(These are notes for a speech, somewhat tidied for the Web)
by Doug Sharp
I am Doug Sharp, the world's leading expert on interactive fiction. I've come here today to share my breakthroughs and to present solutions to all your thorny problems.
How about this, that by the luck of the draw I'm up here today spouting off, rather than one of you. I'll try not to belabor the obvious, maybe raise a bright idea or two, and leave plenty of time for discussion at the end. And if I don't make a few real blunders I'm playing it too safe and not doing my job.
Interactive fiction is tough. I've done one work, The King of Chicago, which I consider an honorable failure. I hope that's a feeling not unfamiliar to people in this room. I'm now collaborating with Kellyn Beeck on a work of IF for Activision. My first game, ChipWits, had nothing to do with fiction and was by far the easiest of the three.
To lay out my biases up front - I come to computers with a background in the arts, primarily literature, so I value words, and that translates into dialog, the speech of my characters. I also love animation and so that's another big part of my approach to IF - moving pictures. I'm not excited by creating puzzles or parsers although I've enjoyed playing games built around them.
I am not inclined to over define Interactive Fiction. For the purposes of this talk it's a computer program with some story and some game in it. It's boundaries are fuzzy and it encompasses games from Ultima to Portal. I've seen much better definitions but the more precise they are the more they seem to invite exceptions.
Interactive Fiction is a battle on many fronts. The computer and the artist have finite resources. The game and the story battle over the computer's RAM, CPU cycles, disk space, and the artist's time, skills, and creative juices.
The story wants to be told and it doesn't like interruptions, much less interrupts. It's the Ancient Mariner who wants nothing less than to buttonhole the guest and pour out its soul. The game wants to play, it's the frisky puppy that wants to fetch the ball and can't sit still for a second. Getting these two to share a disk can be a headache, a disaster, or a challenge, depending on how much sleep you've got in the past week.
Here's a pseudo-systematic look at the two sides in the battle. I've assembled a far from complete set of elements and methods for both the story and the game.
In this corner, The vertical axis,- Fiction- the story. The quality I attach to that is depth. Story elements are headed by plot, I am talking about commercial works here. Plot I've subdivided into PROBLEMS, MOVES, and SOLUTIONS, using a set of terms borrowed from Thomas Pavel in his book, The Poetics of Plot. There have been many plot grammars developed, but this seems to be the smallest useful set of elements. By character development I mean anything that makes a character more believable, real, from distinctive graphic look, to a cumulative pattern of behavior, to verbal quirks. Anything that breathes life into a character. Establishing setting - grounding the tale in some sort of world. These are very traditional elements of human narration and it's hard for me to conceive of a commercial work of IF that isn't built of them.
On to the methods - This is a grab bag of techniques I've used and seen others use to deepen the story. Dialog is primary for me. descriptive text less so, but a vital part of text adventures. Graphics I find important and sound fx and music are becoming more so as machines become more capable and project budgets grow.
In this corner The horizontal axis - Interaction - the game. Chris Crawford will explain this later today in his talk. The quality I attach to this axis is breadth. When I tried to come up with a long list of Game elements it kept reducing to these two - making choices and exercising skills. Even these two are not discreet qualities. Arcade skills can be used to make choices. My incomplete list of game methods includes parsers, arcade action, menu and xy selectors of discreet options and geographic navigation.
The terms Breadth and depth I use narrowly, depth of story development, breadth of gaming freedom. A novel is said to be broad if it portrays a large, well-populated world, while a non- narrative game is often called deep if it is replayable. Perhaps these other meanings make the terms useless for my purpose, but what the heck, the overhead was already made.
There is a tremendous tension between the elements and methods of Fiction and those of Interaction. But the fundamental resource battle is between choices and plot. This is one of the obvious points I promised not to belabor. As player choices are added potential plotlines explode exponentially. This is the reason that works of IF generally fall much closer to one or the other of the axes.
This is my ultra-rigorous taxonomy of works of Interactive fiction. Again the terms narrow and shallow are not used pejoratively but because they add pizzazz to the chart. What is meant is narrowness of freedom and shallowness of story. I've put in plenty of days of wizardrying and hitchhiked far enough to go mad trying to grab the babel fish. I include The King of Chicago to place me on the scale and to mingle with these classics.
On the chart, heavily fortified, lies The Forbidden Zone, unfortunately not a game you can buy today but a dream of what a work of IF can be.
Here lie the games we can play only in our minds - inspired improvisation, where every action evokes a meaningful reaction and anything can happen. But as Eliot didn't quite say, "Between the conception and the creation, falls the software..."
