Download a Free Game - Instantly!

Imagination Serviced and Sold Here.

A definition, as always, is hard to come by. All fiction is interactive, in that each reader brings a different perspective to the story. Interactive fiction (IF) is fiction where the experiences of diffcrellt readers are objectively, measurably different. Usually the reader can influence the outcome of the story. The degree of interactivity in IF ranges from movies where the audience votes on one of two endings to live role-playing games where the participants are given characters to play and placed in a situation of conflict, and each try to steer the outcome to their advantage. l'm going to focus on forms of IF which are enhanced or made possible by computers.

History:
Hypertext fiction

Hypertext is text with links. Links take you from one text to another. Sometimes there is a default linear path which the reader can follow through the narrative, and the links are optional.

For instance, say you were reading the hypertext version of Hamlet on an Apple Macintosh. After reading Act II, you might be prompted, 'Should Hamlet (A) kill his uncle, (B) leave the country, or (C) mope about life and death?' You type 'A', and read a considerably shortened version of Hamlet (This exhibits one problem with interactive fiction - sometimes the action which builds up to more dramatic climax is not the action which a goal-oriented reader would take.)

It is possible to do this on paper by letting the reader decide at each crisis what the protagonist would do next, and telling them a page to turn to depending on her decision. This is like the programmed learning textbooks from the 1960s, e.g. Schagrin, 1968. Now there are many juvenile novels written this way (Brust, 1987).

Jorge Luis Borges described such a book (though he did not write one) in 'El jardin de senderos que se bifurca' ('The garden of forking paths') in 1941 (Fishburn, 1990):

In all fiction, when a man is faced with alternatives he chooses one at the expense of the others. In the almost unfathomable Ts'ui Pen, he chooses - simultaneosly - all of them... Fang, let us say, has a secret. A stranger knocks at his door. Fang makes up his mind to kill him. Naturally there are varios possible outcomes. Fang can kill the intruder, the intruder can kill Fang, both can be saved, both can die and so on and so on. In Ts'ui Pen's work, all the possible solutions occur, each one being the point of departrre for other bifurcations. Sometimes the pathways for this labyrinth converge. For example, you come to this house: but in some possible pasts you are my enemy: in others my friend. (Borges, 1944)

In the same year Borges described a backwards hypertext fiction, the likes of which has never been written, in 'An examination of the work of Herbert Quain' (Borges, 1944). Herbert Quain's supposed book April March was a backwards-branching hypertext. The first chapter described the events of an evening. The next theee chapters describe three alternate prececling evenings. The next nine chapters describe nine alternate evenings before those in the second through fourth chapters with three possible preludes to each of those three chapters. There never was any such book; Borges often pretended to review an imaginary book in order to explain the principles he had in mind for a book without actually writing it.

Julio Cortazar wrote the novel Rayuela (Hopscotch) in 1963, which is a simple non-interactive type of hypertext. He provides two ways of reading it: With or without a set of optional chapters between the required chapters (Cortazar, 1966). To my lnowledge, the only interactive fiction written on paper before it had been demonstrated on a computer was 'Norman vs America', a 20-frame cartoon by Charles Platt based on an idea by John Sladek, published in an underground comic in 1971 (Platt, 1971).

Interactive drama had been experimented with; two early examples were the first British science fiction TV show Stranger from Space 1951), and a movie shown in the Czechoslovak pavilion at Expo 67 in Montreal. The next week's installment of the TV show was based on suggestions in viewer mail (Ford, 1993). Viewers of the Czechoslovakian movie voted on the spot to choose between possible, previously filmed continuations (Elmer De Witt, 1983).

A computer is useful for hypertext fiction because a reader wants to move through the story without filling his book with bookmarks of points to return to and without constantly searching for the next part of text.

Another type of link does not alter the course of the plot, but is a digression. When you read, 'When he himselfe might his Quietus make with a bare Bodkin', you might click on the word 'Bodkin', and see a window come up that says, 'Bodkin: A short pointed weapon; a dagger; poniard, stiletto, lancet.' A hypertext annotation of James Joyce's Ulysses is being assembled at the State University of New York at Buffalo, which should make that book more readable.

The most straightfoward type of hypertext novel would be a plot tree through which the reader chooses one path which takes them along a traditional narrative, or a non-branching narrative from which they may take minor digressions.

