling out a Spiderman action figure might work, but a much different toy, such as the spaceship, might work better.
   
   
3. Find a better way to distract the change in sub-games. If kids are becoming bored with GI Joe action figures and all an adult has to go with is a flying saucer toy, then have GI Joe and friends be abducted by aliens. From there, play with the flying saucer. When the kids get bored of that, the adult can have the flying saucer drop GI Joe and friends on an alien planet with giant trees that look like lamps, or forget about GI Joe entirely and have the flying saucer be attacked by a giant space dragon.
   
   
4. Repeat.

This is easier said than done...

Determining if players are enjoying sub-games

An adult easily knows when children get bored. The same goes for tabletop game masters. A game AI acting as an game master would find this problem much more tricky. To do this, an AI would need to:

• See the players (a video camera) and read the players' emotions (emotion recognition)... I don't think emotion recognition works nearly well enough for this.
  
  
• If a game could get players to wear lie-detector apparatus it could determine how much fun they were having. This obviously won't happen, although an IR camera, a chair that detects player movement and sweating, and a mouse with sweat and pressure sensors might provide some information... Not many players will pay the $500-$1000 this system would cost though.
  
  
• Of course, an AI could detect if a player stops playing, or if the player's pace of play drops off. Such "subtle" hints would inform the AI that the sub-game was boring, but it might be too late by then.
  
  
• Alternatively, an AI game master could occasionally ask players if they're having fun yet. Asking too often could get distracting.
  
  
• For all of these detection mechanisms, the AI would need to identify players that were faking their behaviours in an attempt to fool the game-master AI. Some games attempt to tune a game's difficulty to their players' skill levels, only to have the players game the game-difficulty AI.

Personally, I prefer the "ask the players if they're having fun yet" approach, but only ask once in awhile. To do this:

1. Divide the virtual world into hundreds of quests. (Most virtual worlds already do this.)
   
   
2. When a player finishes a quest, ask them to rate it from 1 to 5. The polling is a bit intrusive, but not that bad.
   
   
3. Use various mathematical techniques (whose names I have forgotten) to cluster players' preferences for quests and come up with a model of what type of quests a player would like. Amazon.com uses this for recommending books, and while the recommendations aren't perfect, they're pretty good.
   
   The selection process is a bit tricker than Amazon.com's because it needs to take into account the player's location and his character's skill level, as well as the player's overall progress in the "storyline" of the VW (if there is one).
   
   
4. Either recommend these quests to the player, or somehow guide the player character to the quests. For example: If a virtual world learns that a player likes to rescue pet cats, it would have NPCs from the pet-cat-rescuing quests seek out the player from miles around.
   
   Some people do not like being second-guessed though; The virtual world would need to provide an option so players could turn off their personal game master AI.

Choosing the next sub-game

In order to choose the next sub-game for the player, the AI game master needs to know:

1. The player's frame of mind. Thus, the AI must have a "theory of mind" about what the players are thinking and what their moods are. At its simplest level, if the player is obviously getting bored of killing rats, swapping them over to killing giant rats won't be much of a change. Having them solve puzzles or sail a ship would be better.
   
   
2. What the players' characters can handle. There's no point in bringing out the dragon "toy" if it will instantly incinerate all the players' characters. Likewise, forcing a thief PC to fight a NPC knight in an arena isn't too fair.
   
   
3. Where the storyline of the virtual world is leading. If the storyline dictates that the players must meet up with the evil overlord's henchmen at some point, is now the right time for him to appear? (If you don't believe in storylines then ignore this statement.)

How can AI solve these problems? I'm not entirely sure, but here are a few ideas:

1. Manually classify each quest (or sub-game), such quests with "lots of combat", or "lots of puzzles". (Or maybe ask players classify the quest type when they finish it.)
   
   When a player finishes one quest, determine what quests the player could undertake. If more than one quest is available, try to recommend the quest that is the most different from the last quest the player completed. (Of course, if the player doesn't like puzzle quests, then don't recommend puzzle quests, even if they'd make a good change from killing things.)
   
