It's clear that the term cybertext is a modification to the root word text, and it seems reasonable therefore to begin defining the term by asking ‘what is text’. While over time, the word has changed meaning several times, at least when the word cybertext was coined (i.e. contemporary times), text means a coherent set of symbols presented as distinct and identifiable from other sets. That text does not have to be unified or decodable or even meaningful. Some have suggested that a text has to be hierarchical or fit some structure, but that doesn't seem to carry much weight. Relatively recently we have extended the term text to non-letter based media, such as radio plays, movies, performances, and these do not have to be hierarchical. Further, I suspect we are all comfortable calling the Dadaist automatic or non-sense poetry texts, so structure doesn't seem necessary either.
As a symbol set, it belongs to the web of other symbol sets that forms our semiotic landscape. No book stands on its on, but each draws from other books either explicitly or implicitly. Every text is related to other texts. And as readers, the texts are related to us, just as we as people are related to each other.
While we are most familiar with the printed word or other fixations as definitions of texts, partly due to the nature of copyright laws and our publication industries, these texts lack something. They are static. While they belong in the web of relationships that form human existence, they wholly differ from how I relate to you and you relate to me since they are always the same, while we change, adapt, learn, and think. Indeed, one of the important results of the print revolution was the exacting similarity of the text from copy to copy to copy, so that everyone might experience the text in the same way. This ideal was so important us, that many launched quixotic efforts to find the 'authoritative' version of texts.
Conversely, in pre-print culture, copying texts was a highly idiosyncratic process, and therefore highly fluid. Even manuscripts were wholly different from one another, which caused owners to revere each copy as being unique and deserving of special veneration. More oral texts like songs, poems, and speeches would shift continuously as they spread throughout culture, as each reteller might change the story for a variety of reasons, like a mistake or an adaptation. This was a critical aspect of our cultural formation, one that has sadly been taken away due to intellectual property stemming from the printing press.
Cybertext is a move to reinvigorate the fixed texts with the dynamism that we have since lost. The difference is that while before texts used to be changed by people due to the imperfect nature of duplication, in the digital network age copying is more perfect than ever. Rather, since texts are now directly manipulatable by the machines at their smallest component, we make them dynamic by giving them the power of the machine. More to the point, one of the great findings of the Turing Machine was the recognition that machine and text were interchangeable (cf. WhatIsSoftware). Thus, our texts might become machines themselves (i.e. software programs).
It's not enough to say that all software programs were cybertexts. First, not all software programs act as texts. They do not belong to the world of related symbol sets. Depending on our philosophy of meaning, we might include software that only interacts with other software, or we might exclude them. But either way, some software does very little. I think it's important to draw upon Weiner's view of cybernetics as a reactive system, in part because that is where the prefix cyber originated in contemporary language, and in part because it helps enrich the definition of cybertext. A cybertext then would have to be a reactive text. And if a text is a symbol set embedded in the web of relationships that constitute the world, then the text should be reactive to the world; i.e. its environment.
To privilege the concept of text for a moment by limiting it only to what is valuable to humans, the environment of a cybertext then is of course the audience. A cybertext then is a coherent symbol set that can change in response to TheAudience reading it. Built into this notion is that the set remains coherent after the change, so a crashed program does not remain a text as it has fallen apart, broken like a dropped clay tablet. A cybertext must be homeostatic in some sense. And while many texts can change in response to the audience, such as an improvisational play, the difference is that the changes are chosen by a machine.
The common criticism that the machines are built by people, and thus it is the builders who actually make the choice, is disingenuous. Really simple machines make very simplistic choices, but even still relatively simple machines can make choices that are unpredicted by their authors, especially when enmeshed with others (cf. CellularAutomata?, EmergentBehaviour?). Higher generation cybernetic machines will continuously learn from their individual environments and experiences, and use that learning to impact the choices they make earlier. It's not only possible but maybe even common for cybertexts to make choices that diverge from the original author's intention or predictions. Thus, we can say that cybertexts pull the locus of control out of the author's hands to share it with the text as well.
If the locus of control between author and text becomes blurred, there is a third party that has not yet been involved. TheAudience themselves might share control with the cybertext. In CybertextPerspectivesOnErgodicLiterature, Aarseth calls this type of text ergodic, where the reader can make choices that influences how the text is experienced and/or structured. While navigation and arrangement are very common, it's also possible for TheAudience to modify the text itself, changing the content and meaning of the text. In this sense, we normally say that the distinction between TheAudience and TheAuthor have blurred or merged, but it's really TheAudience and the text that have blurred, as TheAudience only has access to the text but not the author. When there is no prima auteur ex machina, but merely a collective of readers/authors mutually creating the text (as we see often in OnlineCommunities), then we have recreated pre-PrintCulture? but with more power (and less cholera).
In short, CyberText is the end of the fixation, which is the death of CopyrightLaw as some have claimed. Rather that all the symbol sets being sterile, static, and self-aggrandizing (i.e. the cult of TheAuthor, the reverence of the book), they can become all dynamic actors in the network of semiotic relationships, all exchanging meanings and changes and stimuli and responses just as we likely did in a solely OralCulture. Except, of course, one of the participants is non-human: it's a machine.
And of course there's the possibility that the language used in the conversation itself could be a machine, that the conversation might act itself in the conversation. A conversation with a conversation? Sure, why not? cf. SelfProgrammingWiki.
Benjamin, W. (1968). The work of art in the age of mechanical reproduction. In H. Arendt (Ed.), Illuminations. (pp. 217–51) New York: Shocken. Available from http://bid.berkeley.edu/bidclass/readings/benjamin.html