Chatbots and Literature

As a literary scholar, I am interested in how writers and critics have responded to phenomena and what literary criticism might bring to debates around cultural understandings of chatbots. Reference to George Bernard Shaw and Ovid in my discussions of the history of chatbots hints at the fact that writers have indeed developed conversational agents – whether in their fiction or real life.

One of the earliest modern chatbots (along with ELIZA and PARRY) was Racter (short for Raconteur). Invented by William Chamberlain and Thomas Etter, the program used a very simple syntax, allowing users to input questions, to which Racter would respond with long, nonsensical (if grammatically correct) paragraphs. Although purportedly enabling a conversation, as one critic noted, “It was unclear from the start who was interviewing whom,” indeed, “As computers move ever closer to artificial intelligence, Racter is on the edge of artificial insanity.”¹

In addition to releasing Racter, Chamberlain published The Policeman’s Beard Is Half Constructed (1983). Purportedly, “With the exception of this introduction, the writing in this book was all done by a computer. The book has been proofread for spelling but otherwise is completely unedited.” The introduction further imagines the communicative potential of such a program: while computers are “supposed to compute” and Chamberlain thus ponders the question of “Why have a computer talk endlessly and in perfect English about nothing?” he ultimately suggests that this program offers the potential for “the production of prose that is in no way contingent upon human experience.” For Chamberlain such computer poetry, dialogue, and prose offers “formal communication” of “the Mind of a Machine.”² The volume received acclaim, despite the fact that an editor’s hand was likely involved, indicative less of the technical success than the book’s positioning of itself within a legacy of Surrealist experimentation and the expanding horizons of aesthetic experimentation with computers for communicative purposes.³

In my own work I talk at length about two other examples of literary experiments with conversational and interactive agents – those by novelist J. M. Coetzee and linguist Margaret Masterman. Both individuals experiment with computational poetry writing as a means of reflecting on literature’s role as a cultural interface.

Meanwhile, the Electronic Literature Organization (ELO)⁴ has curated a number of examples of early and notable born-digital “bot” literature. They anthologize one of the most fully formed examples, Galatea by Emily Short, an interactive fiction (IF) writer. A piece of IF which turns on conversation, Galatea won the 2000 IF Art Show and has been well received by scholars. Galatea is a statue come to life and with whom the user can converse. While conversation is restricted, the stories that unfold are the result of topic choice, the progress of the dialogue, and Galatea’s “mood” – recalling the classical roots, Galatea explores the relation between creator and creation. Another piece of notable chatbot fiction curated by the ELO is Michael Mateas and Andrew Stern’s 2005 piece Façade. Set in an apartment with the user as a dinner guest, the piece utilizes NLP of user input to direct interaction (and micro-actions) with the increasingly hostile hosts.

There are also a growing number of chatbots which present literary avatars.⁵ Willbot, by Existor (Cleverbot’s developers), presents users with the opportunity to talk with Shakespeare. The company Fastbot meanwhile offers various character and “Author bots,” including Wonderbot, an Alice in Wonderland chatbot. Via such interfaces, “Readers can find out about plot, characters, backstory and access exclusive content through messaging.”⁶

Media providers are exploring the possibilities of utilizing chatbots in the delivery of content. Building on the recent trend in marketing to use chatbots as a means by which to develop brand identification in users, the British Broadcasting Corporation (BBC) and other media institutions have experimented with the aesthetic possibilities of such interfaces.

Since 2016, the BBC’s Research and Development team have been running the “Talking with Machines” project which asks “How do you design an interface for a device with no screen?” Led by Senior Producer and Creative Technologist Henry Cooke, the project has been created to explore “the potential of devices with conversational and spoken interfaces.”⁷ Given the popularity of spoken interfaces like chatbots, the project notes that:

To date the project has experimented with a number of different pilots.

