Frustration and Trail-and-Error
Published by Jeremy Douglass June 9th, 2005 in HCTI, hyperfic, IF, Software.
So far in this series on frustration and interactive media, I’ve been thinking different definitions, different approaches, and the way frustration is accomodated by expectation (or not). But what I wanted to think towards was chatbots and Interactive Fiction.
Frustration happens when interactions fail. Either the participant didn’t know enough to form expectations as to a result (illiteracy), or the interface led the participant to form bad expectations (inconsistency). Either way, the participant guessed, and guessed wrong. Guessing leads to failure, and failure leads to frustration. If we want an aesthetics of frustration, guessing should probably be part of it. Or, to put it another way, if we want an art of guessing, frustration will be part of the aesthetic.
By convention and by design, IF introduces the interactor to nouns and verbs during the course of interaction which may or may not be be functional. How do you find out if you can EXAMINE LAWN or FLATTER GUARD?
You can sometimes know what should or should not be understood by the parser, if you are literate in the conventions of the genre, but this may not not help you predict what will or won’t be understood. Even then, being understood by the parser seldom predicts what will generate a productive response - that is, what will be implemented. IF is almost necessarily characterized by weak inconsistency. You can examine things, but is the LAWN an implemented object or background description? You can TALK TO people, but will FLATTER work?
The best available coping mechanism is “trial-and-error,” with ‘error’ here being a new synonym for ‘frustration.’ In communities of IF authors and readers such as ifMUD, there is often an expectation that the system should help the participant to pick and choose their trials through hints and cueing. The result is an aesthetics of managed expectation, in which a gracefully written IF reduces the number of dead-end “incapacity” frustrations in favor of progressive “set-back” frustrations which culminate in a complete, successful traversal of the text.
Some poorly cued puzzles are disapprovingly described by IF developers as “guess the verb” - a situation in which the concept “FLATTER GUARD” is correct, but after trying to “PRAISE GUARD” “SUCK UP TO GUARD” “BUTTER UP GUARD” and “COMPLIMENT GUARD”, it turns out that, in fact, “SWEET-TALK GUARD” works! Even in a genre where agency is constantly frustrated, and “I don’t know what to do!” is considered part of the experience, “I know what to do, but can’t do it!” is a bad thing. The interface may allow inconsistent interaction (i.e. some things are implemented, but not others), but agency shouldn’t come unmoored from the interface (you should know how to do the simple actions which are do-able).
This however is only one potential aesthetic of trial-and-error - with Eastgate Hypertexts being a notable counter-example. Unlike HTML, which (by convention and design) reveals the locations of links and their destinations, many hypertext fictions made use of the functionality of Storyspace to hide linked text and conceal the destination effect - mouse rollover “trial and error” would reveal that a particular letter was clickable, but not whether it was part of a phrase or rather three separately clickable words, each with different destinations. Switching, rotation, randomization, and other “strong inconsistency” effects could also be embedded in the link structure - and all were hidden from interactors.
Unlike in IF, which allows inconsistently available interaction but values clear articulation of agency, blind-link hypertext provides interactions that are completely consistent in their availability - but almost entirely evacuated of agency in their results. You know how to do everything, but not what anything does.
In addition to contrasting processes of trail-and-error, hypertext fiction and IF have contrasting products. In the IF the trail-and-error process begins in frequent failure and culminates in predictive power. In hypertext fiction, the process is more exploratory or exhaustive, with little if any ability to predict the contents of a new lexia gained as the text is read.
This comment is somewhat off-topic, but your post is of interest to me. The more I find out about IF the more I see Botfiction and IF as very similiar forms that can benefit from the lessons learnt in each. For instance, I utilise the IF convention of capitalising words as a cue to my bot users. When I put a word in capitals it is a cue to the user that they can enter that word and will be assured of a response that continues the conversation. Conversely, I find it surprising that IF does not appear to utilise the bot programming convention of pre-empting as many user inputs to capture their response. For instance, the situation you cite of the various ways a guard can be flattered:
A botmaster would have already listed many of these inputs and would enter more as they are discovered through analysis of the conversation logs, or would design a wildcard entry. Is it really the case that IF uses ONE input for each command? I find this surprising…but as you intimate, it may be an intended/unintended frustration aesthetic operating.
I think I wasn’t clear in my FLATTER GUARD example and I’ll revise - just as botmasters are expected to list as many inputs as possible, IF authors are expected to avoid creating “guess the verb” situations, otherwise their designs are criticized - so the two are actually more the same than different.
There is however a difference in verb implementation - in IF, authors have a strong motivation to restrict the verb set as much as possible - *only* implementing a few commonly used ones. This is because of the focus on agency and action (rather than conversational topics) - a bot can easily have a few things to say about any sentence that involves the idea of PRAISE - perhaps a response about something the bot admires, a question about “what else do you love?” a joke… but in IF, once you create a place in which the verb PRAISE functions, interactors will expect that it has been consistently implemented throughout the entire text. Thus you should be able to PRAISE CHILD and perhaps have the parent respond, PRAISE TOAST at the dining hall earlier and at least get a response that the cook doesn’t notice… and if you implement BUTTER UP GUARD, you can be sure that someone will attempt to BUTTER UP TOAST.
