McMahan Game Cultures Final
6/15/16, 10:58 AM
INTERNATIONAL CONFERENCE AT THE WATERSHED MEDIA CENTRE GAMECULTURES BRISTOL 29/30th June & 1st July 2001
THE ROLE OF ARTIFICIAL INTELLIGENCE IN INTERACTIVE FICTION: THE CASE OF THE SIMS Alison McMahan, Ph.D. A division has appeared recently in writing about computer games, a division between scholars who favor narratological tools for analyzing interactive media, especially interactive fiction, and those who prefer to apply game theory. In the eighties academic discussions of interactive fiction focused on hypertext, and to a lesser extent, hypermedias. The emphasis on hypertext led to an emphasis on multiform plot structure as the basis for interactive fiction. For a while – a decade or so – this was a productive approach. Some of the most popular computer games, such as Myst and its sequel, Riven, are graphic hypertexts. Structurally similar games such as Blade Runner, Star Trek Borg, Blair Witch I through IV, used the same remediation approach as Myst and Riven: the idea in these games is to present the user/ player with a multiform plot version of a literary or cinematic text and encourage the user to have not only a point of view on the game but a “point of action” – in other words, to play a part in the story. For someone like me, who as a kid always wanted to jump out of the little boat wending its way through Disneyland’s ride The Pirates of the Caribbean and carouse with the pirates, this approach was perfect. Like most of Disneyland’s rides, these computer games exploit a franchise that already exists in another medium, usually cinema or television. This development was so successful that the cinema and television in turn began to incorporate computer game aesthetics. This might take the form of films imitating the look and level design of games (see for example Luc Besson’s The Fifth Element), or a linear presentation of the multiform structures that are experiences through gameplay in films like Run Lola Run, 12 Monkeys, eXistenZ, and Dead Again, to name just a few. But the very fact that cinema began to remediate (to use the term in the sense that Bolter and
Page 1 of 11
McMahan Game Cultures Final
6/15/16, 10:58 AM
Grusin do) computer game aesthetics lent more force to a different, some say an opposite, approach to interactive fiction design. This is the game theory or ludology approach, an approach that focuses on the kinds of computer games willfully ignored by they hypertext or narratological theorists, the games that remediate not movies or TV shows, but board games, card games, and games of chance. I would include first person shooters and most simulations in this category, first person shooters because they remediate survivalist type games and simulations because they remediate human systems which are based on rules, even if no way to win or lose the game has been specifically defined. What has struck me lately about the division between the narratologists – which is what the hypertext people really are – and the ludologists is that the division between the two approaches has moved from being a fruitful discussion to almost acrimonious. The first sign of this acrimony is the revulsion many designers and seasoned players feel when the word “narrative” is applied to their latest or favorite game. For example, when I approached an interactive media designer for an interview and told him I was writing a narratology of computer games, he indicated that we had very little to talk about because his work, according to him, is not narrative. I pointed out that our disagreement lay not in how we each perceived his work, but in what we understood narrative to be. That got me my interview, but his anti-narrative stance remained firm. By the same token, longtime practitioners of the hypertext theory have two responses to the application of game theory to interactive fiction: one is to say that the hypertext approach covers all the bases addressed by game theory analysis; in other words, the ludologists are re-inventing the wheel. I have to side with the ludologists on this one, as I believe that much of the narratological approach only works because narratologists limit themselves to certain kinds of games: role-playing games (RPGs) and games that are essentially multiform plots with a few puzzles thrown in. In other words, games that can be analyzed as graphic hypertexts. This response clearly seems motivated less by critical factors than by a desire to one, not cede the field to those game theorists, and two, not to have to learn game theory, which surely must involve some kind of math…. A corollary response to game theory from narratologists has been to claim that hypertexts are also games. This latter approach could be interpreted as conciliatory and hopefully will lead to further exchange between narratologists and ludologists, but so far this hasn’t happened. The ludologists, filled with the heady power of having discovered a productive approach, are busy getting on with their work, and are unwilling to work with hypertext people. And why should they? The narratologists limited
Page 2 of 11
McMahan Game Cultures Final
6/15/16, 10:58 AM