The people who write hypertext fiction using computers today generally want to be very cutting-edge, and to use this new medium to communicate a fundamentally new reading experience. The computer hypertext stories that have heen written, such as 'Afternoon', try to replace the straightforward following of a narrative with a stochastic sampling of the story that leads you through a maze of links until you (hopefillly) finally have a feel for the entire set of interrelated people and events that populate this piece of fiction¹ (Coover, 1993). Glen Hartley has proposed that the 'ultimate participatory novel' may resemble the 'Tralfamadorian novel' in Kurt Vonnegut's Slaughterhouse Five.

Each clump of symbols is a brief, urgent message - describing a situation, a scene. We Tralfamadorians read them all at once, not one after the other. There isn't any particular relationship between all the messages, except that the author has chosen them carefully so that, when seen all at once, they produce an image of life that is beautiful and surprising and deep. (Kurt Vonnegut, cited in Hartley, 1985)

My reaction to these types of works is that interactivity is actually very low. They are more like the computer game Portal Activision than like IF: rather than affecting the story, the readers merely search through the hypertext until he understands what's going on.

I believe that before trying to create entirely new means of communicating fiction, we should extend traditional narratives with hypertext, especially since that is the only practical way to interest most people in hypertext.

Computer adventures

Suppose that, instead of giving the reader two or three choices at every branch point, you give them hundreds. And suppose that branch points came not every page, but every sentence. The resulting hypertext would be too large to list in a tree fashion. Instead, the effects of each choice must be computable. This means that the fictional world must have a representatin which can be altered in detail and in ways not foreseen by the author. Furthermore, the list of possible choices is too large to present as a menu; it must be presented implicitly; for instance, by allowing choices to be specified using a subset of English. The resulting hypertext is an adventure

You are standing at the end of a road before a small brick building. Around you is a forest. A small stream flows out of the building and down a gully.

That is the first line that greets you upon running Adventure, which was finished in early 1977 by Willie Crowther and Don Woods. The first version, in 1975, was simply a map of Mammoth Cave, Kentucky, which let the player walk from room to room. Commands were added to pick up, carry, and use items in various ways. The player's goal was to find treasure. Various problems presented themselves, ranging from the obvious (a fierce green dragon bars the way) to the subtle (a gold nugget is too heavy to carry up the stairs to the treasure room). Objects or information that could be used to overcome these obstacles were also waiting to be found. Each treasure gained or problem solved added to the player's score.

Since Adventure was written in FORTRAN, which everyone had, it spread rapidly over the Arpanet. It may have set the entire computer industry back two weeks: when it reached a site, work was suspended until everyone had solved it (Anderson & Galley, 1985).

The way this world was consructed has remained the same in all adventures: the world consists of thins contained in other things. For instance, at the start, you are contained in a location described in the above quote. If you enter the building, you will find a lantern in the building. Pick it up, and it is in your. The world is discrete, not allowing you to be 'in transit' between locations, nr (generally) for an item to be in two locations at the same time, even if it should be (e.g. a rope). Each command you issue takes one unit of time; events between moves occur all at once rather than continuosly. Your commands are issued by typing a sentence (in Adventure's case. a verb and a noun) at the start of each turn.

>DOWN STEPS
You are at one end of a vast hall stretching forward out of sight to the west. There are openings to either side. Nearby, a wide stone staircase leads downward. The hall is filled with wisps of white mist swaying to and fro almost as if alive. A cold wind blows up the staircase. There is a passage at the top of a dome behind you.
>NORTH
You are in the hall of the mountain king, with passages off in all directions.
A huge green fierce snake bars the way!
>RELEASE BIRD
The little bird attacks the green snake. and in an astounding flurry drives the snake away.

Unlike almost all traditional fiction, adventures use second-person present. This is because they are immersive: the player projects their self into the role of the protagonist with an immediacy not possible in static fiction. Years later, Brian Moriarty designed Trinity so that the player had to kill a lizard. In an interview, he said,

I was amazed to see how many people were actually bothered by the scene with the lizard, because it was them doing it. It's nice to know that interactive fiction could do that, make you feel uncomfortable about killing things. In no other media could I make you feel bad about killing something. Because there is only one medium where I can make you do it, and make you feel empathy for a thing that doesn't exist. It s only with interactive fiction that you can explore these emotions. (Rigby. 1991)

After playing Adventure, many people wanted to write their own. In a few months Tim Anderson, Marc Blank, Bruce Daniels, and Dave Lebling of the Dynamic Modelling Group in the MUT Artificial lntelligence Lab created Zork (Anderson & Galley, 1985), which was famous for its sense of humour, its anticipation of actions the player might try, the cleverness of its puzzles, and (eventually) the complexity of its parser. Zork was the first adventure which could parse complete imperative sentences, plus a few questions. Zork was also the first adventure whose non-player characters had personality. The thief was a gentleman gone wrong, with good manners, a cynical sense of humour and the willingness to slit your throat in a moment.