One of the major BBC pilots in this project to date is called The Inspection Chamber (2017–19). Produced in collaboration with Rosina Sound, it offered users the chance to engage in a piece of interactive voice storytelling delivered through Amazon Echo. The scenario, a “science fiction comedy story inspired by Douglas Adams, Franz Kafka and Portal” involves “a glitch-ridden computer, two desperately homesick scientists, and one thing in their way, you – an unidentified ‘being’ that just can’t be classified.”⁹ On launching the program the user is asked a number of questions by the scientists, and the plot develops in response. Recalling interactive fiction of the 1980s and 1990s (although Black Mirror offered a recent televisual update with its Netflix episode “Bandersnatch”), The Inspection Chamber experimented with providing an audio offering – what they have nicknamed “conversational radio.”¹⁰ As part of the pilot, participants were interviewed about their experiences, with a summary of the findings and recommendations made available to the public. The conclusions are intriguing – for example, while users generally enjoyed the format and thought it had potential, many found the sci-fi genre off-putting. Users responded well to audio cues within the production as to when they should speak, but would have liked more personalization and to have more impact on the story (they also enjoyed talking!).¹¹

Other pilot projects have an even more explicitly experimental and literary orientation, recalling the BBC’s historic contributions to experimental radiophonic art. One project aimed to provide a new interface for an extant piece of radio drama. In 2010 BBC Radio 3 dramatized B. S. Johnson’s 1969 experimental novel The Unfortunates. The novel consisted of twenty-seven unordered sections delivered to the reader in a box. When adapted for radio by Graham White, the combinatorial format was retained during production, with the final order only determined live on air (but effectively “frozen” thereafter). In 2018 “Talking with Machines” edited that original broadcast to produce separate recordings of the original sections. A playlist was designed to randomize the order of sections, and a voice interface enabled listeners to jump between this and additional content (including a radio documentary on 1960s avant-garde literature and a making-of discussion about the 2010 production).¹² The BBC’s “Talking with Machines” project builds on a long tradition of experimentalism in the arts – in particular in sound and literature – when exploring the aesthetic potential of conversational agents.

One other intriguing example consists of an out of hours project by BBC producer Henry Cooke and his collaborator Tim Cowlishaw, which deploys two Amazon Echo speakers to produce sound art by conversing.¹³ Invoking mid-century OULIPO and Surrealist experiments with dialogue¹⁴ (along with the BBC’s own Third Programme activities), the pilot was inspired by Alvin Lucier's 1969 tape recorder piece I am sitting in a room. As Cooke explains:

So, our loose idea was to have two voice devices repeat Lucier’s original text to each other, amplifying and compounding small errors in their speech-to-text systems by speaking the text out loud and repeating what they’d transcribed. The quirks and biases of the complex natural language processing systems at work in these devices would be revealed by the errors that crept in through repetition, the byzantine pathways of their networked operations replacing the Lucier’s original room and tape decks.¹⁵

Although the eventual piece deployed a new script to respond to the Echo software’s tendency to remove error, it does offer a fascinating example of mainstream software utilized for aesthetic, Surrealist purposes. You can view a video of the piece at https://www.youtube.com/watch?v=TGmrltCuTZQ.

Projects such as those instigated by the BBC demonstrate the ways in which writers and producers might respond to the new talk interfaces and fuel innovative uses of these technologies. Within the computing industry, these contributions are increasingly identified as vital to the future development of AI.

In recent years the skills that writers, artists, and videogame designers can bring to chatbot design has been recognized. A number of companies now offer developers tools to create “compelling characters” and tell gripping stories through talk interfaces. SpiritAI for example offers the Character Engine which helps designers “create distinctive personalities who will show emotion and respond dynamically to the player – yet still hit the key beats of your story.”¹⁶ A product managed by none other than Emily Short, Character Engine draws on experience from IF heritage and the creative arts, “making AI more human.”¹⁷

More generally, there has been a growing awareness in the computer industry that the arts and humanities offer potentially valuable expertise and knowledge for the development of AI. This heartening erosion of the “two cultures” mantra can be seen in more than discussions of good storytelling. Steve Jobs was one of the more explicit voices, arguing in 2011 that “It is in Apple’s DNA that technology alone is not enough—it’s technology married with liberal arts, married with the humanities, that yields us the results that make our heart sing.”¹⁸ However his position has been slow to establish.

