December 11, 2006

New Interactive Drama in the Works (Part 2)

by Andrew Stern · , 9:25 pm

Yikes, I’m over two months late with Part 2 of this post announcing production of The Party. My only excuse is that life is hectic for me these days. As Part 1 described, pre-production, authoring-tool building and prototyping of The Party has been underway for about a year now.

Building next-generation interactive characters and stories has many design, technology, production and fundraising issues. In this post I thought I’d lay out a series of issues we’re grappling with, and give my initial take on each. (I may have subtly different takes than my collaborator Michael.) We’d like to hear any thoughts, opinions and suggestions you have on these, and anything major you think I’ve left out. This helps us think through the issues and figure out solutions.

First, I have two assumptions that I hope won’t require debate in this thread, but can be taken to another thread if requested. For the sake of discussion, let’s assume that interactive comedy and drama games are marketable if and when the games are well-made and compelling to play, even though this market is largely unproven to date; and therefore in theory there will be investors interested to fund their production. Two, assume it is theoretically within our current means to build a satisfying interactive story, though this is largely unproven to date. My gut tells me this is true, based on what we’ve learned so far trying to build one (but not yet succeeding) plus my experience building related products.

In these efforts we’re building upon the design techniques and technology gained from creating Façade. But Façade is a mixed bag; it has some successful entertaining elements mixed in with some not so successful elements. I’ll try to unravel these into a set of questions and issues, that get to the heart of what I believe is required to create a compelling interactive story/game that can succeed in the marketplace. In my estimation these are necessary, though perhaps not sufficient, requirements for commercial success; again, if you feel I’ve missed anything important, please say so. (There are obvious requirements I haven’t listed here, such as the need for good writing, animation, etc.)

Issue 1) Our approach is to break down and represent a story space as a large collection of small, procedural narrative machines called beats, each of which are composed of a collection of behaviors and lines of dialog. Beats are designed to be able to intermix nicely and be sequenced in many different orders. Can this technique allow for a satisfying richness of non-linear real-time generation of interactive stories, and did this occur in Façade?

Issue 2) Even if there is richly generated story, can the player affect the generation of the story in significant, meaningful ways? I.e., is satisfying agency possible, and did this occur in Façade?

Issue 3) Using an open-ended natural language (NL) interface, Façade attempts to allow the player to speak anything they like to the characters. What succeeded and what failed here?

Issue 4) Of the failures in the NL interface, can they be improved enough in the short-term to justify its use in a commercial product? What would be gained and lost by using a different interface?

Issue 5) Thematically, Façade‘s drama was about a marriage in crisis, which was not much fun for many players, although most said it was at least “interesting”. How endemic is this lack of fun-ness to the form of interactive drama in general?

Issue 6) By rendering its characters in a more illustrative style a la alternative comics such as Optic Nerve and Eightball, Façade attempts to break away from mannequin-ish, stiff-faced polygonal characters, hyperrealism and the uncanny valley. Simultaneously this technique allowed for the faces to be drawn procedurally, and therefore results in them being quite fluid and expressive. Yet, many players and reviewers considered Façade‘s graphics to be crude. What’s the solution here?

Issue 7) How can you fundraise for a product that proclaims to innovate a new genre of interactive entertainment — a market has to be proven before investors will invest, no?

Here I’ll give my brief take on all but issues 3 and 4; I’ll (and hopefully your comments will) save Part 3 for the discussion of NL interface questions.

Answer 1) How do you achieve satisfying richness of non-linear real-time generation of interactive stories?

Any richly non-linear interactive story implementation will involve breaking down the narrative into lots of little pieces of some sort. A large number of those pieces will always be needed, because by definition, richness requires a great deal of variety and quantity of content. (In any one play-through of the interactive story, perhaps only a fraction of the total amount of available content is experienced, but the system needs a very large pool of potential story content to choose from.)

Although Façade is comprised of 2500+ behaviors employing 5 total hours of voice-recorded dialog snippets, it didn’t add up to enough beats to achieve a satisfying richness. This was primarily because of how long beat behaviors took to author — not a function of the nature of beats per se, or how the architecture as a whole is designed. Of course we intended to create much more content for Façade, but we got bogged down on how long it took to make good beats, and had to cut many beats we hoped to make.

The solution for The Party and future projects is three-fold. One, create authoring tools to speed up the authoring of beats and their behaviors; we’re doing that now. Two, build as much of the experience as we can in a simplistic text-bubble version, requiring all of the AI and dialog to be implemented but almost none of the animation — roughly equivalent to writing the screenplay for a movie before it’s shot/animated. This should allow for more rapid authoring. Three, put more people on the team than just me and Michael!

Once you have lots more content, your architecture and authoring process need to be able to handle it all, and work with it well; so far, we think our architecture can do this.

A longer-term solution for creating lots of content is to procedurally generate beat and behaviors themselves; that way you can have an order of magnitude more available for the system to choose from. This is long-term R&D work, it could take decades really. (Currently our architecture doesn’t procedurally generate beats, but does procedurally generate sequences and intermixing of beats.)

Answer 2) Can the player affect the generation of the story in significant, meaningful ways? I.e., is satisfying agency possible?

This is the most important issue of all; agency is the holy grail of interactive stories. Get this right and it can go a long way towards forgiving any other problems in the system. I’ll spend the most time in this post on this issue.