Zork like the Apple Computer, got its name because no-one came up with another. 'Zork' was a nonsense word; the Dynamic Modelling Group usually called its programs 'zork' until they were ready to install. Since Zork never was officially installed, it was never named (Anderson & Galley, 1985).

Zork was written not in FORTRAN, but in MUDDLE, a LISP variant which was not very widespread. Zork gained fame because, although people couldn't distrib-lte it widely; anyone could log onto the MIT Artificial Intelligence Lab computers and run it. They ran an MIT-grown OS called ITS (Incompatible Time-sharing System), which had no security (Anderson & Galley, 1985) (In fact, ITS had a command KILL SYSTEM which would do just that, so that it wouldn't pose a challenge to crackers (Levy, 1984). The Implementors (as they were known) later formed Infocom (below) which gained fame as the best (some would say the only) publisher of interactive fiction.

But it wasn't Infocom that first brought adventures to the masses. Zork, like Adventure, originally ran only on mainframes, since it took a megabyte of RAM (Anderson & Galley, 1985) (Adventure took 300K (Adams, 1980)). In 1977, few believed that a personal computer (which then had 16K RAM if you were lucky) had enough memory for an adventure. A systems programmer named Scott Adams thought they did. He worked so hard to prove this that his wife, feeling neglected, once put his floppies full of source code in the oven (Adams, 1985). But after a year, he amazed the personal computer world with an adventure interpreter which allowed adventures (though not nearly as large as Adventure or Zork) to run in l6K. They were distributed on cassette tape because they didn't leave enough RAM for a disk operating system to load them in. Adams quickly wrote twelve text adventures which became the standard literature. The simple verb-noun parser he borrowed from Adventure is still often referred to as an 'Adams parser', since he was the last major author to get away with using it.

Scott Adams touched on all the genres. He published Tolkienesque, pirate, mystery, gothic horror, spy, science fiction and western adventures. To this day, adventures generally remain genre vehicles.

In 1979, various people from the MIT Dynamic Modelling Group formed Infocom. They cut Zork in half, squeezed the first half onto one 140K floppy disk for personal computers, and called it Zork I. They licensed it to Personal Software Inc. PSI sold about lO,OOO copies, then gave up on Zork I since that was all they expected from a game. Infocom regained the rights to market Zork I themselves, and by 1986 had sold 250,000 copies (Anderson & Galley, 1985; Gerrard, 1987). Byte magazine said:

That the program is entertaining, eloaquent, witty, and precisely written is almost beside the point. Unlike the kingdoms of the Adventures for machines with 16K bytes of memory and far from the classic counter-earthiness of the Colossal Cave in the original Adventure, Zork can be felt and touched - experienced, if you will - through the care and attention to delail the authors have rendered... [A] most excellent and memorable work of computerized fiction (Liddil, 1981)

Infocom's games kept getting better. They updated their game engine to simulate the world more realistically, and to parse more sentences. After Zork, their games always followed a central narrative in a uniified setting. They learned to avoid the cardinal sins of mentioning something in a description but not understanding references to it; requiring a specific sentence to accomplish something so that the game degenerated to synonym guessing ('Move curtain? Examine curtain? Open curtains?'); presenting problems that were simply exercises in combinatorial search; or selling buggy software.

In 1983 they released Deadline, a murder mystery which advanced the state of the art in several ways. Characters in the story played their parts out, moving throughout the mansion and its grounds on their own business. But if you interfered with them, you could thwart their plans. Simply following someone around could cause them to do something more innocent than what they had in mind. Deadline was also the first adventure to use the plot tree form of hypertext; it had about 30 possible endings (Hartley, 1985).

Graphic adventures

Roberta Williams, co-founder of On-Line Systems (now Sierra Software), was hooked on Adventure, and wrote her own adventure which took place in a Victorian mansion with a killer on the loose. Her husband and co-founder Ken told her she needed a new angle to sell it, and she thought it would be great if a game had pictures as well as words.

The result, Mystery House, was released in 1980. It had a picture for every location. Despite the fact that its pictures were monochrome line drawings with stick figures for people, it was an instant success (Levy, 1984).

