Machine Intelligence and Electronic Literature
by: James J. Pulizzi
Computers have been programmed to decrypt enciphered messages, convert text to speech, parse and act upon human speech, generate fonts, and even design patentable electronic circuits.(Note 1) There remains, however, much computers cannot do and may never be able to do without a body.(Note 2) Despite such limitations the computer is an excellent extension of some deficient areas of human cognition, such as sorting large lists, mathematical calculations, memory and retrieval, and error correction. If one combines those skills with human superiority in pattern recognition, adaptability, natural language processing, and the ability to distinguish irrelevant from relevant information, the computer and the human have a potentially powerful symbiotic relationship. The eventual introduction of machine intelligence into works of electronic literature opens many possible areas of cooperation and collaboration between the user, computer, and author in the creation and generation of electronic narratives. My specific focus for elaborating that potential further will be a genre called interactive fiction (IF), which is primarily text based.
Before proceeding with a closer examination of the impact of machine intelligence, some explanation of why I prefer the term machine to artificial intelligence is necessary. The research program commonly referred to as artificial intelligence has been superceded by cognitive science, machine intelligence, and neuroscience. Artificial intelligence began with Alan Turing’s 1950 article on computability and human intelligence, but it was not until the work of Allen Newell and Herbert Simon at RAND around the same time that researchers saw the digital computer not merely as a calculator but as a device able to represent the real world through physical symbol systems in its circuitry. Repeated failures to design a software or hardware architecture using this information processing and physical symbol system theory of the mind led to the eventual repositioning of the field in the disciplines mentioned earlier. Cognitive science has no specific interest in the physical structure of the brain but is instead interested in how thought processes function. Machine intelligence seeks to produce machines—a term used broadly to refer to computers or other relevant hardware—capable of exhibiting intelligent behavior. Neuroscience studies the physical and chemical structure of the brain.
By not attempting to equal human cognitive abilities in all domains, machine intelligence presupposes a complementary relationship between the human and the machine. Interactive fiction programmed with machine intelligence has the ability to engage with the human interactor in a way that makes the word reader a poor description for what human beings and computers could do in the creation of an electronic piece. The intelligent machine can be a partner with the human and integral to the creation of the resulting piece of electronic literature.
A piece of IF called Galatea by Emily Short and another titled Whom the Telling Changed by Aaron A. Reed exemplify some of the current capabilities of computer literature while also pointing to future potential through their limitations. Nick Montfort has published a thorough and provocative study of IF as a distinct genre within the realm of electronic literature, but I will highlight some of the most salient aspects for this discussion.(Note 3) IF constructs a simulated world in which the interactor, or player, navigates and explores through an entirely text-based interface. A parser interprets simple natural language inputs from the interactor, usually in the form verb noun and then passes the encoded information to the program that determines how the action should advance. Several programming languages have been developed for writing IF, but one could also use a high level programming language such as Pascal, C, or C++ to produce a piece.
Just as with print literature, the complexity and quality of the experience varies depending upon the platform used to develop the piece and the imagination of the programmer. The IF work may use a simple branching narrative that presents the interactor with a limited number of choices for how the narrative will develop. Such a structure would result in a constrained narrative, more akin to the set print of a book than what one would expect from a dynamic device like a computer. The IF work may conversely employ a complex and loose collection of narrative elements and characters that make every session through the narrative highly variable and that offers no set conclusion.
It is important to keep in mind that IF unlike conventional printed literature is always a potential narrative.(Note 4) The programmer need not have determined in advance a predefined path through the story. In this respect their literary heritage would be better traced back to the Ouvroir de Littérature Potentielle (OuLiPo) than to the Victorian novel. The story the interactor participates in arises through his or her interaction with the computer program.
Blending aspects of IF and also of hypertext narrative, Whom the Telling Changed places the interactor in the role of a healer from the time and land of the Gilgamesh epic whose village may soon go to war with newcomers to the valley. Your task is to listen to the story being recounted during an annual ritual and to try to persuade the crowd of fellow villagers to seek peace by asking questions or emphasizing specific themes in the inner story. Your adversary has the opposite goal. One can interact by commands of the verb noun formed common to IF or by typing words highlighted in the output text, which is the hypertext aspect of the interface.