themselves willingly to hypertext-like games, such as most of the games based on movies, games which the ludologists don’t even think are worthy of the name. At the same time the narratologists willfully neglected the games the ludologists know for a fact are really interesting: games like Tetris, any first person shooter, and simulations. This gleeful rejection of the narratological approach seems to be especially pronounced in ludologists who used to be narratologists…. The idea I’d like to get across in my talk today is that as long as this division continues both groups are likely to miss the point. We are in danger of getting so focused on our respective approaches that the true nature of our object of study will continue to elude us. By our true object of study I mean what I have been calling interactive fiction and non-fiction, that is, all types of computer games, on-line games, MUDS and MOOs, and interactive television. The danger of becoming blind to our object of study is the same even if we use both approaches, but apply them separately. I don’t claim to have any special insight into the true nature of interactive fiction, but I do have some ideas about where we, academics engaged in the study of interactive fiction, should go from here. Rather than focus on what is different in these two polarized approaches, and rather than having each camp defend their approach by focusing exclusively on one type of computer game over another, I would like to focus on what both types of computer games have in common. I was started on this train of thought by the annoying fact that industry genre categories – Adventure, RPG, simulation, strategy, puzzle, sport, etc, etc, -- are so inadequate for the task they are meant to fulfill. Even the industry has come to recognize this recently and their solution is to label games with two or even more genre labels – “sport/simulation” for example, or “adventure/rpg”. Clearly a new system for describing genres of computer games is needed. I once proposed a system based not on game mis-en-scene elements – which is what the current system is based on – but on game narrative structures. I realized how imperfect the system was even before the essay made it into print. I then tried to use a reception study approach and create a genre system based on how the player’s position is built into the game. This IS a useful approach and I am still pursuing it – if any of you are going to Console-ing passions you can hear me apply this classification system to interactive television in a paper I will give there. But a reception studies approach still doesn’t solve the essential problem of creating a genre system based on the texts themselves. Such a genre system would clearly have to take the type and degree of interactivity of each genre into consideration. But “interactivity” is too vague of a concept, so it can’t be the only, or even the main defining factor.
Page 3 of 11
McMahan Game Cultures Final
6/15/16, 10:58 AM
The approach I am working on now is based on an assumption that all games have some sort of artificial intelligence (AI) as an integral part of their structure, and than an analysis of the AI of markedly different games can lead us to greater insight into the nature of games in particular and interactive fiction in general. Furthermore, I believe that both hypertext theorists and ludologists need to take AI into consideration, so the study of AI can be a discursive bridge where the two approaches can meet. The problem with this approach is that, to be done properly, it requires some knowledge of programming, and that’s knowledge that I don’t have, yet. So for the preliminary analysis that I have relied upon descriptions of AI in different games provided by the game designers themselves, either directly to me or in the press. Until recently, when someone said “artificial intelligence” to me my first mental association was with HAL, the computer of 2001. I guess after this summer most people will associate AI with the Pinnochio character in Steven Spielberg’s film. In other words, when we think of AI, the first thing we think of is robots that simulate human beings in voice and often appearance. Such forms of artificial intelligence dominate the popular imagination and the movies; I call this the “synthespian” approach to AI, using a term coined by James Cameron to refer to digital characters Though it is the most popular form of AI in the public imagination, it is also the least practical. Remko Scha, a Natural Language Processing specialist at the University of Amsterdam, makes fun of this conception of AI with his creation, Huge Harry. Huge Harry lives in a computer and speaks with a digital synthesized voice (speech synthesization is Scha’s specialty), and he has to do a lot of speaking, because he is the representative of the Institute for Artificial Art in Amsterdam, (www.iaaa.nl) and goes around making speeches to educate people on the differences between artificial and human intelligence (for example when he scans the crowd he is addressing he is much more sensitive to the presence of tape recorders and videocameras, machines he can identify with, than he is to people) and educating humans on the higher aesthetic value of paintings and music created by computers over that of humans. One of Huge Harry’s favorite speech topics is the superiority of computer interfaces over the human face as a means of expression. To demonstrate this he uses a human (usually Arthur Elsenaar) as his avatar, by attaching electrodes to the human’s face and then making the human’s facial muscles twitch to form different expressions. The laboriousness of this task shows once and for all, at least to Harry’s satisfaction, that computer interfaces are superior to human ones and that the search for the perfect
Page 4 of 11
McMahan Game Cultures Final
6/15/16, 10:58 AM