From then on, the trend was towards graphics adventures. Even Scott Adams rereleased all his adventures with accompanving pictures, and found he could charge twice as much as he did for the text-only versions. Fans of text-only adventures complained about the smaller scenarios, concentration on graphics to the exclusion of other issues such as plot and ease of use, and the limitation of the imagination. But pictures sold programs. Infocom included some in Zork Zero shortly before the company was dissolved in 1989.

Publishers used Infocom's failure as proof that text adventlures were dead (Goetz, 1987). But Infocom's failure was not because their text adventures weren't selling, but because their relational database Cornerstone, an expensive long-term development project, didn't sell (Forbes, 1993).

Now commercial software publishers deal only with graphic adventures, claiming they have more mass appeal (Goetz, 1987). But the large group of amateurs who write and swap their own adventures write only text adventures².

Issues in Interactive Fiction:
Freedom vs drama

A fundamental problem with interactive fiction concerns open-ended interactivity vs drama. A dramatic story is one crafted by a writer to be so. Though there may be 30 possible endings to an interactive fiction, that is still a finite number. The player does not have true freedom. Yet if you let the player wander outside the storyline, the author cannot provide a dramatic experience.

Jim Gasperini, author of both a text adventure (Star Trek: The Promethean Prophecy) and a simulation (Hidden Agenda, a narrative simulation of Central American politics), contrasts the closed-endedness of an adventure to the open-endedness of a simulation. His comments on adventures apply equally to IF if we substitute 'plot has been played out' for 'puzzles have been solved':

Even in the best 'interactive fiction', once all the puzzles have been solved the plot is revealed in all its naked linearity. A finished 'closed-ended' work is like a punctured balloon, emptied of all ambiguity. There is little reason for anyone to go through it again.

By contrast, an 'open-ended' work becomes more ambiguous, not less, the more it is played. It is through repeated playings, comparing different plots chosen through the same web of potential plots, that the experience becomes most meaningful. This can be most clearly seen in the genre known as 'simulations'...

Each subsequent time the player enters the election campaign comparisons naturally arise between what happens this time and what happened other times. This serves to deepen the player's awareness of the range of structural possibilities. (Gasperini, 1990)

How can we experience open-ended interactivity that isn't boring? If we set up the characters in the story and let them work out their problems around the player, they are likely to find sudden and unsatisfying resolutions (A con man comes to Dave's town to trick old ladies out of their retirement funds. Dave threatens to expose him. The con man shrugs his shoulders and moves on to the next town.) James Meehan wrote a program called TALESPIN in 1976 that generated simple stories based on rules about how characters interact. Sometimes they turned out like this:

Betty was famished. Betty wanted to get some berries. Betty wanted to get near the cranberries. Betty walked from her cave down a pass through the valley across a meadow to the bush. Betty took the cranberries. Betty ate the cranberries. The cranberries were gone. Betty was not hungry. The end. (Meehan, 1980)

Stories rely on conflict. Conflict is implicit in a simulation of a battle, a dogfight or an economic system (survival vs collapse). When the conflict is resolved, the simulation is over. But a novel is constructed by suststaining a major conflict, continually introducing new complications that prevent the protagonist from resolving the situation. The gradual escalation of conflict we find dramatic is unnatural, a failure on the parts of both protagonist and antagonist. It needs artifice to maintain it. Furthermore, the consequences of the resolution must be commensurate to the magnitude of the conflict. In real life, the war may be lost for want of a nail, but in IF, the protagonist had better have to work harder than to provide someone with a nail

Brenda Laurel, in her 1986 dissertation, proposed the development of a computational theory of drama, possibly based on Aristotle's theory of dramatic structure, which would be a sort of grammar for drama (Laurel, 1986). This would allow a computer to construct a dramatic turn of events whatever the participant does.

Writers often complain that 'everything has been written before', meaning that there is a small number of basic plots. Georges Polti claimed in 1921 that there are 36 dramatic situations (Polti, 1921) and others have tried to fins similar 'basic plots'. Joseph Bates of CMU, David Graves of Hewlett-Packard and Jurgen Appelo all advocate compiling a library of standard plot segments and writing a computer composer capable of combining them in sensible ways (Bates, 1990; Graves, 1993; Appelo, 1993). This calls to mind Mozart's dice minuets, in which before performing you would roll dice to choose which phrases to play when; or, on a more mundane level, Mr Potato-Head. Which of these two the products of an automated playwright would most resemble remains to be seen.

An automated playwlight would have an enumeration of plots, match the current state of events and past history to one of them, and be responsible for the other parts in the current plot besides the protagonist.