For me, the vision of what interactive fiction might be is rooted in my childhood. My friends and I loved to play and live in the middle of wild fantasy worlds, dangerous, inhabited by a menagerie of exotic animals, deer with ruby eyes and gold hooves, flying monkeys borrowed from Oz, and lightning transformations of setting, relationships, personal powers, weapons. If only such games could be played among an illusion of reality more real than cinema.
These are games that tailor themselves to the participants needs, wishes, whims. Games that unfold of years as the player returns and leaves and grows and forgets and ages.
These games of the future are obviously generated on the fly. Not even a million Shakespeares pounding on typewriters could script all the dialog needed to cover every fork in the potential story space. The other characters are genuinely intelligent and coordinate behind the scenes to present a seamless interactive narrative.
The bulk of this talk and my own efforts in the field of IF have to do with story making, story crafting, and not the story generation in the games we can imagine. It seems like a doable task though, the basic plotlines have been well analyzed and certainly there are software artists clever enough to encode the available story elements and craft algorithms to order them meaningfully in infinite combination. I do believe that such games could be made right now, games that create a story as they are played. The only problem is that the interactive stories that could be generated today wouldn't give me what I want in a story - living characters speaking vivid prose.
This is the dialog gap, the Artificial Intelligence gap standing between us and the story games we hope our children can play. People tell stories and computers ain't people yet. I believe that story-telling is one of the highest forms of human communication. In telling a story or writing a novel the creator becomes many people and each one has a voice. It's a long way from slicing and dicing plotlines to generating one memorable verbal exchange between two characters.
With that in mind I present the magic lantern slide which will repay you for getting up this early on a weekend. A glimpse of the future of Interactive Fiction.
Doug Peers into the Future, this graph reveals the next 41 years of IF history. Horizontally I give you Time in years, vertically ART measured in milliprousts, or one thousandths of the qualitative narrative output of Mr. Proust, who wrote some good books. The era that most concerns me is that between now, accurately pinpointed at 1989 I believe, and the spot marked Doug retires, in about 30 gloriously successful years.
The two quantities I track over time are the maximum narrative quality, measured in milliprousts, of the best work of Interactive fiction in each year using each of two approaches - IF STORYMAKING, making interactive stories through hand plotting, scriptwriting, and crafty use of interaction. The other approach, IF STORY GENERATION, generating interactive stories through algorithms and AI.
The term Story Making, as opposed to Story Telling, I owe to Brian Moriarty. There has been quite a passionate and entertaining exchange concerning these two approaches on the Journal of Computer Game Design BBS. I recommend it.
You can see how I'm betting, that during my working years the best results will come from the STORYMAKING approach. Again, in my unhumble opinion, this is because of the dialog gap, the AI gap. After I'm put out to pasture, though, I do see a reversal, when computers first equal, and then surpass, human intelligence. At that point of course all bets are off, but I would say there's a good chance that some excellent games could come out of collaborations with true artificial intelligences.
It's nice to finally have a reliable milestone for computer-human equivalence. I owe this to Hans Moravec and his fun, radical little book Mind Children, which I recommend. Fasten your seatbelts when you read it.
(MIND CHILDREN) The slide that goes here is missing. It ís a page straight out of the Moravec book which Iíll have to scan for this doc.
A little side excursion to look at the page in Mind Children in which Moravec proves that we'll all have a conscious intellects on our desktop in 2030. His data points are all the familiar machines of history from Babbage's clockwork, had it been built, through ENIAC, various IBM's to Apple II, Mac, Sun and Cray. His vertical plot is not of raw computational power but of Computational power per unit cost or Cost of hardware for human equivalence. His chart estimates that today it would cost somewhere around 100 billion dollars to build a computer to equal a human mind. What I like about the guy is that he doesn't just pick his numbers out of thin air. He lays out all his assumptions for defining what human equivalence means and explains exactly how he went about plotting each data point, so you're free to disagree or laugh at any step. It's good to find another guy who believes that if you put things on a graph, they're that much closer to being true.
Enough of that. Actually I think it's very reassuring that the human equivalent computer continues to hover around 30 to 40 years in the future. That seems to be one of the few constants elements of cybernetics. Just give us a few more decades and then we'll have this AI problem licked. If anything, that goal seems to recede farther into the future which each passing year.
I am enough of a cyberhead, though, to believe smart computers will actually arrive some day. I'm hoping they arrive soon enough that I can download into one to enjoy my dotage. Following Moravec again, I really want to retire into a nice cozy little CPU in the heart of a neutron star and play some really good games while I wait for the universe to collapse. A common dream of old age.
