|
Argument
Agents By Dirk Scheuring Abstract: Conversational Agents with Natural Language Abilities are usually programmed, or "written," using "dialog branching" strategies. Writers following this "resolution-based" approach generally encounter the problem of "state explosion" - i.e., after several of these "branching" operations, there are just to many separate branches, and therefore, dialog possibilities, to be handled effectively with current technology and technique. This leads to what Sengers[1] describes as "agent schizophrenia"; put informally, the agent seems "un-believable" to the human User (a.k.a. the "User" or "Dialog Partner"), which, especially in the context of automated business transactions, often translates to "un-trustworthy." The author´s perspective on Conversational Agents is based on the premise that they represent, not "chat machines," but a novel Storytelling medium. "Stories," in this case, are defined as goal-oriented Human-Computer Interactions; e.g. online sales transactions, "troubleshooting" conversations, etc. Stories of this type can be modeled using the Grand Argument Story model developed by Phillips/Huntley.[6] An Argument Agent, as represented by this Blueprint, uses a "table-based" strategy in order to "argue" with the User about progress towards the Story Goal, thereby dramatically (in two senses of the word) restricting dialog possibilities. This leads to the AA appearing much more "reasonable" to the User than the average Conversational Agent using a "resolution-based" dialog strategy. Keywords: Intelligent Agents, Narrative Agents, Deliberate Agents, Believable Agents Aims of Research The goal of the Argument Agent Project is to develop a framework and software environment to support the development of Argument Agents. The ArgumentAgent Class which these agents are instances of is a Subclass of the ConversationalAgent class. This is a GNU Project. Conversational Agent: Description A Conversational Agent, a.k.a. "Chatterbot," can be loosely described as a Database with a Natural Language Processing User Interface. It processes User Queries and responds by outputting a matching Template Value for each Query it receives, the process following this pattern:
This basic mechanism is employed by a Designer of Conversational Agents to enable computers to be "simulated dialog partners." Conversational Agents are, as of now, generally thought to be subject to the design premise TheBiggerTheBetter, i.e. the larger the matchtable, and therefore, the higher the number of User queries the agent (in the majority of Use Cases) "knows" a (for most Users) "satisfying" answer to, the more Users believe it to be "intelligent." And UserConsidersAgentIntelligent is the key quality when it comes to evaluating a Conversational Agent, who is a Subtype of the Intelligent Agent. This is the Growth-Is-Good-approach, as effectively represented by Alice (www.alicebot.org): Alice, as of May 2001, matches about 24.000 User Query Patterns to its Template Table. "Dramatica®" And The Argument Agent: Introduction An Argument Agent represents the instantiation of an alternative approach. The underlying premise of AA-Design is that the agent, rather than representing a KnowledgeDomain, which is, potentially, theWorld, represents an ArgumentDomain, which is aStory. In contrast to KnowledgeDomain, which is an abstract Class with infinite members, ArgumentDomain represents a Story Argument, which can be looked at only from a given number of perspectives, and is therefore finite. The Type of Story employed in AA Design is the "Grand Argument Story©," as defined in the "Dramatica®" --Theory and implemented in the software of the same name. Successful Grand Argument Stories can be loosely described as providing the Story Recipient with a feeling of "Closure" when their end is reached. A Set of well known Instances of this Story Type could include "Star Wars - Episode 4" by George Lucas; "Blade Runner" by Hampton Fancher and David Peoples; "Casablanca" by Julius Epstein, Philip Epstein, and Howard Koch; "Hamlet" by William Shakespeare; "Reservoir Dogs" and "Pulp Fiction" by Quentin Tarantino; "A Doll´s House" by Hendrik Ibsen; "The Sun Also Rises" by Ernest Hemingway. Each Grand Argument Story can be described using one specific Storyform out of the Set of "Dramatica®"´s 32.768 Storyforms. Argument Agents, like other Conversational Agents, have Knowledge -- but just enough to completely render their ArgumentDomain. Unlike past AI efforts which were meant to eventually lead to Superbots featuring all-encompassing WorldKnowledge and eventually