June 9 - 10, 2001
Phrontisterion 3 was undoubtedly the best so far. Phrontisterion 1, with 28 attendees, was too big, and Phrontisterion 2, with 8 attendees, was too small, but this conference, with 14 attendees, was just right. The group was big enough to provide a useful diversity of opinions but small enough that those opinions could be fully explored.
Phront3 was also more focussed because we used this report as the organizing structure for the conference. I presented the attendees with a list of 14 questions. We discussed each question, with each person taking notes. I gathered up all the reports and digested them into this one report. Thus, we had 14 speakers, 14 reporters, and one editor.
The other organizing construct of Phront3 was the hypothesis that there are two fundamentally different approaches to the creation of interactive storytelling: the evolutionary and the revolutionary. Our purpose in Phront3 was to compare and contrast these two approaches. The hypothesis was not automatically assumed to be correct; it was simply a working statement that generated a great many interesting questions. Late in the conference we addressed the question of just how serious the gap was between evolutionists and revolutionaries.
Question 1: What do each of the two sides mean by ‘storytelling’?
The discussion raised all sorts of secondary issues, but on the main topic, we first agreed that the evolutionists consist of two very different groups: game designers and frustrated old media storytellers. The former see storytelling as a means to add depth to their games; the latter treat interactivity as a feature that can be tacked onto traditional storytelling. The revolutionaries, on the other hand, are harder to characterize, but the general thrust of their beliefs is that interactive storytelling must not be pursued as an extension of some other medium, but rather a completely new medium in its own right.
Game-evolutionists see story as emergent from gameplay. They argue that, technically speaking, the playing of any game can be described as a story; why not simply improve the quality of the emergent stories we already have? They think in terms of an open-ended simulation with player actions somehow generating an emergent story. Current products allow players to move around, to jump, to shoot a gun, or some other action-oriented decision; we need only add commands that permit more dramatically interesting behaviors.
The revolutionaries dismiss these strategies as feckless. They charge that the evolutionists seek not story but dramatic context: a means of dramatically justifying the action-intensive gameplay. The evolutionists’ story is only a means of explaining why the player is mowing down hundreds of monsters. Sometimes this takes the form of backstory: the player learns more and more about the game’s universe before and during play. The revolutionaries claim that "story-enhanced games" remain action-oriented games, and dramatic development can never proceed as long as it must play second fiddle to the primary demands for action and excitement.
All agreed that a core task must be the creation of algorithms that capture "dramatic logic" -- the laws of drama -- in the same fashion that many physical simulations use the laws of physics. What those algorithms might be remains beyond our grasp as yet.
The data structures for games and storytelling are profoundly different. Almost all games have maps as internal data structures; almost no stories have functioning maps behind their architecture.
Evolutionists see storytelling as the delivery of a product to a user, while revolutionaries think in terms of a collaborative creative effort between the artist and the audience.
Evolutionists would use the environment to advance the story; revolutionaries would use the character(s).
Question 2: What existing products best represent the two sides’ ideals?
Here the group had an easier time agreeing. The products listed as good examples of the ideal evolutionists storytelling game were Half-Life, Myst, Starship Titanic, and many MUDs. The list of products that look good to revolutionaries was longer: Freak Show, Bad Day on the Midway, The Sims, Siboot, the Erasmatron, the Dramatron, and many MUDs. Several were of the opinion that paper role-playing games should also be counted in the revolutionary list, but most rejected this on the grounds that such games don’t involve computers and as such have little utility to our deliberations. The group also came up with a graph showing how many of these products line up:
Question 3: What would be the characteristics of a product that best realizes the dreams of each party?
At this point the group realized that it had to come to terms with the issue of ’emergent story’. This is the notion that a good story will automatically emerge from the playing of a program endowed with sufficiently robust dramatic algorithms. To clarify our thinking, we postulated four basic approaches to storytelling:
1. The conventional fixed story. Zero interactivity.
2. The branching story. A story with specified branchpoints generating a few hundred different paths. This scheme had one proponent, but was otherwise roundly dismissed as long since discredited by hard experience.
3. The immersive world. A grand simulation of all things, as in the Star Trek Holodeck, in which a story could be emergent by means of the interpretation of the user. In other words, the simulation knows nothing about dramatic laws, but provides enough interesting behavior that the user interprets the results to constitute a story. Some argued that this approach cannot reliably yield entertaining stories, bereft as it is of any knowledge of constitutes a story.
4. The dramatic engine. This is a program specifically designed with algorithms reflecting the "laws of drama".
The group decided that stronger dramatic engines would mean weaker emergent storytelling -- the story would be more explicitly designed by the artist. Drama engines can insure desired plot movement by either carrots (additive approaches) or sticks (subtractive approaches). Additive approaches insert new behaviors that stimulate dramatic development. Subtractive approaches preclude behaviors that inhibit dramatic development. For example, a carrot would be a message to the player saying, "Would you like to meet beautiful Miss Mary now?"; a stick might be the steady reduction of the player’s options until the only option left to the user is the command "I will meet with Miss Mary now." Such subtractive approaches inhibit the development of emergence. In multiplayer games, the more players there are, the greater the role that must be played by emergence.