Good fiction takes creativity on the part of the author. No artificial intelligence (AI) program in the next twenty years is likely to be able to choose good descriptive details, or to provide humour, pathos or provocative ideas. Writing is 'Al-complete' (Shapiro, 1992): we'd have to solve all the problems in AI before writing a computer author. The systems these people advocate don't need full intelligence because they will be hack writers, at best able to churn out westerns, space opera and romances. Mysteries and sitcoms will remain beyond them. As for me, I will not abandon closed-ended, human-controlled fiction.

Interactive fiction does not equal adventures

Adventure was not really a story, since it suffered from a lack of a plot (other than 'gather treasure') or motivation (magic wands, lanterns, and gold nuggets were just lying around for the taking). The final point needed for a perfect score of 350 was infamous for its arbitrariness: you had to take a certain item among hundreds and drop it off in a certain room among hundreds. Bruce Daniels, one of the authors of Zork, had to disassemble the Adventure object code to discover the solution (Andelson & Galley, 1985).

Infocom preferred to distinguish between adventures, such as Zork, and interactive fiction, such as Deadline. In an adventure, players solve puzzles. Interactive fiction requires plot and characterization (Hartley, 1985). We want to do in IF the things we do in traditional fiction: make readers care about the characters, create suspense and concern, and a feeling of dramatic completion.

Serious researchers are squeamish about the term 'player' because of its connotation of frivolity. Since reading flction is entertainment, and interactive entertainment is a game, the term 'player' is justified. Please understand that this does not imply that all IF will be like adventure games, played to win.

The path not taken

In role-playing games, there are two types of players. Some, who are often found playing Dungeons & Dragons, are very goal-oriented; they will do only that which increases the power of their character. They play to win. Others devised their own games, such as Paranoia or Toon, to emphasize the role-playing aspects. In Paranoia, you have six lives; in Toon, an infinite number. This frees the player to take actions which lead to their characters' deaths if those actions are in character. To these players, playing is winning.

The former class will never be able to appreciate many forms of IF. The development of a satisfactory story depends on both the author and the player. If the player cannot take on another persona, they cannot enter into the world the author has devised and cannot explore the nature of that world. Hamlet would jump straight to the final act, and Kafka's The Trial would turn into 1984, because the player would never take the actions, dictated by the character of the protagonists, which make those stories what they are.

Unless players can find reasons to play other than to win, IF will not escape the literary ghetos of genre fiction. Even some traditional traditional-genre stories would lose their charm under the imposition of a different character; imagine The Hobbit with a self-confident and aggressive Bilbo Baggins, or an interactive Father Brown mystery played by a Humphrey Bogart fan.

In particalar, truly tragic fiction might never work in IF. I'm not referring to 'tragedies' such as Hamlet, which are melely sad. I'm referring to works such as 1984, Brave New World, Lord of the Flies, Heart of Darkness or Deliverance, in which it is dramatically necessary for the main character to be psychically crushed. The IF player might feel that giving them the freedom to choose how to act had been a cruel farce.

One way to keep players from identifying too closely with the protagonist might be to have them interact with several characters. They might change viewpoints, or might simply have a display panel with a point-and click interface controlling the emotional response of each character (level of anger, contentment, fear, urgency, etc.) and see how the story unfolds. But this defeats the intimacy of IF.

Computer Science Problems:
Physical simulation

In linear fiction, the author creates a suspension of disbelief only with great care. In interactive fiction, there are more opportunities to shatter this illusion. The world and the characters in it must respond realistically to the player, even in situations the author has not foreseen. If he player drops a crystal vase, it should shatter, and they should be able to cut a plastic wrapper with the shards. Open-ended stories cannot begin to be developed unless the enture simulated world is complex enough to run on its own without authorial control. Special problems include liquids, fire and transparent items, amd accessibility to sight, sound and touch.

One problem is a result of time passing in discrete steps. If, in time interval I, character X decides to leave the room and character Y decides to shut the door, X may successfully leave the room (if he acts first), or he may run into a shut door.

The different types of entities in the world (people, mountains, candy bar wrappers) require different types of simulation. It may take elaborate calculations to decide how a pile of leaves will blow in the wind (Wejchert and Haumann, 1991); these aerodynamic computations should not be applied to a falling safe. Saying one representation should handle all situations is like saying text, photographs, movies, and music should all be stored with the same representation. A Context mechanism must be found for deciding when a particular level of abstraction is appropriate - see, for example, Guha, 1993.

Methods for limiting computlion will be important. For example, areas outside the current deictic (narrative) centre might be given less processor time, and be simulated at a cruder grain.