intellectually crushing humanity, Argument Agents are intended to be relatively small and, essentially, dumb. They are designed to argue one case at a time, but to do it well. One particular "Dramatica®" Storyform represents an abstraction of one particular Argument. There is no room for the human User to leave this context and still provide dialog contributions that other humans would rate as "making sense" -- the User would not appear "intelligent." The Argument Agent, on the other hand, will always "make sense," and therefore, appear reasonably "intelligent" inside the story´s context, because he relays each and every User Query to the limits of the Argument. An AA is "dumb AI" that still "makes sense." On Inspiration Science Fiction characters, like the ones out of Isaac Asimov´s "Fantastic Voyage" or Neal Stephenson's "Snowcrash," have a habit of inspiring scientific research. In particular, there is one movie character that has been an inspiration to many AI researchers and engineers: It is HAL 9000, the "living computer" out of Stanley Kubrick's 1968 movie "2001 - A Space Odyssey." For instance, HAL has provided a blueprint for the CYC project, an extremely ambitious venture to capture and replicate human "common sense knowledge" with a machine. CYC´s main author, AI pioneer Doug Lenat, initiated the project in 1984. His stated goal was not only to drag the HAL "character" out of the movie and into the "real world," but to build a better version than the one Kubrick and Arthur C. Clarke - who not only co-wrote the "2001" -- script, but also the book on which the script was based - had written. Unlike HAL, CYC would never attempt to kill the crew of a spaceship, because of hasCommonSense (in this use case, "morals" of some sort), and common sense would "tell" the machine that killing unsuspecting humans is a very uncool thing to do. Says Lenat: "To me, HAL's biggest crimes were his conceit and his stupidity." [2] Although CYC provided the inspiration for what would eventually become this Blueprint for Argument Agents, CYC´s founding concept can easily appear flawed in the eyes of a screenwriter, or, indeed, in the eyes of anybody who knows the basics of dramatic storytelling. A storyteller cannot somehow "extract" one character from a finished story and make it "live on its own," by scraping "the rest of the story" without any substitutions. This is because a finished story represents a complex "weave" of Genre-specific, Plot-specific, Theme-specific and Character-specific aspects (we´ll call them Appreciations, each connected to one of said four Levels), finely tuned (we hope) by the storyteller to present an argument in order to convey a message. In analysis, of course, we are able to look at each Level separately; however, trying to present one Level of the argument -- in this case, the "Character" -- Level -- outside of the context that the others provide will not work, since we loose the message, and therefore, the meaning. Put simply, without plot, there is no character; without character, there is no plot, without either, there is no meaning. A Case For Drama To many of today´s AI practitioners who research and/or engineer Intelligent Agents capable of "understanding" and outputting natural language, the notion that their creations should be seen as "dramatic characters," and that the rules of dramatic storytelling are applicable in order to make these "characters" more life-like and "believable" as dialog partners, probably, at the moment, seems like stretching the idea of the art-science-crossover a bit too far. On the other hand, the concept of Narrative Agents has found quite some resonance in the AI community during recent years, and there exists at least one storytelling system -- BRUTUS_1 -- apparently capable of producing decent dramatic "short-short stories."[3] Moreover, few would argue that the latest flock of Embodied Conversational Agents -- heirs to Weizenbaum´s ELIZA, currently servicing websites as virtual "Insurance Consultants" or "Site Guides" -- represent a new Class of Actors that can be stereotyped as "fictional": ECAs are made up, after all. But "dramatic"? Where´s the "drama" supposed to be? To a screenwriter looking at today´s ecommerce realities, the answer should be obvious: Wherever there is an "abandoned shopping cart," there has been "drama." Many screenwriters start working on a story by conceiving of one first character. Taking this approach, the author starts modeling by introducing Character Actor_X. Actor_X isWriter, i.e. the author attributes him with the Role "Writer" (note that Actor_X