Agency in Façade was spotty; the level of agency varied throughout the experience. Façade gave more local plus global agency than perhaps any interactive story I’d played in the past, but that’s not saying much. I think many would agree that overall, Façade did not give satisfying agency. There are several reasons for this.

One, an interactive story needs a very rich pool of story content to achieve agency, and Façade didn’t have that. We hope to fix that in The Party, as described in Answer 1 above.

Two, to have agency, the system needs to understand what the player is saying, in order to respond meaningfully. Façade doesn’t always understand the player properly; in Part 3 I’ll discuss those issues (issues 3 and 4 above), and potential solutions.

Three, even if the system understands you perfectly, to achieve agency the player needs to be able to understand what effect they’re having as they act. The system needs to give clear and immediate feedback about how your actions changed the state of the story/game. I agree with my friend Katherine Isbister when in an interview she said Façade doesn’t give clear feedback. (She was too kind to say “that’s not necessarily a bad thing”.) In Façade, too often the player says something to the characters and gets a response, without a good understanding why they said that back. This was most noticeable in the second half of Façade, when Grace and Trip face off on opposite sides of the room (what we call “the therapy game”). Though, overall, it was much better than the status quo of chatterbots.

Why wasn’t feedback better in Façade? (Again, assuming the system understands the player properly.) The subject matter of our drama hampered us here. In a psychological drama, the game state the player is manipulating is a heterogeneous collection of the characters’ attitudes, feelings, awareness, knowledge and so on. These are very difficult to fully and completely express back to the player in a naturalistic way. Naturalistic interactive drama, that we were striving for in Façade, by definition has no sliders, bar graphs, or HUDs to show the player an information display of internal state of the characters’ minds. (One reason The Sims succeeds as a game with decent feedback is that you can see bar graphs of the characters’ state changing as you act; simultaneously, that’s one of the big reasons The Sims fails as compelling interactive drama, IMO.) In Façade we attempted to give feedback of Grace and Trip’s relatively complex cerebral state by way of dialog, facial expression and gesture, but it often was not enough, or not clear enough; it’s difficult to pull off, while avoiding the writing being too “on-the-nose“.

(In real-life as well, it can be very difficult to understand what effect you’ve had on someone’s internal state when you’re speaking with them. But, this is drama, and our characters should have license to communicate in more efficient ways. Good writing, in any dramatic medium, does this without seeming forced or unnatural.)

One solution to the feedback problem, that we’re pursuing with The Party, is to reduce the complexity going on inside the head of any one character, and spread that complexity among a greater number of slightly simpler characters. Also, make the story more action-based in general, and less cerebral; when the player acts upon a character the effect is more noticeable in their physical action, not just a change inside the mind and relatively subtle facial/body expression. In other words, don’t build a purely psychological drama, build a more soap-opera-like comedy-melodrama.

Answer 3) How well did the NL interface work?
Answer 4) Can the failures be improved?
Answers postponed till post Part 3.

Answer 5) is Façade‘s un-funness endemic to interactive drama?

That’s easy: of course not. A drama about a marriage falling apart is probably bound to be more uncomfortable and less fun to play than a comedy-melodrama about a collection of over-the-top characters partying, seducing and manipulating one another. Next question.

Answer 6) can you have non-photorealistic characters that don’t look crude?

First, I don’t think Grace and Trip look crude. Sure they are flat-shaded, and don’t have a lot of detail, but I think they look great, nicely matching the alternative-comic look I was going for, and their faces are more expressive than any I’ve seen in any game (although their body animation is weak; this can be greatly improved with an animation staff).

Ask me right now, I’d say I think it makes sense for The Party and future products to avoid hyperrealism, and to continue rendering the animation in a non-photorealistic style of some sort — though, the hyperreal HD game trailers of late (1 2 3) are very seductive, I have to admit. That said, I want to see more richness and detail in any non-photorealistic look of our future projects. (One nice thing about taking so long on the AI side of things is that more animation tools and technologies become available over time, as well as an increase in the speed of player’s hardware! :-)

Answer 7) How can you fundraise for a product that proclaims to innovate a new genre of interactive entertainment — a market has to be proven before investors will invest, no?

It’s tough. (The Atlantic Monthly article suggests we’ve already raised money for the project; we haven’t. The $2MM deal mentioned in the article was actually to a deal to work to fundraise that budget — and, we’ve since moved on from working with those particular producers.)

Basically, you have to make a convincing case that your product will sell well, and you have to find investors willing to take a chance. Even though we got incredible press and close to half a million downloads for Façade, as a prototype of a commercial product, Façade is is no slam dunk for proving the viability of The Party, for two reasons. It doesn’t work well enough (issues 1-4) to totally prove this will fly; a case has to be made for how our in-the-works improvements will get us to a solid product-level, and/or build a new prototype actually demonstrating it. Two, the thematic nature of Façade (issue 5) doesn’t give an investor simple proof we can make something fun, although that shouldn’t be too hard to make a good case for (write new sample scripts, make an animated trailer, etc.), and my track record with Petz helps too.

While I hope we’ll be able to raise the full budget up front, it could be we can only raise a small amount at first, with which we’ll use to build a good playable short prototype of the The Party, proving solutions for issues 1-5. With that prototype we’ll then raise the remaining full budget.

(To answer Kenneth’s question from Part 1 whether we’re focused on making stories/games versus licensable technology: we’re primarily focused on making stories/games, but are open to licensing the technology if we can justify the time and expense.)