Simulated characters

Joseph Bates says we don't need to create intelligent characters, just ones that aren't obviously stupid (Bates, 1991). He calls them shallow but broad agents. They need some knowledge in many areas, to avoid acting unbelievably, e.g. standing in the path of a steamroller. This brings to mind reactive agents as popularized by Rodney Brooks' subsumption architecture (Blooks, 1985).

At the other extreme, we would like characters to reason about 'symbolic' statements. A word in English is a symbol for a concept; hence reasoning about statements like those in English sentences is called 'symbolic reasoning.' The rules that tell how the world works are usually stated procedurally in IF. That is, if the program contolling the IF world wants to know whether a character can unlock a box, it calls a subroutine which returns a 'yes' or 'no'. If instead you stated the rules declaratively, e.g. 'if a box is locked, it cannot be opened', then characters could apply standard Al symbolic planning techniques to form plans on the fly, adjusting them when problems occur. Then, if you asked Jack to fetch a pail of water, he could figure out how to do it.

There are two difficulties. It's difficult to come up with a representation powerful enough to say all the things you want to but simple enough to apply these techniques. It's also not known if the standard techniques will work in a world as complex as an IF world, or if there will be too many things for the computer to 'think about' in a reasonable amoumt of time (Goetz, 1994).

SNePS, a Semantic Network Processing System (Shapiro and the SNePS Research Groop, 1994), has a component called SNeRE, the SNePS Rational Engine (Kumar, 1993), qhich unites planning and acting in one formalism. This lets it integrate reactive behaviour with symbolic reasoning, since a reaction can be expressed as an action taken whenever the agent finds itself in a certain type of situation. This may be useful for creating broad and shallow (reactive) agents with particular deep and narrow (symbolic) capabilities. Doug Lenat proposed the use of Cyc, a vast ('encyclopedic') database of commonsense knowledge being developed by the Microelectronics and Computer Technology Corporation (Lenat and Guha, 1990), to maintain simulated worlds because it has a lot of knowledge about the physical behaviour of objects (Lenat and Guha, 1991). Cyc's social knowledge would also help characters reason. James Meenan's 1976 dissertation, using his program TALESPIN, explored the type of knowledge needed to simulate realistic interaction between characters (Meehan, l980). Joseph Bates of Carnegie Mellon hopes to use Cyc in Oz, the CMU interactive fiction platform, to provride such knowledge, as well as drawing on cognitive structures such as Soar (Bates, 1990; Newell, 1990).

Why IF is interesting to AI

When you have an idea for a representation or technique and write test cases, your imagination is restricted by the techniques you have in mind. Plan forming and natural-language-understanding systems such as STRIPS (Flkes and Nilsson, 1971) and SHRDLU (Winograd, 1972) gave impressive performances only because they dealt with a world consisting of nothing but a table with blocks on it. IF forces you out of the blocks world. You have to bring things in for the story, and you quickly find the weaknesses in your system. Players are much more thorough testers than you can be.

Future Directions:
Virtual reality interactive fiction

Real-time 3D rendering is still beyond the capabilities of personal computers, as is thorough real time 3D physical simulation. Silicon Graphics claims they will provide real-time rendering in late 1995 for around $5000 (as opposed to $100,000 today) (Simerman, 1993). Real-time polygon-based 3D is already available for Personal computers, and some systems, such as Autodesk's Cyberspace Developer's Kit, Sense8's WorldToolKit, and Robert Grant's Multiverse, provide some aspects of physics simulation (friction, gravity, and elasticity) in real-time (Autodesk, 1993; Brill, 1993; Grant, 1993). Knowledge Revolution's Working Model is a detailed real-time 2D graphical simulation of physics on the Macintosh, taking into account velocity, mass, inertia, gravity, collisions, static and kinematic friction, elasticity, electlical charge and torsion, among other things (Schaff, 1993).

As it becomes easier to render good 3D graphics on personal computers and to simulate physics for a 3D world, graphical IF will approach a detailed physical model of the world. We can imagine an interface which is more like a Virtual Reality system than a text adventure. You are looking at a 3D world on your computer screen, with colours, shadows, reflections, surface reflections and textures.

The command parser is used only for speaking with other characters. Your physical interaction with this world is done entirely with an arm. This arm has no joints. It sticks out straight in front of you. You can pull it out or push it in, and control the speed at which this is done. You can make it sticky or un-sticky; you can also rotate it (roll, for you aerospacers).