is a productive set, in the sense given by [Bringsjord], i.e. capable of true creativity). A Use Case has Actor_X writing a dramatic story about the sale of a computer, involving two Characters, represented as Actor_1 and Actor_2. Actor_1 isBuyer; Actor_2 isSeller. The StoryGoal is closeTheDeal. The Writer writes: "The Buyer enters the shop." Immediately, he has a lot of dramatic potential -- possible scenes -- to play with. Maybe the Buyer is not yet sure which machine might suit her needs -- the Seller makes a suggestion, but fails to convince the Buyer that the machine in question actually is "the right one." Or: The Buyer is looking for a particular model -- the Seller doesn´t stock this brand and scrambles to offer a substitute product. Or: The Buyer finds a computer she likes -- the Seller, however, for some reason does not appear trustworthy to her, and she refuses to hand him her credit card. And so on -- the argument evolves, with dramatic tension flowing from side to side, and is only resolved once the buyer signs the credit card receipt (in writer´s terms, this is called the "Moment Of Truth") and leaves with the merchandise. Or else, the Buyer´s argument against the offered merchandise is simply stronger than the Seller´s argument for it, and the Buyer leaves without buying anything. In this case, the Seller, who isEmployee, when (in a story following this one) reporting to his boss, who isEmployer, should at least be able to prove that, even though he was ultimately unable to reach the Story Goal and closeTheDeal by meeting the Buyer´s unrealistic premises, he did, in fact, argue as well as he possibly could. For the Sake of Argument Watching human salespeople argue with potential buyers can be a fascinating experience: The really good ones, without a doubt, are "creative storytellers," and, according to the suspicions of at least some AI researchers, true creativity might forever remain un-computable, since the Church-Turing Thesis might be false But even average salespeople "do" Arguments, as good as they´ve learned to. Websites, usually, don´t "do" Arguments at all. Potential buyers might find that a website "stating good reasons" for buying a particular product -- but "stating good reasons" is not the same as "arguing," since Statements are, as the term implies, static by nature, and Arguments are dialogical in form, representing the contributions of, at least, two Co-Constructors and evolving dynamically. "Conventional" Conversational Agents don´t "do" Arguments, either: They do not succeed in helping the User construct enough context so that he is able to understand a complete Argument, by having access to all the necessary information at all the necessary moments in time. This is because their architecture focuses on deducting the meaning the User seems to try and give the current conversation based almost only on the Users last input line. Though it is true that, for instance, Conversational Agents Authoring Software often features the possibility of tagging keywords to the Agents´ output, writers currently use this feature only to provide local context for pronoun disambiguation at random times during the dialog, not as a way to introduce contextual structure to it as a whole. Deductive Agents These Conversational Agents act like automated chess players: they wait for the User to make a move, then -- at their best -- they try to deduce his next move by "thinking through" some possible combinations for own next moves and their consequences, then choosing the answer that seems fitting, yet might be "loose" enough to be "excusable," in case it doesn´t fit.. Because the machine is not smart enough to do any of this "thinking through" on its own (playing chess is much easier for computers than dealing with natural language [Bringsjord]), human writers have to jump in and fill the gap, and they do their very best to deal with the the constant explosion of possible meanings that goes with trying to construct dialog this particular way, i.e. by Deduction. Inductive Agents Contrastingly, an Argument Agent is programmed to seemingly operate by Induction. The writer might, and probably will, make it capable of displaying certain deductive behavior whenever this serves its function, illustrates its character, or otherwise supports the storytelling. Ultimately, though, the AA will not try to "follow" the User by deducing the possible meanings of each query (or Input Event); instead, when "in doubt," it will jump to a conclusion in order to advance the Story. A Brief Look at Real Life Human sales and service people do this all