I’m not saying that AIML is easier to implement than IF - each design problem faces different challenges. However, AIML very much approaches some problems by aggregated responses, whereas when IF approaches the same problems an aesthetic emerges that involves aggressively limiting the terms of interaction in an attempt to head of the combinatoric explosion created by each new verb or noun. One solution might be something like “GIVE MIRROR TO GUARD” or “ASK GUARD FOR AUTOGRAPH.”
Ah yes, so IF users (interactors/player/etc) need to choose the correct verb for action to occur, and that verb is from a narrow set that all IF utilises. Eg: EXAMINE, GET, MOVE, LOOK. So the experience of IF includes the discovery of verbs with predefined illocutionary force. IF constructors/programmers/creators (is there a term?) should not capture the alternate verbs because that would negate the trope. I can see now how frustration is an aesthetic, as well as usability issue.
Bots usually don’t include the attempt to find the right input as part of the game or storyworld. Indeed, because natural language is the aim, as you mention, there is the problem for botmasters of trying to capture as many possible inputs as possible. As a writer, this is a very interesting exercise — not only prescripting every possible way of saying something; but also guessing what the user/reader is thinking at that exact moment, what they would want to find out, is a fascinating stretch. But as is well known, unframed natural language interaction is a huge usability issue which takes bots beyond the frustration aesthetic to a failed aesthetic experience. But there are many ways of framing that can be implemented on a bot by bot level or perhaps, like IF, botfiction needs to settle on some global rule of interaction? But then that, as you say, takes away the point of natural language. So, to answer my own question: I think more botmasters need to implement some localised rules of play where the experience of frustration can actually be rewarded when the system is figured out. Frustration and Reward, or Frustration and More Frustration, that is the question.
Good discussion - which has spawned more at markbernstein.org and grandtextauto.org, btw — http://grandtextauto.gatech.edu/2005/06/10/bernsteins-bait-redux/
Thanks for the heads up, andrew. I enjoyed reading the GTxA conversaion - stop by here any time. When I have long responses to discussions you’re having over there, do you prefer complete cross-posting, teasers, or just a trackback?
Jeremy — this is a great series of posts. This more complex treatment of frustration will be quite useful for IF authors and critics.
Of course, you’re welcome to comment at whatever length you like over at GTxA, but we don’t mean to head over to your blog in a raiding party and bring the conversation over there. (Well, I guess I should speak for myself - maybe that is what Andrew means to do?)
I wanted to at least briefly mention one classic puzzle that might highlight some interesting things about setbacks vs. incapacity. In a sense, it seems that a setback that the player triggers is a particularly positive failure or frustration: players can at least understand from it that there is such a thing as progress toward a goal, even if they haven’t achieved the goal.
I was thinking that the intermediate steps involved in solving the Babel Fish puzzle in HHGTTG seem to work progress and setback together quite well. You’re glad to have done something that “works” and is recognized as a seemingly-effective action, but at the same time you’re thwarted. Of course, the actual process of trying to solve the puzzle would no doubt involve the player feeling some of the incapacity sort of frustration at times.
The endings of Varicella provide another case that’s similar to this one: You can see you’re making progress in some sense, but are cruelly dealt with and sent back for further thinking and further work. As you hinted at in your next post, Varicella isn’t all that far away from Hamlet - it’s at least Titus Andronicus, which isn’t too shabby for a piece done 25 years after the invention of IF.
Christy,
Yes, the verb set is generally conventional and constrained. Many texts have a relatively small number of specific verbs that have been added. Of these texts, only some require that you discover the verb itself when appropriate (REMEMBER, POUR etc.), while others simply include a description of the unconventional verb in the introduction or help material.
As to your idea about localized rules of play for bots - absolutely. Perhaps these rules could be communicated through error messages which more aggressively steer you towards the bot’s areas of conversational strength - it communicates to you what your localized expectations should be. Which, come to think of it, is what people often do at cocktail parties. Mention math to your cocktail-bot, and it replies… “Ah, math. You know, my husband doesn’t understand math either…” to which you could reply “Why on earth not?” but it is less rude and more productive to say “What does your husband do?”
Thanks Nick,
Nice - the fish from HHGTTG is a perfect example of elaboration - each success making possible a new and more interesting failure, until all the failures have been dealt with. I can see where my language starts to slip when stages of progressive solution are labeled “setbacks,” but in terms of frustration I think the label still works because at each stage a new error / failure statement is generated.
Varicella is a showcase of exteme setbacks, because the process is elaboration / resource management, and the major form of feedback is “learn through dying” - hard to get more frustrating than that. Yet Adam Cadre makes it work by generating unique interest in each road to failure, which is what brought him to mind for Hamlet - come to think of it, Lock and Key also takes the form of elaboration / resource management and learn through dying.
Speaking of Cadre, in the next frustration post I also cite Cadre’s 9:05 when talking about dramatic irony - and I might as well have added shrapnel to the mix. From now on perhaps my advice to people interested in either frustration or dramatic irony in IF will be to read some Cadre first - not that he has the market cornered, but his sensibility consistently leads him in that direction.