When true AI does arrive, I think it will pretty much sweep away the groundwork we're laying today and set up its own rules for interaction, narration, and a few other details.
In the meantime, there's a lot of fun to be had making the best games we can with our primitive machines and methods. Daydreaming of the IF games of the future is stimulating, inspiring. In fact it's inspired my unofficial motto "Counterfeiting the games of tomorrow today."
Which brings us back to the purported topic of this talk. Game vs. Story, the battle of Interactive Fiction.
After all, for a lot of us Interactive Fiction is more than a quest, a dream, it's how we're trying to earn some bucks.
In working on the talk it became clear to me that I couldn't just hang the whole thing out in the air and talk about the battle of Interactive fiction in the abstract. A lot of that topic is really pretty obvious. Letting players butt into a story blows the amount of plotting into the stratosphere if you don't impose constraints.
So rather than making myself look foolish by pontificating a set of general rules for creating works of Interactive fiction, I've decided to make myself look foolish by postmorteming one of my own projects. Believe me, this is not a commercial. As I've said, I consider the King of Chicago an honorable failure. I'm not going to do a book report about the plot. I'll go light on the development details, heavy on the design decisions, and avoid talking about the publisher to avoid a costly slander suit.
I hope my approach to producing a work of IF is odd enough to slant a different light on some of its problems.
For those few or many, who may have been too busy to play the King, a one line synopsis of the story. A young punk gangster named Pinky Callahan tries to take over the Northside gang and then all of Chicago right after Al Capone is sent to prison. Now a line about the game. The player chooses between courses of action for Pinky by clicking on alternatives presented in thought balloons and, oh yeah, there are a few arcade sequences sorta tacked on.
In setting out to do the King, there was no question in my mind but that the story would take precedence over the game. Again, my biases showing, but also a conscious decision about how I would allocate my time during the development period. I wanted to write tons of tough, streetwise, smartcracking gangster banter. I watched many, many gangster films.
I had a mental image when I set out to put the King together as a work of Interactive Fiction. A guy in a projection both with hours and hours of film about a group of gangsters. The film is not on reels but in short clips of from a few seconds to a few minutes long. The clips hang all over the walls of the projection room. The projectionist knows exactly what's on each clip and can grab a new one and thread it into the projector instantly. The audience is out there in the theater shouting out suggestions and the projectionist is listening and taking the suggestions into account but also factoring in what clips he's already shown, because he wants to put together a real story with a beginning, middle, and end, subplots, introduction and development of characters and the whole narrative works. I wanted to minimize hard branches, to keep the cuts between clips as unpredictable as possible. Yet the story had to make sense, guys couldn't die and reappear later, you couldn't treat the gangster's moll like dirt and expect her to cover your back later.
With that image in mind here's a loose description of what I came up with to implement it. Basically big bags with smaller bags inside which hold the clips, which I called episodes. These little things that are supposed to look like a film clip with a little tag on top. An episode specifies things like who is to be onscreen, chooses background graphics, switches between close-ups and medium shots, drives the character's animation and facial expression, and feeds dialog to them. The episode script reads a lot like a playscript or screenplay. A single episode might put Pinky and his girlfriend Lola in her bedroom, and have her badger Pinky about earning more money. This episode might take two minutes of screen time.
The little bags holding the episodes I called sequences, and these I grouped in sets by attaching an owner, Pinky's Sequences, Tony's sequences. Tony is the head of the rival Southside gang. The big idea is that the episodes are not flowcharted together in any set way. There is a routine I called Narraton who was in charge of picking the next episode whenever one ended. I'll leave the how of it's selection mechanism till later.
The nicest thing about this setup was that it made implementing subplots pretty clean. Narraton would choose one of Pinky's episodes, then one of Tony's, and then one from another bag, in an unpredictable rotation. The way that the subplots got played out to the end was the way they were grouped in sequences.
The Pinky sequences contain episodes in the which the player can select actions for Pinky, while other episodes carry narrative freight, reactions, foreshadowing.
A sequence contains episodes all having to do with a medium sized hunk of story. A problem, a move, a solution. For example a sequence in which Pinky tries to get Lola to come on to Tony to set him up for an ambush. This has three phases: Pinky tries to get Lola into the plot, Lola goes to Tony's hangout and gives him the eye, and the ambush. This doesn't imply a straight storyline with these things happening in the most predictable ways. Each phase has a number of episodes grouped in it and some contain real twists.