the time -- they will often -- stealthily -- interpret their customers' utterances, manipulate them in their own mind and come up with a "reframed" version to "shoot back" at the customer. Without most people noticing, this transition technique is used to redirect the conversation to another "scene." For instance, if a woman working behind a bar receives yet another sexually motivated "come-on" prompt by yet another horny male customer, rather than probing the depths of his mind, she will just find a way to establish that, yes, she understood him, but he has "stepped out of his role." To "rope him back in," she will use a proper transition, then "enact" a scene with him that has worked for her on past occasions of the same type, starting with him being obnoxious and ending with her selling him yet another drink. She uses generally accepted "role conventions" for that ("Drinker" and "BarTender," in this case). With enough experience on her side, this will work in most cases. (Sometimes, of course, the customer insists on stepping out of the boundaries of his role and cannot be roped in. This is how horny men get thrown out of bars.) One good thing about this tactic is that the lady may learn to use exactly the same "scene" in other contexts as well, by generalization. For instance, she might decide that "rampant horniness" is just a more specialized case of "male frustration," and use the same "scene" on other men with other reasons to be frustrated as well -- just with a different transition! "Dramatica®": Some Vital Definitions Before progressing, it is important to accurately define at least some fundamental "Dramatica®" terms: Story Mind: "The central concept from which Dramatica was derived is the notion of the Story Mind. Rather than seeing stories simply as a number of characters interacting, Dramatica sees the entire story as an analogy to a single human mind dealing with a particular problem. This mind, the Story Mind, contains all the characters, themes, and plot progressions of the story as incarnations of the psychological process of problem solving. In this way, each story explores the inner workings of the mind so that we (as the audience) may take a more objective view of our decisions and indecisions and learn from the experience." Throughline: "The sequence of appreciations over a story that describe one of the four Perspectives in a story. "The Throughlines are the four structural Perspectives which each movetoward facing its own problem as the story reaches a climax. The Objective Story, Subjective Story, Main Character, and Obstacle Character Throughline all have their own distinct appreciations which have to be illustrated to create a Grand Argument Story, but Storytelling choices can be made to accentuate a particular Throughline and emphasize it more than the others (. . .)" Appreciation: "Appreciations are items of dramatic meaning that are common to all stories. Meaning is created when an identifiable topic is seen from a particular point of view. This creates perspective which takes into account both the observation and the observer. In complete stories, there are four principal viewpoints at work: Objective Story, Main Character, Obstacle Character, Subjective Story. Each viewpoint has its own unique Appreciations, though they parallel and match item for item the Appreciations from another viewpoint. In addition, some Appreciations are from a wider view, describing the relationship among the viewpoints and the dramatic results of their combined perspectives. In this manner, a story structure built on Appreciations will cover all the topics and viewpoints necessary to fully explore an issue central to them all. Common Appreciations include such dramatic items as Goal, Requirements, Problem, Concern, and Outcome." Argument: "The progression of logistic and emotional meanings that combine to prove a story´s message. "A story´s message is proven by a progression of logistic (dispassionate) and emotional (passionate) meanings which are created by the interaction of Character, Plot, Theme, and Genre. The dispassionate argument is the story´s contention that a particular approach is the most appropriate one to solve a particular problem or achieve a goal in a given context. The passionate argument is the story´s contention that one world view is better than another in terms of leading to personal fulfillment. An author can use his story´s argument to convey his message directly, indirectly by inference, or by making an exaggerated argument supporting what he is against." Grand Argument Story: "A story that illustrates