In any case, several of the attendees felt strongly that the very concept of emergent story is ridiculous, and the opposition to their position was half-hearted.
Walt Freitag: Is ridiculous too strong a word? I thought others shared my opinion that emergent story does occur, it just doesn’t occur reliably from the point of view of an individual user. An intelligent editor selecting from a large body of emergent chronologies is as valid a way to generate good stories as is an author crafting them from scratch (though that approach does not appear to lend itself to interactive storytelling systems).
Question 4: What methods or techniques are the two sides considering?
The group opted to confine this question to matters of product distribution -- for which there was plenty of controversy! The evolutionists’ strategies were fairly obvious: they are quite successful with physical distribution of boxes containing CDs. Indeed, this distribution system is now one of the major barriers to entry for new competitors -- hence the unwillingness of currently successful publishers to abandon that model. Unfortunately, online delivery of software threatens this cozy situation, and places publishers in a difficult position with their distributors if they (the publishers) pursue online delivery options. Therefore, the publishers are likely to follow the market into online distribution, not lead it.
For revolutionaries, the situation is more complicated. We listed seven different strategies a revolutionary might use:
1. Shareware. Sell rates are typically only 1% for entertainment software, so this doesn’t look like a good strategy.
2. Advertising. Yeah, right.
3. Subscription-based. This approach has already been tried by several hot startups, all of which are now cold corpses.
4. Pay per turn or pay per game. This model might have promise, but has not yet been properly tested.
5. Online play of a skeletal version of the game is free; once they’re hooked by the gameplay, you sell them a CD with data assets to enhance the playing experience. No solid data on this method is available.
6. Give it away using the open source model. Don’t even try making money, do it for love.
7. Ask for donations from your players.
The group concluded that we must rely on shareware models until the market for interactive storytelling is developed, at which point we can move to subscription-based models. We strongly agreed that any distribution system must be viral in nature, motivating satisfied customers to bring in new customers, but we were unable to agree on the characteristics of such a system.
Walt Freitag: I don’t recall the group being quite so pessimistic about the profitability of shareware, despite the low sell rate. Also, the potential of asking for donations (list item 7) was discussed in a rather positive light, with several variations enumerated. It was speculated that this (as well as shareware fees) is likely to be more successful when PayPal or equivalent mechanisms for convenient risk-free small online payments become more familiar and pervasive.
(A possible variation I didn’t get a chance to suggest at the conference: instead of asking for a voluntary "fee" before beginning the next chapter, ask for a "donation" after the end of the previous one. Same process, different psychology. This idea comes from street performers, who seem to know what they’re doing. Call it "pass-the-hat-ware." Requires a very large and well-established PayPal installed base to work well.)
Question 11: Referring to Crawford’s hoary anecdote, can we successfully turn Grandpa into algorithms?
The group decided to skip ahead and tackle the questions in a different order; this was the next question answered. "Crawford’s hoary anecdote" is a simple example of genuine interactive storytelling in the form of Grandpa telling a story to little Annie, who frequently interrupts him to make her desires known; Grandpa responds by integrating Annie’s desires into the developing story. Crawford asks, if Grandpa can do it, why can’t we?
The discussion covered a lot of territory. We talked about the movie "The Princess Bride". We developed a concept we called "The Man in the Moon Effect". This is the notion that we can see a face in the moon, even though there is no such face there. Our minds impose patterns on reality, and so too will they impose story-patterns on any sequence of events. This suggests that we want to leave plenty of room for the player’s imagination to fill in the details that we couldn’t compute anyway. It may be that the Man in the Moon Effect is the only way we’ll ever get interactive storytelling working.
Another concept discussed was "audience engineering", the notion that we must educate our audience in the forms of interactive storytelling.
It was also noted that we must not be too literal in studying Grandpa. Airplanes don’t flap their wings like birds, but they do copy the same underlying principles.
We agreed that the completely free-form input that Annie is permitted in the anecdote is completely out of our reach. However, some suggested that an input system similar to the Rogerian therapy used so stunningly in Weizenbaum’s ELIZA program might have utility in interactive storytelling.
The overall conclusion was that we can approximate Grandpa’s performance. We can’t put Grandpa’s knowledge of the world into our algorithms, but we can narrow the context enough to get away with it. Worse, we don’t have Grandpa’s knowledge of Annie -- this will be a serious problem. For now, we must strive to build a persistent database on each player.
Questions 12 - 14: How successful have the various design strategies been?