   
2. When players complete a quest ask them how difficult they thought it was. Plug the difficulty, along with the players' skills, into some clever AI and produce a heuristic. When considering the quest for future players, plug in their characters' abilities and guesstimate how difficult they would rate the quest. If it's too difficult (or too easy) then don't recommend it.
   
   Some players, with the same character skill as others, may still find quests too difficult or too easy. The AI will need to guesstimate how skilled the player is (as opposed to their character) in relation to other players.
   
   
3. The quest's recommendation score is also affected by how important it is for the storyline, and whether it must be run in order for the story to proceed.
   
   A user may sometimes have to complete one out of two possible quests for the storyline to advance; choose the one which the player would enjoy the most. For example: To kill the evil overlord the player may need to acquire a cloak of morphing. Since all players must eventually acquire the cloak, a virtual world might provide 2-4 quests that result in the player getting the cloak, but recommend only one of them to the player.

Which quest the AI recommends next can also be affected by:

• How far away the quest is.
  
  
• How many other players (or groups of players) are already participating in the quest.
  
  
• If a person (or players) somehow enters how similar two quests are, the AI can ensure that similar quests are not recommended too close together. Two FedEx quests in a row are undesirable, even if one is combat intensive and the other is puzzle intensive.

Transitioning between sub-games

You may not have noticed, but the examples I gave used quests as atomic units, not sub-games. The reason is simple:

Producing AI to transition between quests is fairly simple. Producing AI to transition between sub-games is very tricky.

To transition the player from one quest to another, the AI merely needs to (a) unobtrusively inform the player that the quest exists, and if the quest is a long ways away, (b) find a way to get the player there. Both of these problems have simple solutions, although more complex solutions would produce smoother (and less obvious) transitions for players. For example: A really clever way to get a player from point A (where they finished their last quest) to point B (where the next quest is), is to use a small quest in-between that coincidentally transports the players from A to B.

Transitioning between sub-games is much more difficult because there are more transitions; a player may transition from one sub-game to another dozens of times an hour. If these transitions are not 100% smooth, players will notice them and become annoyed at the artificiality of the system. Players will also notice transitions between quests, but because a transition is only needed once every few hours, the transition won't grate as much.

To do a proper sub-game transition system, the game master AI needs to have a huge expert-system knowledge base. For example: Lets say the AI determines that a player is getting bored with killing orcs, and decides that a car chase would be more fun for the player. How does the AI explain the transition? Do some of the attacking orcs suddenly jump into a car and take off down the road? Where did the car come from? Where do the players get their car? Can orcs even drive? What if the players were attacking a heard of wildebeests? Do they drive? What is a car doing in a fantasy setting anyways? What happens if the players decide to keep the car they pilfered?

To make things even more complicated... What happens if there are no roads nearby? Does the AI invent a road? If it does, what will players think when they suddenly see a road appear? What do players in a different group think when a road suddenly appears, and only a few moments later a carload of orcs comes hurling towards them on the newly invented road? What happens when the world is filled with too many roads?

And, that's not the end of it... What happens if the car-load of orcs, in an attempt to escape their attackers, have a head-on collision with a dragon that another group of players was attempting to kill? The dragon ends up with a broken leg and must be put down, while the orcs are all flattened and rushed off to hospital. The experiences for both the orc-chasing players and the dragon-slaying players have just been ruined.

That's not the worst: If the game world starts dealing with individual sub-games, it should conceptually tie these sub-games together into quests. (It doesn't have to, but people seem to like having overarching goals.) Creating automatic quests is fairly tricky, and not something that contemporary AI is up to, except for simple quests.

Automatically generated quests

If you read, "Automatically generated content", you'll know that I like the concept of automatically generating quests. I don't think, however, that such quests will be very interesting, and they will pale in comparison to the hand-written ones.

Automatic quests could easily be written and effectively used for transition quests, however. In the example I used above, where a quest is needed to get the player characters from the city where one quest ends (A) to the distant city where the next quests beings (B), an automatic quest would be a cinch... I can imagine a few scenarios:

• A FedEx quest where the player is asked to deliver an object to B.
  
  
• The players hire transportation (such as a ship, airship, train, etc.) and have a small quest along the way, which could involve a bit of combat or maybe even a simple murder mystery. (What other quest would you put on a train?)
  