all four throughlines (Objective Story, Subjective Story, Main Character Story, and Obstacle Character Story) in their every appreciation so that no holes are left in either the passionate or dispassionate arguments of that story. "A Grand Argument Story covers all the bases so that it cannot be disproved. From the perspective it creates, it is right. There are four views in a complete story which look at all the possible ways the story could be resolved from all the possible perspectives allowed; these are represented by the perspectives created by matching the four Domains with the four Classes -- (the Objective Story, Subjective Story, Main Character and Obstacle Character Domains matched up with the Classes of Universe, Physics, Psychology, and Mind) to create the four perspectives entirely so that their view of the story´s problem is consistent and that they arrive at the only solution that could possibly work, allowing the givens built into the story from the start. When this is done, a Grand Argument has been made and there is no disproving it on its own terms. You may disagree that the things it takes for givens really are givens, but as an argument it has no holes." (all quotes)[6] An Example: "Star Wars" Using the "original" "Star Wars" movie from 1977 (a.k.a. "Episode 4"), which features a Grand Argument Story structure, as an example, the concept can be explained this way: Luke Skywalker is the story's Main Character - his Throughline (the "I"-Perspective) is a substory in the Story Mind. The Character who, throughout the Story, continually tries to make the Main Character "change his ways" in order to prepare him for the "Moment Of Truth" is the Obstacle Character - in the case of "Star Wars," that would be Obi Wan Kenobi, who has his own Throughline (the "You"-Perspective). The Objective Story Throughline (the "They" Perspective) deals with the rebels fighting the empire. Finally, the Subjective Story (the "We" Perspective) illustrates the relationship between an old, experienced, but somewhat out-of-date jedi knight (Kenobi) and his young, hot-headed apprentice (Luke). Any deeper discussion of storyforming and storytelling using the "Dramatica®" tools is out of scope of this Blueprint. Nevertheless, the Story Engine Settings for the Storyform of the Story for the ArgumentAgent <<misssalesman>>Pizzaman this Blueprint will from now on refer to as its main example is appended to this document. Pizzaman: A Prototype Actor_X, who isWriter, meets Actor_Y, who isClient. The Client manages a pizzeria called "Don Camillo," where the best pizza in the city of Cologne is produced. The Client commissions the Writer with writing an Argument Agent that sells said pizza to Users of the pizzeria´s website. Storyforming First, the Writer casts the Agent as his story's Main Character ("I"-Perspective). This Throughline shows the Agent´s point of view: What he plans to do -- sell pizza -- what ideas he has about himself and his job, how he tries to fulfill his role to accommodate the User, and how he strives to, finally, become the perfect Pizzaman. The User is cast as the Obstacle Character ("You" Perspective). The "worst" User is assumed by the Writer -- one that is totally uncooperative and tries to subvert the Agent´s approach at every corner. This Throughline describes how the User has to understand that the Agent is not HAL but a pizza salesman, how he can choose from the food and drinks menu while other services are out of scope, how he actually will have to pay to have his pizza delivered, and finally, that he has to give his home address to the Agent in order to get it. This defines the most "negative" view on the Argument. The Objective Story ("They" Perspective) describes the general setting: the pizzeria, its history, why its talented team could progress to leadership in pizza-baking, how the pizzerias R&D strategy assures constant further improvement, and the daily struggle of everybody involved to maximize customer satisfaction. The Subjective Story ("We" Perspective) deals with the emotional relationship of User and Agent during the ordering process: Whether the User is a first time customer or has used the service before, what food and drinks he might need, how his hunger will soon be satisfied, and the value the Agent gives him for his money. This defines the most "positive" view on the Argument. Then, the Writer has to decide on an Objective Story Theme that defines the most general Story Context. At this point in the history of User-Agent relations, a good Theme obviously is "Fantasy v. Fact": Users very often, when conversing