Here we evaluated four different design strategies for interactive storytelling:
1. Event-driven, in which the central data structure is an event
2. Character-driven, in which the central datå structure is a character.
3. Plan-driven: in which characters build plans based on a hierarchy of goals.
4. Plot-driven: we really don’t know what this would be.
We discussed these strategies and concluded that each has its merits. We then constructed a table showing which of the strategies is most likely to be successful in various time frames. For the short to medium term (1 to 5 years into the future), we feel that event-driven methods are most likely to bear fruit, followed by plot-driven, with character-driven in third place. However, for the long term (further than five years out), we feel that plot-driven methods will take the fore, with character-driven methods in second place and event-driven methods in third place. We could see no future in plan-driven methods.
Walt Freitag: Is that really the way we defined "plan-driven?" Was it the characters making the plans autonomously, or some sort of storyteller agent doing so?
I’m a little worried that the description "in which characters build plans based on a hierarchy of goals" might be confusing in the context of the Oz/Petz/Zoesis work. Those characters do possess and use a hierarchy of goals, but they most definitely do not build plans. (I for one would put them in the "character driven" category, reserving "plan-driven" for AI approaches based on planning by logical reasoning over a set of rules or predicates--whether it’s the characters or a story manager building the plans. I would certainly be cautious about declaring no future for the Oz approach.)
Question 10: Data versus process: is there a viable middle ground?
The next question we took up concerned the issue of process intensity versus data intensity. A detailed explanation of the concepts can be found here. Some participants objected to the very question, suggesting that it is really nothing more than a pointless theological arguement. However, the discussion continued and remained lively for nearly an hour. Some of the points made (but not necessarily embraced by the entire group): it is cheaper to generate data than process. In storytelling, characters are a form of data while plot is a form of process. The failure of multimedia was predictable (and was in fact predicted by some) because it is a data-intensive form, and computer software is best when it is process-intensive.
We came to three conclusions: 1) Interactivity demands process; process, however, does not necessarily confer interactivity. 2) A major challenge in interactive storytelling is the division of different components of the technology between process-intensive approaches and data-intensive approaches. 3) Revolutionaries need to consider problems of process-intensity more carefully than evolutionists do.
Questions 6 through 9: not worthy of our time
The group then looked at four other questions and decided that they were unanswerable. Among them were "What kind of people tend to be evolutionists or revolutionaries?"; "Is the gap between the two sides related to the Two Cultures divide?"; "How deep is the divide between the two sides?"; and "What do the two sides think of each others’ agendas?"
Question 5: What market demographics are anticipated by the two groups?
By this time the group was gaining more confidence and people were coming to firmer conclusions. There was rough agreement that, despite all the skewed-data studies so frequently published by games publishers, computer games remain the province of young males. Women and older adults just aren’t interested in games -- but they could be very interested in interactive storytelling. Surprisingly, the group concluded that, in the long run, the revolutionary approach is more likely than the evolutionary approach to achieve broad market penetration, but there are many hurdles to jump. The revolutionaries will likely fail many times, but their group progress will still be faster than the evolutionists. It is vital that interactive storytelling clearly differentiate itself from gaming and establish its own distinct identity as an entertainment medium. We use the word ’medium’ in the double sense of the products and their distribution system. Interactive storytelling will probably establish itself on the web, and the first big success in the field will likely define the entire field for a long time. Lastly, we noted nervously that many other media got off the ground by means of pornography.
Our discussions complete, we made plans for next year. Phrontisterion 4 will be held on June 8-9, 2002, (the weekend before Father’s Day), at the same place -- my home. Its organizing principle will be "demonstrations of actual working products". Saturday morning will be devoted to brief show and tell presentations to the whole group, and Saturday afternoon and evening will be used for more in-depth demonstrations to small groups. Sunday will be devoted to discussion of the strengths and weaknesses of the various products. To make a presentation at next year’s Phrontisterion, a candidate must first submit a proposal to Chris Crawford. He will gather up proposals with their technical specifications and send these out to attendees prior to the conference on a CD-ROM. All attendees are expected to examine the technical specifications documents prior to the conference; the show and tell sessions will presume considerable advance knowledge on the part of the audience. Each presenter will be limited to 15 minutes of presentation time, and a variable amount of time for Q&A. All editorial commentary on products will be restricted to the Sunday discussions. I will also provide genuine name badges for the attendees next year -- and without the funny names this time!
Rick Smith has provided some thoughtful essays on the questions raised at Phrontisterion 3; you can find them here.
The attendees of Phrontisterion 3
Top Row, left to right
Kevin Gliner in the black jacket, Walt Freitag with the white shirt, Barbara Lanza in the yellow jacket, Benjamin Fallenstein ducking low, Rick Smith with a hat, secret attendee in checked shirt, Irene Boczek with the imprinted sweatshirt, and Verin Lewis in the glasses.
Binky Burro in the red halter
Bottom Row, left to right
Tim Emmerich in the light blue jeans, Moose the Dog, Gordon Walton resplendently but only apparently bald, Laura Mixon in black jacket, Chris Crawford in white shirt, Auggie Doggie, Kalev Tait, the only person who remembered to bring his duck (the rest of us were so embarrassed!), and Joseph Breitreiter in beret.