If you want to pick up an object ftom the floor, you bend over until the object is in the centre of the screen at the end of the arm. Then you extend the arm, change it to sticky, and retract it. You can 'drop' items from your arm into your inventot,; and pop them back onto the arm from your inventory list. Say you want to throw a ball. You pick it up as described, then rapidly extend the arm. At the end of the arm's movement, the ball flies off the arm.

With the head-mounted display and body motion sensors based in W Incorporated's 1992 game Dactyl Nightmare, we can envision this system being fully immersive virtual reality. This type of IF may be produced by Hollywood movie moguls. Real-life actors may have their images copyrighted and licensed to be used to generate actors in IF games. Despite all the interface changes, this type of IF still has the same problems as all-text IF: offering the player freedom while keeping them in the plot, or generating a plot on the fly; creating believable characters, and escaping from genres. (VR IF will probably run in continuous time, eliminating the discrete time-interval problem.)

Textual IF will survive, just as text novels haven't been entirely, replaced by movies. It is a matter of time investment. A graphical presentation takes longer to 'play', just as a two-hour movie can't communicate as much as two hours of reading. It also takes much longer to create. Individual authors simply don't have the time to stop every time they write a scene, and create every object in that scene as a 3D object, as well as the background.

Some aspects of textual IF will change. Time may run continuously, rather than always waiting for the player's next move. A more detailed physical model may lie behind the text (which can only communicate so much detail).

Multi-reader interactive fiction

Many computer games have developed from single-player to multi-player, such as Nettrek and Conquest (space war games), Maze (a tank-war game), and many multi-player dungeons. But having multiple readers in an IF is not as simple as introducing another person into the same scenario. If the two people act independently, how can they both experience a dramatic unfolding of events? How can they both even understand what is happening in the world? In order for a reader to experience drama, what unfolds before them must be important to future events. But if what unfolds before each reader is central to the plot, then each sees only one half of what they need to. If they find each other and exchange information, the plot will not unfold but be thrown on them in disordered chunks.

Writers using the third person limited point of view must be careful to ensure that their central character has an appropriate view of the story; no events should occur without explanation, but when suspense is intended, the outcome should not be known in advance. Can these restrictions be ignored safely, as they are in live-action role-playing games? Perhaps the solution is that each participant experiences a different drama: one may be the protagonist, and the other the antagonist.

For adventures, the task is difficult, but different. Since the primary purpose in an adventure is puzzle-olving, players can interview each other. It doesn't matter in what order information is revealed. Half of the fun might be divinig which players are telling the truth and which are lying. Some might pretend to be computer-controlled, so as not to be feared as competitors. A different problem for adventures is that in traditional puzzles, particular items are often necessary. If there is only one key to the attic, and one player keeps it, what can the rest do? Perhaps the scenario can be designed so that each participant has a different set of tasks.

What if one player plays for hours on end, and the other only an hour a day? Can a narrative be such that one can drop in and out of it and still enjoy it? In a narrative with competing players, perhaps the players can take turns, with large sections of narrative between each turn - a play-by-mail (or email) format.

Conclusion

It is inevitable that future IF will have more real-world knowledge and more realistic interfaces. It is not clear whether authors and players co-operating can communicate the same range of emotions and thoughts to the players as in traditional fiction, whether a theory of drama would enable the player to have an exploratory literary experience rather than a controlled one, or if IF will escape from genres.

Phil Goetz is a doctoral student of computer science at the University at Buffalo, specializing in artificial intelligence. He has played and written role-plying and adventure games for most of his life. His Internet address is goetz@cs.buffalo.edu.

Notes

¹ Based on a demonstration of 'Afternoon' by Michael Joyce at the University of Buffalo in 1992.

² Based on the contents of the interactive fiction archive held on computer at ftp.gmd.de:/if-archive