The last phase, the ambush, for example, has some episodes that are just that. You get a chance to shoot Tony before he shoots you, while Lola gets out of the way. In others, Lola has double crossed you and warned Tony, and you get gunned down without a chance of escape. It seems that going from episode to episode within a sequence could be handled pretty well with hard-branching. If this-condition then choose Lola double-cross episode, else choose straight ambush. Let's take a peek at the episode selection algorithm and I'll explain some of the advantages it brought to the project.
Here is a stylized look at two episodes in the pool of those currently available. The episode selector's whole purpose in life is to pick the one that best fits in with the current state of the game. The game state is represented by a set of game variables such as Lola_happiness, Boss_cash, Boss_turf - how much territory you control, and Heat - how much pressure is on the police to make a move against the gangs. These could vary from 0-100, which gave me more discrimination than I could handle. Each episode has a key attached to it which is looked at by the episode selector, this is a target level for one or more game variables. Episode 1 has Lola_happiness at 15 and Boss_rep at 30, while episode 7 is looking for a GANG_MORALE of 40 and a BOSS_REP of 50.
The episode selector looks at all the keys of available episodes and selects the one that most closely matches current game variables. Its method is to look for a least square fit.
At the top here are the current levels of some game variables at this moment in the game. LOLA_HAPPINESS is currently 20, BOSS_REP 40, and GANG_MORALE 80. The game is just beginning and Pinky is just a gang member with ambition. Lola is angry, the world doesnít think that much of Pinky, and the gang members are really happy with the way the gang is being run, bad for Pinkyís ambition.
The selector subtracts the current variable from the target of the key and squares the difference, adding the results of both operations. For episode 1 it found a close match between Lola's current happiness and the target happiness, a difference of 5 and a square of 25. Lola is pretty upset and the episode features an angry Lola. Pinky's rep his reputation in the gang world, is a bit low at 40, the episode is looking for a 30, for a difference of 10 squared to 100. The match score for episode 1, which is titled "Lola plays Lady MacBeth", is the sum 125.
Episode 7 is called "The butcher with the sharpest knife has the warmest heart." It's looking to match a gang morale of 40 and a Boss-rep of 50. In this episode Pinky overhears a couple of gang members talking about the Old man, the current head of the gang he wants to take over, and he has to decide whether to butt in and bad mouth him to try to win the boys to his side in the power struggle. In the current game state the gang's morale is at 80 so the difference squared is 1600, a relatively close match on the BOSS_REP variable can't overcome this huge difference. The gang is happy with the current leadership with their morale so high, and for the episode to make sense, the gang would have to be pretty unenthusiastic about the Old man as boss.
So were these two the only episodes available at this point in the game the selector would choose episode 1 by a long shot 125 to 1700. Generally there were from 4 to 20 episodes that the episode selector could choose among at any point.
This selection algorithm is obviously a make-do rather than the ultimate answer, but it got some pretty good results. The thing it did best was to exclude episodes way off the mark because the difference squared weighed heavily against them.
Inside an episode I had more traditional if-then kind of branching.
In episode 1, Lola is trying to goad Pinky into bumping off the Old Man, "Sweetie, ya can't wait no longer." and Pinky sends up 2 thought-balloons, "She's getting my goat." and "Maybe she had a hard day." The player chooses one or the other. If the player chooses to play hard guy Pinky sneers and replies," Lay off, will ya?" and Lola_happines variable is decremented by a small amount. If the player wants to play patsy Pinky says, "Take it easy, Lola." and her happiness is incremented. So choices made inside episodes impact the variables which are used to choose future episodes. Make Lola angry enough and she'll double-cross you. Make her too happy and you're spending too much time away from your gang duties and risking a revolt of the troops.
Some of the branches where invisible to the player. Based on the level of a game variable, say BOSS_CASH, Lola might try to wheedle more money out of Pinky, or command him to go make a bundle.
A big advantage of using an episode selector rather than a more explicit type of branching was that it let me improvise more freely. If I got a great idea for a new episode, I could set it up in its own sequence, assign it keys and trust that it would be selected appropriately. I could also come up with a new alternative plot twist for a sequence, write it and throw it in without rewriting any of the others episodes in the sequence.
It also made collaboration on the script a lot easier. I worked with a playwright who had never worked in games before and he was able to write whole sequences of episodes which I put in script data form, assign variable keys. He loved working on a work of interactive fiction. He said, "When you work on a play, you have to cut out so much good stuff. With this, all your good ideas get thrown in." If only it did work that way. In fact I wound up cutting plenty of his spaghetti plotting when he got to much into hard branching between episodes. Pick-a-path plotting bores me to tears.
One disadvantage was that I could never try out all the combinations even within a single sequence. I'm sure there are some episodes that have never been played because I assigned some very unlikely keys to them.