with a Conversational Agent on the Internet, try to figure out how this Agent matches the general hype around this type of software devices, or the AI fantasies (like HAL, "Star Trek" etc.) the media have spawned. On the other side of the "thematic scale" resides the Fact. Without going too far into the discussion of Dramatica Theory, the choice of Objective Story (and Obstacle Character) Problem and Solution in the Storyform <<misssalesman>>Pizzaman should be briefly explained: The Problem isEnding because the biggest danger is that the User will end the dialog before the Agent could closeTheSale; the Solution isUnending when the User decides he wants to "play by the rules," i.e. he continues with the Agent through the whole process in order to get his pizza. Storyencoding, Step 1: "The Floor" When the dialog starts, the Agent starts out "pre-loaded" with an initial Topic. Each current ECA authoring software has some function to deal with "topics," and with each it is done slightly differently; for instance, in Artificial Intelligence Markup Language (AIML), an XML specification developed by Dr. Richard Wallace, the markup <topic>X</topic> is used. Interaction with the User might change the "active" Topic in the Agent´s memory, thereby indicating a change of context, and thus, a change of Scene. Scenes are composed of a Sequence of Story Events. When a Scene´s end is reached, the Agent, also, can query the User and, according to his reaction, go to one of several follow-up Scenes. But only when a User´s Query completely falls out of the Agent´s scope; i.e. the Agent doesn´t "understand" a single word in the current Query, the next event of the current Scene is called up. In AIML, the markup <pattern>*</pattern> (the "Star"-pattern) is used to call up a Value in response to a Query that the Agent found no other matching pattern for. Now, if the Agent is forced to match a Query with <pattern>*</pattern> while at <topic>N</topic>, he advances the Story by responding with the next EventValue - depending on the Answer Rank -- from Scene N: Main Character Scene N Key: <topic>N</topic> Event N1 Answer Rank: 1 StringValue: Event N2 Answer Rank: 2 StringValue: Event N3 Answer Rank: 3 StringValue: Maybe you´re skeptical - pizza salespeople are not widely known to deliver on their promises. New Topic: <topic>IV</topic> Event N4 Answer Rank: 4 StringValue: At <pattern>YES< /pattern> und <topic> IV</topic>: Agent goes to Scene D, Event 1, new Key: <topic>E</topic> At <pattern>NO< /pattern> und <topic> IV</topic>: Agent goes to Scene O, Event 1, new Key: <topic>P</topic> At <pattern>*< /pattern> und <topic> IV</topic>: Agent goes to Scene D, Event 1, new Key: <topic>T</topic> Taking the opening Scene of the Main Character Throughline as an example, this illustrates the concept: Whenever communication breaks down, the Agent supplies a Story Event that, though not matching the current User Query, matches the current Context, thereby advancing the Story. Storyencoding, Step 2: "The Transitions" The next "layer" consists of another Class of User Queries: Queries that the Agent has found a match for in its Database, at least a partial one, but that also cannot be rated as "direct" contributions to the process of negotiating a pizza sale. These Queries will not directly bring up Story Events, but they will change the current <topic> tag. Therefore, if after such a "tag change" a <pattern>*</pattern> is detected, the Agent will jump to a another Scene. Those "tyg changes" might happen in the background for quite a while -- as long as the Agent understands some part of the Query, it will respond with "transitional" Values, trying to focus the User on the task at hand -- ordering pizza. But when communication breaks down, it will "know" what the last context was, whether the User has responded positively or negatively to it, and will go to a matching Story Event. One might think of it being a table with four columns of output Values, each representing a different thematic context, like A v. B, but linked to the other contexts (C v. D, E v. F, X v. Y). Each describes a Throughline, consisting -- at the current level of resolution - of four Scenes. The Scenes escalate in relation to the Throughline Goal as well as in relation to the general Story Goal (the "Moment Of Truth). The Scenes consist of four Events each; the Sequence of Events is also escalating, leading to a mini "Moment Of Truth" at the end of each Scene, where the Agent queries the User an, as a result to his response, keeps the current Throughline or changes to a more "positive" or "negative" one. The length and number of Scenes -- the resolution -- can of course be adjusted according to the complexity of the transaction the Story is supposed to model. Theoretically, using the current resolution, a total of 64 Story Events could be fired by the Agent. In praxis, this will rarely be the case, because of the existence of Act Breaks. The Argument Agent uses an overall Story Structure consisting of three dynamic Acts. This is one of the oldest Storytelling techniques, first proposed by Aristotele, circa 350 B.C.