synthespian or humanoid form of AI is a waste of time. Even if we give up on the idea of a human form, we still want the AI to communicate to us like a person. That leads to AIs like ELIZA, the bot programmed to ask questions in the pattern of a Rogerian therapist. ELIZA was a computer program written in 1966 by Joseph Weizenbaum. This is Marie-Laure Ryan’s assessment of ELIZA’s significance: The relevant feature of ELIZA is that it was able to carry on a reasonably coherent conversation without using any sophisticated language parsing techniques. Rather than building a syntactic and semantic representation of the user’s input (as do programs seriously aiming at language understanding), it relied on rather crude patternmatching strategies, such as detecting key words and responding with canned formulae, recycling the user’s input, changing pronouns from first to second person, and, most importantly, answering almost every input with a question of its own. Yet despite the system’s total lack of understanding of the human mind, ELIZA’s conversation was clever enough to fascinate users. Many people turned to ELIZA as a help toward self-understanding. The strategy of answering questions with questions was ideally suited to the context of the psychotherapeutic exchange, since the role of the therapist is to help the patient produce his own analysis, rather than to impose an interpretation. But the truly important legacy of ELIZA to Interactive Drama was her ability to promote a ludic attitude. It did not matter that ELIZA did not understand a conversation, as long as she could fake understanding. To the user willing to play a game of make-believe with the computer, ELIZA was the perfect prop. Synthespians – synthetic actors – depend on their seeming intelligence because they can incorporate the pronouncements of the user in a way that makes sense. In other words, the machine’s apparent intelligence is really its ability to replay the human intelligence. Another digital character, Keiko Suzuki, took this principle to an extreme. Suzuki was an avatar that absorbed the e-mails of the members of the 7-11 mailing list and re-printed them after filtering them through her own personality. Ironically, the inconsistency of her pronouncements – since they originated from different people – made her seem more human. What started out as a whimsical “oracle” device to liven up a listserver developed into a digital persona, complete with erotic websites designed by her fans. (Unfortunately Keiko Suzuki’s career as oracle ended when the 7-11 company objected to the use of their trademarked name for such “off-color”
Page 5 of 11
McMahan Game Cultures Final
6/15/16, 10:58 AM
discourse.) Suzuki and Eliza both demonstrate the limitations of the synthespian approach to AI. Stephen Grand, the creator of the computer game Creatures, came to the same conclusion: that the goal of AI research should not be to simulate life, as Eliza does, but to duplicate it. He created Norns, the creatures that inhabit Albia, the virtual world of the Creatures series, first released in 1996. Players rear their norn and help it through various encounters. Grand programmed his norns with over 300 genes and simple drives to satisfy, such as the drives to eat, breed, and avoid pain. Most importantly, norns can learn. Productive skills are reinforced, unused ones fade away. This kind of design characterizes most virtual pets, but Grand also designed a “biochemistry” for his norns that behave much as hormones and neurotransmitters do in the human body. Players can influence the norns’ learning pattern by doling out rewards and punishments, and when norns breed, the “DNA” patterns are mixed, often with unexpected results. One of Grand’s surprises was to find two of his norns playing ball with each other, when he not programmed them to play together. The idea that the norns can learn is profoundly shocking. As Norbert Wiener put it, “Learning is a property that we often attribute exclusively to self-conscious systems. It is a phenomenon that occurs in its most characteristic form in Man,”. As Wiener predicted back in the early 60s, not only would intelligent machines learn, what they would learn would be games and game-like systems, because games are subject to a clear-cut objective criterion of merit and the rules of play are simple and inexorable. In other words, easily programmable. Wiener was one of the first to point out that war and business were conflicts that resembled games (hence the use of texts like Sun Tzu’s The Art of War and Musashi’s Book of Five Rings as game manuals for business.) He warned that one of the first uses AI would be put to would be to the waging of war, and his book, God and Golem, Inc, was a call for the need to remind ourselves that this isn’t just a game, it is also life. This brings me back to the norns of Albia. The norn-bots fit Wiener’s definition of artificial intelligence, because they can learn how to play games simply by playing games. They are so successful at learning through doing, in fact, that the British military is now using the same software to invent adversaries for pilots flying missions in flight simulators. Research is already being conducted to see if the virtual pilots could replace human pilots altogether. The June 2001 issue of Popular Science had a cover article on this type of pilotless bomber jet, which are expected to be in military use as soon as 2010. The idea that of a bot like ELIZA flying a bomber jet that fits into a
Page 6 of 11
McMahan Game Cultures Final
6/15/16, 10:58 AM