In 2018 Microsoft released the book The Future Computed: Artificial Intelligence and Its Role in Society with much fanfare. Contemplating the future of AI, the book offers a number of familiar (if important) platitudes concerning the status of the human worker in this future: “Many jobs will continue to require uniquely human skills that AI and machines cannot replicate, such as creativity, collaboration, abstract and systems thinking, complex communication, and the ability to work in diverse environments.”¹⁹ It is also notable, however, that the book makes tentative moves towards proclaiming the value of disciplines it has long-denigrated: “As automation and AI take on tasks that require thinking and judgement, it will become increasingly important to train people — perhaps through a renewed focus on the humanities — to develop their critical thinking, creativity, empathy, and reasoning.”²⁰ Conditional and in parenthesis though it might be, it is a step.

The book’s foreword offers a message of slightly more fulsome support for the arts. Written by Brad Smith, President of Microsoft, and Harry Shum, Executive VP of AI and Research, the foreword was also picked up extensively by industry news commentators and widely quoted. Their initial mention of the arts is again framed in the rather familiar trope that recalls IBM’s “THINK” promise of seventy years prior: “Before long, many mundane and repetitive tasks will be handled automatically by AI, freeing us to devote our time and energy to more productive and creative endeavors.”²¹ However, Smith and Shum go on to offer a more explicit suggestion for the future of AI:

This pronouncement received a lot of attention in the IT and business industries; headlines such as “Microsoft’s president says liberal arts majors are necessary for the future of tech” proliferated.²³

We can also see moves within the tech industry to engage with those trained in the arts and humanities in order to develop AI technologies. IBM’s Watson blog has promoted “Fuzzy or Techie?! Why AI needs more interdisciplinary thinkers,” an IBM Think Leaders podcast that discusses the importance of liberal arts perspectives for developing AI.²⁴ While some might protest the label “fuzzy,” L. A. Zadeh’s notion of the “fuzzy logic” pivotal to computing with words has at least reclaimed some of the status of the term.²⁵ It is also notable that companies such as Microsoft, Google Creative Lab, and Intel have worked with playwrights, poets, and novelists in their development of AI. In particular, it is in the development of conversational agents—Alexa, Siri, and Cortana, amongst others—that literary authors are considered to be invaluable in their development of characters that are able to respond appropriately to conversation, display culturally sensitivity and a personality that has human characteristics, while still acknowledging its AI nature.^26,27 The answers that a chatbot will give to questions such as preference for presidential candidates, or favorite movie, and the number of exclamation marks, ums, or emojis it uses, become important expressions of the bot’s character and thus the willingness of the user to deploy the interface. Just as SpiritAI explicitly presents itself as offering support for bot engineering via creative writing expertise, so too the chatbot industry is increasingly recognizing the expertise that literary authors can bring to the design of conversational agents.²⁸

Nevertheless, while being valuable, these workers are also, apparently, far from indispensable. Google’s mass sacking of temporary workers employed on Google Assistant’s “personality” team in Spring 2019 drew attention to the humans behind these conversational agents. As the team announced in an open letter, “We’re responsible for the voice of Google — the Google Assistant — across the world. We are the human labor that makes the Assistant relevant, funny, and relatable in more than 50 languages.”²⁹ Personality might be an important aspect of user engagement, but this has yet to translate into job security for those producing personality. Far from technology freeing us up to perform creative tasks, it often works to elide the human labor behind computational assistance.

Where to next?

A Literary Guide to Natural Language Processing

Chatbots

Chatbot Lives

How to Read a Chatbot

Talking Interfaces

Literature, Talk, and Computing