Bibliography

Adams, Scott (1980). Pirate's adventure. Byte Dec 1980. p. 192-212.
Anderson, Tim, and Galley, Stu (1085). The history of Zork. The New Zork Times vol. 4 nos. 1-3.
Appelo. Jurgen (1993) . Posts to rec.arts.int.fiction.
Autodesk (1993). Press release for Cyberspace Developer's Kit
Bates. Joseph (1990). Computational drama in Oz. In Working Notes of the AAAI-90 Workshop on Interactive Fiction and Synthetic Realities, Boston, July 1990.
Bates, Joseph (1991). Broad agents. In SIGART Bulletin, vol. 9. no. 4. Aug. 1991. p. 38-40.
Borges. Jorge Luis (1944). Ficciones. NY, NY: Grove Press, 1962.
Brill. Louis (1993). Kicking the tires of VR software. Computer Graphics World Volume 16 No. 6. Jun 1993: 40-53.
Brooks, Rodney (1985). A robust layered control system for a mobile robot, Technical report 864, MIT AI Labs, MIT.
Brust, Steven (1987). Dzurlord, NY: Tom Doherty.
Coover, Robert (1993). Hyperfiction: Novels For The Computer. New York Times Book Review, Aug 29 1993, p. 1.
Cortazar, Julio (1966). Hopscotch. New York: Pantheon Books. Translated from the Spanish by Gregory Rabassa.
Elmer-De Witt, Philip (1983). Computers: Putting Fiction on a Floppy. TIME. Dec 5 1983, p. 76.
Fikes. R. E. and Nilsson. N. J. (1971). STRIPS: A new approach to the application of theorem proving to artificial intelligence. Artificial Intelligence 1 (2).
Fishburn, Evelyn, and Hughes, Psiche (1990). A Dictionary of Borges. London: Duckworth and Co.
Forbes, Scott (1993). Rec.games.int-fiction FAQ. Usenet. rec.games.int-fiction.
Ford, Dave (1993). Personal commumication.
Gasperini, Jim (1990). An art form for the interactive age. Art Com (an online magazine). Dec 1990. Vol. 10 Number 10.
Gerrard, Mike (1987). Interview at the end of the universe. Atari ST User. May 1987.
Goetz, Phil (1987). Personal correspondence with Activision, Broderbund, Datamost, Datasoft, Electronic Arts, Firebird, Origin, and other computer game publishers.
Goetz. Phil (1994). Notes on using SNePS for interactive fiction. Usenet. rec.arts.int-fiction. Feb. 1994.
Grant, Robert (1993). Article 7581 (SUNY distribution) of sci.virtual-worlds. 24 Mar 1993.
Graves, David (1993). Rec.arts.int-ficdon FAQ.
Guha, R. (1993). Contexts. Draft, MCC. Austin, TX.
Hartley, Glen (1985). Toward the ultimate participatory novel. EPB vol. 3 no. 4. Jun 1985. p. 12-15.
Kumar, Deepak (1993). From Beliefs and Goals to Intentions and Actions: An Amalgamated Model of Inference and Acting. Technical report 94-04. Department of Computer Science. State University of New York at Buffalo.
Laurel, Brenda (1996). Towards the design of a computer based interactive fantasy. system. PhD dissertation. Ohio State University.
Lenat. D. B., and Guha, R. V. (1991). Ideas for applying Cyc. MCC Technical Report No. ACI-CYC-407-91.
Lenat, D. B., and Guha. R V. (1990). Building Large Knowledge-Based Systems: Representation and inference in the Cyc project. Reading, Mass,: Addison-Wedley.
Levy, Stephen (1984). Hackers: Heroes of the computer revolution. Garden City, NY: Anchor Press/Doubleday.
Liddil, Bob (1981). Zork, the great underground empire. BYTE Feb 1981. p. 262-264.
Meehan, James (1980). The Metanovel: Writing stories by computer. New York and London: Garland Publishing.
Newell, Allen (1990). Unified Theories of Cognition. Cambridge, Mass.: Harvard University Press.
Platt, Charles (1971). Norman vs. America. Reprinted in Breakthrough Fictioneers, ed. Richard Kostelanetz, Berlin: Something Else Press. 1973. p 85-90.
Polti, Georges (1921). The Thirty-Six Dramatic Situations. Franklin, Ohio: J. K. Reeve.
Rigby, Paul (1991). From here to Trinity ... and back again. Adventure Probe vol. 5 no. 5. May 1991. p. 15-19.
Schaff, Robert (1993). Physics for the rest of us. Computer Graphics World. Sep 1993, p. 76-77.
Schagrin, Morton (1968). The Language of Logi4c. New York: Random House.
Shapiro, Stuart (1992). Artificial intelligence. In Encyclopedia of Artificial Intelligence, 2nd ed., editor-in-chief Stuart Shapiro. New York: Wiley, p. 54-57.
Shapiro, Stuart, and the SNePS Research Group (1994). SNePS 2.1 User's Manual. Dept. of Computer Science. State University of New York at Buffalo, Last revised March 29 1994.
Simerman, Tony (1993). Evaluating the impact of the reality engine. Computer Graphics World Aug 1993. p 17-19.
Wejchert, J., and Haumann, D. (1991). Animation aerodynamics. Computer Graphics Jul 1991. p.19.
Winograd, Terry (1972). Understanding Natural Language, New York: Academic Press.