  
• If the players go it alone and buy horses, part way through their journey they are ambushed by bandits or find a small tomb.

Of course, players will realise that the quests are automated, but I don't think they'll mind the occasional automatic quest.

Theory-of-mind in quests

Even though only a small percentage of quests can be automatically generated, every quest can be tailored to its players by understanding what the players like:

• One obvious tailoring solution is to adjust the difficulty of the quest based on the player characters' skill levels. This is not without its perils though, since many players will attempt to fool the AI into making the quest easier.
  
  
• Adjust (or provide) solutions to quest problems based on the player character's abilities. A wizard character needs to be able to complete a quest using magic alone. A fighter should be able to use combat alone. Or is this a mistake? Maybe it would be better if a quest encouraged players to team up. Maybe quests should only be solvable by certain classes so that players are encourage to replay the game using a different class?
  
  
• Provide friendly NPCs who team up with single players and help them finish a quest. If a group of players comes through the NPCs won't join up.
  
  
• Adjust the gender of NPCs based on the player's genders. Male players get to kiss a female frog and turn her into a princess, while female players turn a male frog into a prince.
  
  
• Likewise adjust the race of NPCs based upon the player's race. A party of PC elves will be attacked by orc bandits. A party of orc PCs will be attacked by Elvish bandits.
  
  
• Other aspects of quests could be adjusted, but doing so may completely ruin the impartiality of the game: For example, adjusting the treasure given out to suit the player... Is such treasure properly earned?

Did I go off topic?

How did I get from toy rooms to game-master AI?

Simple really...

1. A toy room is a source of entertainment, but it's much more entertaining when an adult selflessly partakes in the play.
   
   
2. Virtual worlds are conceptually toy rooms. They're more fun when a game master is selflessly intervening to make the world more fun.
   
   
3. Since game masters are too expensive to employ, an AI will have to do. (Some people may object to AI second-guessing their desires, so provide players an option to turn off the AI.)
   
   
4. Given the lack-of-sophistication of current AI, the AI's functionality is limited to recommending quests.
   
   
5. Important quests must be manually generated. However, simple automatically-generated quests are possible, and are particularly useful as "bridging quests" between the manually-generated quests.
   
   
6. Even important quests can be minimally adjusted to take the players' likes and dislikes into account.

Some additional thoughts...

If several players frequently team up into parties, how does the AI handle each individual's different motivations? Are there quests specifically designed for teams vs. loners, and does the AI use its knowledge about the party?

If an AI notices two players (or small parties) that seem to have similar habits, does it try to get them to bump into one another? Is there a gender and age bias? (See Build it and they will come.) What if a player wants to play alone?

Does the AI attempt to produce conflict by encouraging enemies occasionally run into one another?

Alternatively, does the AI pretend it's Charles Dickens and have the same NPCs reappear from time to time? Some favourite NPCs from the past could be kidnapped by enemy NPCs from the past.

Do the quests, with the help of AI, attempt to intertwine the players' experiences? Maybe one quest will call for a character to steal an item from a NPC. The NPC might then choose to hire another PC to get his item back. Which PC out of 1000's is targeted for the retrieval quest?

Is the amount of PvP in a quest also monitored? Does the AI remember how much PvP a player likes and recommend quests based upon that?

How does the AI attempt to create strong memories, as mentioned in The authoring equation? Are some sub-games only brought out occasionally in order to create an extra-sharp memory of them? Or are combinations of sub-games used? Or the context? How are other players involved? Are important choices necessary for strong memories?

How are inconsistencies and extraneous quests minimised? The storyline that the player selects is one method. (See Intertwined storylines.) Are there others? Are there sub-themes to the storylines that should be emphasised for specific players?

If a player only follows recommended quests they will finish in (lets say) 100 hours. If they play all the quests in the world, play with several characters, or try several storylines, they might get 300 to 1000 hours of entertainment out of the virtual world. Does this solve the problem presented in The anti-MMORPG? Does this make the virtual world too expensive?

The possibilities are limitless... and that's the point. Adding a game master AI can significantly enrich the world. Some players won't like it; they can either turn off the game master, or go back to their sandboxes.