crate but can destroy a line of army tanks on the ground makes me shudder. The article in Popular Science promises that human operators on the ground will control the automated bomber jets and be able to override them if they fix on the wrong target, for example. But I have seen enough movies about robots run amuck to be highly skeptical of this scenario. This is exactly the kind of thing Norbert Wiener was referring to back in the early sixties, when he said that the only solution would be to always combine a machine intelligence with a human one, in other words, to combine an intelligence based on game rules with an ethical intelligence. He believed that the religious nature of humans would rein in the game-play nature of machine intelligence. This last point seems to have been eviscerated by his very proof for AI. If God made humans in his own image, in the sense that God is intelligent and humans are intelligent, i.e.,we can learn, and now humans have made machines that can learn, and in addition, as Wiener and the norns of Albia have demonstrated, we can makes machines that can make other machines in their own image (not pictorial images, but operative images), then have we become more like God (that was Wiener’s assumption) or has God become more like a machine? Let me put this another way: the basis for predicate logic, the logic most used for AI program languages such as Prolog, is the algorithm. An algorithm, “is a finite procedure, written in a fixed symbolic vocabulary, governed by precise instructions, moving in discrete steps… whose execution requires no insight, cleverness, intuition, intelligence or perspicuity, and that sooner or later comes to an end.” Just as an algorithm is the formula that can enable a machine to take variables into account and react accordingly but logically every time, a simulation takes a series of algorithms to simulate a whole system – like a game, or dropping bombs on tanks. This sounds very mechanical and mathematical, but that’s not how it felt the first time I saw a documentary (I forget the title) about a cybernetic engineer at MIT? Who designed insectlike robots. The robot-insects only had to avoid crashing into walls and each other and go to get power before they ran out. They were run by a total of six algorithms, and yet such a short list of simple commands made them seem completely insect-like, at least to my untrained eye. And I was overcome by the feeling that if it was that easy to duplicate an insectintelligence, then it couldn’t possibly be that much harder to duplicate me, the human intelligence. In other words, what if there is no God and we have no souls at all, or rather, our souls are really composed of a series of algorithms, of mathematical formulas? Just as I never thought I would feel that way about my soul, I never
Page 7 of 11
McMahan Game Cultures Final
6/15/16, 10:58 AM
thought I would feel that way about narrative. Anyone who teaches screenwriting, as I do, knows that the heart of a screenplay is the human effort to work through a very human problem. Certain elements of drama have been described in almost the same terms since Aristotle wrote his Poetics two thousand years ago. Terms like “epiphany” remind us of the religious origins of most drama. And just as it has been impossible, to date, to create a real human-like machine intelligence, it has also been impossible to create a machine that generates what we would recognize as narrative. However, we do have the next best thing, what Aespen Aarseth has called “cybertexts”, machines that produce signs, which vary from reading to reading while still maintining a continuous coherence. Of course, hypertext falls into the category of cybertext, so I have now succeeded in putting games based on hypertext structure and games based on gamesystems on a similar footing. At their core, both types of games are characterized by an algorithmic logic; though extremely different, they are both machines that reproduce some kind of human system, whether it be a narrative or a game. Until Will Wright’s game The Sims appeared I thought that the principle use of AI in a game was to create characters. In other words, I was still making the error that Remko Scha ridicules via Huge Harry: I was still anthropomorphisizing artificial intelligence. I still wanted the Bicentennial Man or Commander Data to keep me company, even in the forms of speechless norns. One of the discoveries of Grand’s research with the norns is that a rich environment, one that constantly changes, is crucial to speed up the development of virtual character intelligence. This insight has been applied to The Sims, a game developed by the same company that brought us SimCity. Here the virtual characters look like real people, with family feuds and TV sets. The player’s principle objective is to provide a stimulating environment for their Sim characters, which includes building a house, getting toys, and putting in other characters. Lone Sim characters will often pine away and go broke or even die! It would be impossible to do a cultural analysis – or a level of narration analysis -- of the Sims without taking the AI into account. For example, although the player influences the Sim environment he doesn’t have that much control over the Sim characters, who often shake their head “no” if their programming impels them in a direction other than where you want them to go. How are these characters programmed? There are five characteristics for each sim character – Neat, Outgoing, Active, Playful, and Nice. The player starts out with a set number of points that can be distributed among
Page 8 of 11
McMahan Game Cultures Final
6/15/16, 10:58 AM