The program has been so intricately designed that it is difficult for the interactor to determine exactly the story’s conclusion or how his or her questions and word choices will sway the crowd gathered to hear the inner story. This indeterminacy is not a failure on the author’s part but rather a consequence of the interactor not being the sole force that drives the narrative of the piece. If it is difficult to anticipate future states of the program, it is because the computer is also deciding how the story should proceed given the input from the interactor and other factors that may or may not be related to the interaction with the human player.
Emily Short’s Galatea places the interactor in the role of a visitor to a museum exhibition of artificial life forms where he or she interacts with a female statue. The interactor can touch, look, and interrogate the statue, but the successful player will convince Galatea to turn around and face the interactor. Short has programmed the statue to be rather fickle making any specific line of play to the various conclusions highly variable, so that any one player would have to play hundreds of games before finding a repetition in sequence.
The navigation of this narrative space, as already noted, can occur through a joint effort between the human and machine intelligences. Aaron Reed, the author of Whom the Telling Changed, has already called for a parser able to understand and interpret more complex natural language, and even complete sentences or human speech. The addition of more powerful artificial intelligence to the IF implementation platform, either in hardware or software, would further enhance the capabilities of IF and other genres of computer literature.(Note 5) In their current forms, the IF programs are not necessarily co-authoring the piece in the way a literary critic would find the term meaningful, i.e. making creative and unforeseen decisions about the development of the plot or adding characters and situations not specified by the interactor or programmer.
Such a radically different system of narrative generation would blur the distinction between author, writer, and reader. For a piece of IF or other electronic literature that employs a machine intelligence, one could reasonably ask to what extent the programmer and author of the base elements of the piece actually authors the work. Both the machine intelligence and the human intelligence would input text, or write, parts of the piece—assuming that we are dealing with IF or other text based works, though such writing could apply to multimedia narratives—so one could not ascribe authorship solely to one party or the other. Both machine and human intelligence would be reading the piece. To talk about a solitary author distinct from the reader would have little meaning. Indeed, the very capacity that differentiates it from print literature is this collaboration and cooperation between human and machine that produces the piece. I would call such a narrative cybernetic given its incorporation of human, machine, and text. In this respect it has more affinities with a musical composition or play that while written by a composer or playwright, human beings must perform and a conductor or director must interpret.
Human narrative and machine narrative, rather than competing with one another, mutually enhance one another. Consciousness, language, culture, and narrative rather than being distinct entities caused one by the other exist in a dynamic relationship. With the assistance of computers and the artificial intelligence programs designed for them, it might be possible to imagine a literary narrative that can generate itself in the same way people generate conscious narratives of their lives, through interaction with the environment.
- Dreyfus, Hubert L. What Computers Still Can’t Do: A Critique of Artificial Reason. Cambridge, MA: MIT Press, 1992; 1972.
- Hofstadter, David. Fluid Concepts and Creative Analogies: Computer Models of the Fundamental Mechanisms of Thought. New York: Basic Books, 1995.
- Koza, John. Genetic Programming IV: Routine Human-Competitive Machine Intelligence. Norwell, MA: Kluwer Academic Publishers, 2003.
- Montfort, Nick. Twisty Little Passages: An Approach to Interactive Fiction. Cambridge, MA: MIT Press, 2003.
- Reed, Aaron A. Whom the Telling Changed. Interactive Fiction http://aaronareed.net.
- Short, Emily. Galatea. Interactive Fiction http://aaronareed.net.
- Turing, Alan. “Computing Machinery and Intelligence.” Mind 59 (1950) 433–60.
- Douglas Hofstadter has undertaken significant research in producing computer programs capable of performing human-like tasks in limited domains, i.e. a program that identify items on a dinner table but not on the living room table. For more details on that and the font design program see his monograph Fluid Concepts and Dynamic Analogies. John Koza’s most recent work on genetic programming, Genetic Programming IV, details the genetic programs capable of designing electronic circuits and solving other problems in different domains of research.
- An early but still relevant consideration of the problems with artificial intelligence from a phenomenological perspective is Hubert Dreyfus’s What Computers Still Can’t Do.
- Nick Montfort, Twisty Little Passages: An Approach to Interactive Fiction (Cambridge, MA: MIT Press, 2003).
- Montfort 6.
- See his release notes for Whom the Telling Changed at his home page.