Another disadvantage was that I had to make sure sequences didn't have any undesirable side effects. You couldn't be moving through a subplot about Lola and Tony plotting Pinky's betrayal while in another subplot Tony sends a hitman after Pinky.
There are about 270 episodes in the King which I estimate would play for about 8 hours if laid end to end. A game might take anywhere from 5 minutes to an hour and a half. It was very replayable. people played it over and over trying to figure out the paths through the narrative. They didnít realize that I was rerouting most of the paths each time they played.
One of the big decisions in designing an interactive story is imposing constraints. This is one big feature that separates our present IF works from the interactively generated stories of the future. To keep the possible plotlines from growing out of control, we must constrain the player. A lot of the craft in our art lies in choosing the constraints so that the player doesn't feel straitjacketed.
One of the common constraints is geographic, limiting what can be explored. There was no exploration in the King so I imposed constraints in a number of other ways.
Most obviously I constrained the player by limiting choices to those that appeared in Pinky's thought balloons. Not an original invention, which fact I counted on to help the player accept it. It gave some value to the player in its ease of use, and helped pace the game because if the player didn't make a choice within a few seconds, Pinky made one for him. The value of that constraint for me as the designer is obvious and fundamental. I didn't have to plot and write script for the zillion off-the-wall or just unanticipated directions the player might take the story. It collapsed the script writing problem from mathematically impossible to a question of do I have enough time to write enough episodes to make the story-space feel like it's got some breadth.
Secondly, I constrained the over-all length of a game by making the gang impatient if the player takes too long to win. Gang morale goes down every month a piece of turf isn't won and eventually when it hits a certain low level the episode selector is cued to reach into a special sequence. Angry gang members confront the boss and kill him if he doesn't get some quick success. So the game didnít play endlessly until it ran out of closely fitting episodes and started grabbing inappropriate ones.
Wrapping up the King I've got to tell a story on myself that also says something about how I make story games. I like the tale because it makes me seem modest and yet it's really a bit of a brag. I do read reviews of my work and I take them to heart, especially when they're good reviews. A few months after the King came out on the Mac the MacWorld review appeared and I nearly died of an ego feeding frenzy when I read it. The words you would kill to see in a review of your latest game "I've never had so much fun playing a computer game." It really said that. Well, I thought, this reviewer has an extraordinarily refined critical faculty, I better give him a call and say hi. He'll probably get a big kick out of talking to me. Steven Levy is his name, so I tracked him down and called him up. Hi, I'm Doug Sharp and I wrote King of Chicago. The first thing he says is, wait, you know MacWorld misquoted me. I didn't really write that thing about "I've never had so much fun playing a computer game". My stomach sank. I felt like a jerk. Then he said "What I wrote was 'I've never had so much fun reading a computer game.'" I sorta had the wind taken out of me. I ended our chat a bit quicker than I had planned. It took me a while to realize that that wasn't such a bad thing for him to have said, given my approach to storymaking. Especially as he never asked MacWorld to print a retraction.
So my one work of IF to date is squarely in the story camp with constraints applied to the playerís gaming freedom. In my next work Iím going farther into the game camp, which is one reason I wanted to work with Kellyn.
When exhausted from a bout of coding, debugging, or playing and replaying until all pleasure is ground away, it is easy to question the value of what we are trying to do. The value certainly isn't just in the vast sums of money we all earn for very little work. My little two year old daughter Margaret is a good reminder to me of just how vital stories are to the human mind. She devours fiction, it nourishes her emotions and stretches her mind. She would be a duller and poorer girl without the gifts of Peter Rabbit, ET, and Babar. And now, best of all, she is making stories daily and asking me to share them, to listen or to jointly create. And at night, she dreams, mysterious fictions that we all must make or go mad.
We all have different ideas about how to make interactive stories on computers. Some like text and parsers. Some puzzles, some graphics. Some look to movies as inspiration, some to the novel, some to D&D games. Our interests diverge and yet we all want to use the computer to grab an audience, to create a solid world populated with characters that live, out of shadows on a glass screen.
Thanks for letting me talk about a subject that means a lot to me.
There are limits to how much technical discussion can contribute to the advancement of Interactive Fiction, for above all, it is an art form. Individual artists and small creative teams burning to share a good story with an audience in a new way will carve out the frontiers of what is possible. Masterpieces will emerge and inspire. We work hard to do better than we have done. I look forward to experiencing the works of Interactive Fiction to come out of the people in this room.
Thanks for letting me talk about something I care about.
Back to Doug's Home Page.