the various qualities, and depending on how you distribute the points, the character is given a sign from the zodiac as a personality – Aries, or Leo, or Gemini, or whatever. Will Wright originally called the game “Dollhouse” because the emphasis for him was on designing the home environment, using the same software engine that was used for his SimCity. This emphasis on environment is still there, but The Sims also represents a shift in AI design. In this game, it is not only the characters (and the user) who are intelligent but also the environment. The objects in the environment such as chairs, refrigerators, and bathtubs are programmed to “advertise” their ability and the degree to which they can satisfy those needs, thus acting as a magnet drawing the needy character. For example, if the Sims character hygene is at a low ebb, and it walks past the bathtub, the bathtub will scream at him “I can satisfy hygiene seven”. If the character is also hungry and walks past the fridge, the fridge will scream “I can satisfy hunger four!”. This can make it very difficult for the character to do what you just told it to do. In other words, it is not only the Sim character that has artificial intelligence, but the objects in the environment as well. EA is working on a multi-player online version of The Sims which will commence in 2002. The Sims on-line will take some cues from Ultima and Everquest, in the sense that players will have up to 20 Sim neighborhoods in which they can “live”, and the neighborhoods or cities will vary by type, like western or sci-fi. Cooperation will be the key to success in this on-line environment, that is, your sim will have to cooperate with someone else’s sim who is your roommate or work colleague. The more contact your sim has with other people’s sims the more likely both are to become rich, famous, or both. There will be certain limitations – sex is limited to kissing – enforced by game masters who will keep things in order the same way wizards do in MUDs. As John Hopson has noted in his article on “Behavioral Game Design”, it is not only the machine intelligence that can be programmed, but the human one. According to Hopson, behavioral psychologists have discovered that there are general “rules” for learning and how human minds – Hopson claims that these rules apply to all animals from fish to birds – respond to their environment. He points out that if these insights of behavioral psychology were more systematically applied to game design, games would be better. In other words, he is talking about how the game programs the player, with things like rewards (reinforcers), contingencies (the rules of the environment), the way the game responds to the player and how long it takes it to respond, whether it responds the same way each time, etc. In other words, at least according to Hopson, algorithms can be
Page 9 of 11
McMahan Game Cultures Final
6/15/16, 10:58 AM
designed that humans can be predictably relied upon to follow. Open-ended interactive fictions such as the planned Sims online or the average MUD give the user more options that simulations do, options such as adding to the world or making adjustments to their character, but these adjustments still occur with a limited framework with certain rules. Though such worlds are singularly lacking in traditional forms of closure, the goal of interactive and non-interactive fiction is the same: to provide the user/ spectator with an experience that leads to an epiphany or a revelation, or a series of epiphanies and revelations. In other words, the way the game programs the player ultimately best be analyzed using narratological tools. For example, in the early simulation game Hidden Agenda, the goal of the player, who was playing the role of president in a banana republic, was to stay in office – and alive – for a three year term. The only way to accomplish this was to balance extreme right wing and extreme left wing elements – appointing representatives of both sides to the cabinet, for example – while also keeping the economy running in a satisfactory manner (handling hunger strikes peacefully). Once the player realizes that this is the “theme” of the game, closure can be said to be achieved, even if the game itself has not been completed. The stated goal of the game’s designer, Jim Gasperini, was to use the interactive medium, which he referred to as a new species of drama, to enlarge or alter their own cultural perspective. Gasperini had his own, slightly left of center political agenda, which can be analyzed at the narrative level only once the way the AI of the game has been understood, at least at the level of play. In addition to the time and the left-wing ideology of the game, what is interesting about this game is that it is all played by diplomacy. There is no shooting, you are never given a choice of weapons, etc. Instead you find yourself in a series of confrontations with different characters and you are forced to come to terms with their political perspectives. You are forced to empathize with the different denizens of a third world country, which Gasparini thought was a necessary experience for Americans to have, especially in 1985. Hidden Agenda’s interface seems very outdated to us now, with its black and white line drawings, huge pixel graphics, and minimal use of cinematics, but in terms of its conceptual design it is still worth studying, as it has not been improved upon by its descendants such as Civilization, Age of Empires, Populous, and so on. Hidden Agenda demonstrates my point: that in order to truly understand a game, we need to understand how its AI is designed, which means resorting to game theory; but in the end, the final algorithm to be analyzed is the one that programs, or attempts to program, the human
Page 10 of 11
McMahan Game Cultures Final
6/15/16, 10:58 AM
intelligence on the other side of the games interface. This ultimate analysis is one best carried out with narratological tools, as narrative has taken over the functions that Wiener once attributed to religion (this is why porn reform advocates recommend adding narrative to porn as a way of improving it: if it is narrative, it must be moral). Don’t misunderstand me: I am not advocating a return to the narratology of Aristotle’s Poetics. Narrative, like every other form of artificial intelligence, is in a constant process of flux and change, and interactive fiction calls on us to redefine what narrative is. Thank you. Alison McMahan
Page 11 of 11