[7] All Throughlines start simultaneously at Signpost 1. Following the User´s whims, some Events from this, that or the other Throughline might get triggered. But: as soon a one Throughline reaches its Act Break at Signpost 2, the context shifts. According to his reaction to the Agent´s End-Of-Act query, he will get transferred to the first Scene of the matching Throughline´s second Act, the Throughlines will get synchronized again, and the Story gets advanced from there. The same happens at Signpost 3, starting the third act. Storyencoding, Step 3: "The FAQs" All this, of course, is nothing but a sophisticated safeguard mechanism against User Queries that, whether motivated by benign or mischievous User intentions, do nothing to advance the Story in relation to its Goal, which is closeTheSale. An extremely well-behaved and cooperative User might be able to buy his pizza without ever touching the "Floor" and forcing the Agent to move on with the plan. This is because every input the Writer can anticipate as a "valid" (in relation to the Goal) Query will -- hopefully -- be matched by a direct response, like a straight answer to a straight question. This is the part of the world the Argument Agent is supposed to have "hard" Knowledge about: everything that has to do with the pizzeria, the products on sale, his own role as a salesman, etc. Values of this Class will not change the current Topic, because, as long as they are output, everything can be considered to be running smoothly. This, however, will rarely be the case in User-Agent dialog. Status/To-Do Currently, the prototype Agent Miss Salesman, presenting a Limited Argument (one "amalgamated" Throughline and about 800 Transitions and FAQs) demonstrates the validity of the concept - for an Agent with this little content, Miss Salesman appears unusually in the context of ECAs -- "intelligent." Miss Salesman was designed using a commercial ECA Authoring Software ("Lingubot Creator®," a Kiwilogic®product). The author currently works at three main tasks:
The first two of these tasks are supposed to be completed before September XX, 2001. Conclusion Of course, from the point of view of the "classical" AI Scientist, the Argument Agent is a fluke. The clever things it might say that convince people that it is "intelligent" do not originate from any "personality" that the machine has somehow developed, like human beings develop theirs. We might soon reach a stage where the AA appears to be the caring, sharing, customer-loving machine any internet vendor would like to employ -- the author, for one, dreams of the day he´ll start to write his first AA for the CYC engine. But the machine can only do this after a clever human being, an artist, a con woman, a trickster, has put the caring, sharing stuff into it. Basically, the AA is just another doll on stage, with artists and engineers pulling the levers from behind a black velvet curtain. It is probably time for a re-read of Weizenbaum.[8] And of Aristotele. References: [1]Sengers, Phoebe. "Symptom management for schizophrenic agents." In AAAI-96. Menlo Park, CA: AAAI Press, vol. 2, 1369, 1996. [2]Lenat, Douglas B. "From 2001 to 2001: Common Sense and the Mind of HAL." In Hal's Legacy: 2001's Computer as Dream and Reality. Edited by David G. Stork. MIT Press. 1998. [3]Bringsjord, Selmer & Ferruci, David "AI and Literary Creativity: Inside the Minds of Brutus, A Storytelling Machine," Lawrence Erlbaum Associates. 2000. [4]Weizenbaum, Joseph. "ELIZA -- A Computer Program for the Study of Natural Language Communication Between Man and Machine." In Communications of the Association for Computing Machinery 9 .1966. [5]Bringsjord, Selmer & Noel, Ron. "Why did evolution engineer consciousness?." In Evolving Consciousness, edited by Gregory Mulhauser. Forthcoming. [6]Phillips, Melanie Anne & Huntley, Chris. "Dramatica - A New Theory of Story,"© 1993, 1995, 1996 Screenplay Systems Inc. [7]Aristotele. "Poetics." Translated by S. H. Butcher, 350 B.C. [8]Weizenbaum, Joseph. "Computer Power and Human Reason." Harmondsworth: Penguin Books. 1984. |
