Dreaming Cognitive Assemblages: Artificial Intelligence and Media Shifts in Robot Narratives from Philip K. Dick to Annalee Newitz

In science policy conversations, and lately in the popular press, controversy swirls around the question of just how smart and independent Artificial Intelligence can ever be. In these conversations, a strong “Critical AI” faction argues with renewed vigor that claims for AI’s autonomy and value are overly hyped. Still, popular narratives continue to insist that we will soon see autonomous AIs whose cognitive abilities rival or surpass those of humans. What explains this gap between AI fact and popular AI robot fiction? At one time, we might have explained such a gap between actual technological capacity and possible technological achievement simply by gesturing toward the notion that the science fiction genre exists to offer prospective views of probable or at least possible futures. Yet while new narratives about humanoid robots may still seem like visionary glimpses of possible futures, those futures seem more tenuous than ever. (2) In a culture saturated with dystopian renderings of ecological and political catastrophe, the very ground for technological achievement seems something to be hoped for, not to be taken as given. And while we’ve seen tremendous advances in recent years, months, and, one might even say, days, in such areas of AI as natural language processing, we’ve seen such success nearly outweighed by well-publicized failures to advance substantially in other areas. Truly autonomous vehicles remain to be developed, much less deployed, while the latest reporting finds robotic delivery services stymied in their major test beds. More generally, we inhabit a climate of cynicism about scientific progress, and feel nostalgia for, or embarrassment about, past visions of the future that now seem unrealizable. If the virtual AIs we create in our fictions can’t be explained away as extrapolations from present-day trends, what kinds of expectations, hopes, or wishes do they represent? What kinds of fantasies are these?

As a first, limited, and speculative step toward describing one species of robot narrative and identifying the stakes of taking these stories seriously, in this paper we take a media studies approach to the question of how embodied, autonomous AIs are represented, an approach that enables us to posit some patterns that hold across representations of AI, whether their mode be realistic, speculative, or fantastical. Whatever their mode, we propose, such representations typically depict AIs as what Katherine Hayles “cognitive assemblages”: as plural entities that process information through socio-technical interaction. We suggest that this is an accurate way to represent some actual emergent AIs, although others may differ on this question. Whatever one’s stance on the appropriateness of this description to emergent AIs, however, we argue that their representation as cognitive assemblages actually aligns AIs with traditional information tools such as books and films, which likewise unfold thought in unpredictable ways, leveraging sociotechnical disjunctions between authors, audiences, and media artifacts. Might the futurity of AIs be more continuous than we have previously thought with the history of the media imaginaries that remain, after all, the domain in which such AIs have been most fully realized? Exploring this intuition, we locate case studies of cognitive assemblages in two novels--Autonomous, by Annalee Newitz, and Kazuo Ishiguro’s Klara and the Sun--as well as in the supertext encompassing Philip K. Dick’s short story “Autofac” and its recent adaptation for Amazon Prime’s Bladerunner-flavored science fiction series “Electric Dreams.” This illustration is from the original, Galaxy Science Fiction publication of “Autofac”; its full caption reads, “Naturally, man should want to stand on his own two feet … but how can he when his own machines cut the ground out from under him?”

What gets worked through, perpetuated, or overthrown by stories about humans and robots? Many say these stories recount the history of human relations with other humans, with nature (including non-human animals and the environment); and with extant technologies. We emphasize how these stories work through media history and specifically through what we consider to be that cognitive assemblage practice called “the book.” Moreover, we suggest, stories from the media era dominated by the book dream, specifically of robots, because within the heterogeneous, discontinuous, fluid universe of cognitive assemblage practices, books and robots are categories of assemblage that share a particular relation to autonomy and commodity. It may even be that the era of the book and the era of the robot represent comparably equilibrated sociotechnical regimes. To see how this might be so, we want to look not just at the media shift from print to postprint, from books toward robots, but also at how books and robots alike understand each others’ domains.

The balance of this presentation has three parts. First, we briefly describe how these speculations about book/robot relations have emerged for us from within a broader program of empirical research; second, we contextualize them within a framework of the media history of print; finally, we illustrate them with a few brief examples from imaginary robot literature. So, then, first to Book/Robot relations as seen from the POV of a web of sociotechnical artificial intelligence research programs at the University of Texas.

Hannah and I participate in a web of sociotechnical artificial intelligence research programs underway at the University of Texas at Austin, many of which are connected up in various ways under the auspices of the Good Systems Texas Grand Challenge program, which, like many such programs around the country and the world, aims to better understand what changes new technologies will bring, predict how those changes will unfold, and mitigate the harms or unintended consequences they could cause while still leveraging the benefits AI provides. What we hope makes our program distinctive is how thoroughly it integrates computer scientists and engineers with social scientists and humanists, as for instance in the Living and Working with Robots core research project I help run, along with Elliott Hauser of our School of Information, which has just won funding from the NSF Growing Convergence Research program. While most subprojects involve AI application areas, other subprojects have focused on posing questions about the representation of AI in and to the public, and this present paper grows out of one of those, an investigation called “Bad AI and Beyond: Exploring How Popular Media Shape the Perceived Opportunities and Threats of AI,” which I led with the fan studies scholar Suzanne Scott from the Radio-Televion-Film department and with the game design studies maven Paul Toprac from Computer Science. As a member of this group, Hannah helped organize a creative arts and criticism festival that culminated in a crossover event with the Rapoport Center for Human Rights and Justice on our campus exploring Science Fiction and the Future of Work, which featured our colleague Simone Browne and our alum Bruce Sterling.

The Bad AI and Beyond project accumulated a great deal of focus group and survey data about conceptions of AI in various subcultures and among the general public, data which a philosophy graduate student, Karim Nader took the lead in analyzing and writing up into an article in *AI & Society*. The questions we posed in those surveys, and thus the results we reported in our article, “Public Understanding of Artificial Intelligence through Entertainment Media,” mainly concerned popular films and video games. We remained, however, curious about—and our “reviewer number two” highlighted the need to address—the obvious further question: What about books? While we were able to dodge our reviewer’s question by arguing for the representative quality of the increasingly dominant media we did study, we saw the necessity of exploring how books—perhaps especially, but not exclusively, fictional entertainments—likewise drive and respond to public understandings of AI, and perhaps have a special importance, given how they inspire, or even provide blueprints to, AI designers. Some such explorations have been undertaken, of course, but the need we perceived for more such study related to our sense that this present paper’s more fundamental question requires attention: What are books in, and in relation to, the computational era? What do our books tell us about the age of intelligent machines?

So now on to recontextualizing Book/Robot relations. How might the interaction of Books and Robots appear from the point of view afforded by the efforts of Media History to reposition Book History within the broader story told by that field, and within the problematic that has developed as Media History has been transformed by the experience of digital media? In part, our goal here is to remember all that we’ve learned about the interplay between print media, visual media, and digital media as we think about how autonomous intelligent machines get represented. But we also seek, again, to look not just at the media shift from print to postprint, from books toward robots, but also at how books and robots alike understand each others’ domains, paraphrasing W.J.T. Mitchell’s great question about pictures to ask, What do these media want?

As we seek to frame up this question, we draw rich resources from the new, digital-media-informed media history of the book. Landmark studies we hope to make matter for us in this enterprise include Lisa Gitelman’s Paper Knowledge and Always Already New, Matt Kirschenbaum’s Bitstreams and Track Changes, and Katherine Hayles’s Postprint and How we Think, among other works. Of these studies, it is the work of Hayles, and in particular her flagship concept of cognitive assemblages, that is the most avowedly, as her title puts it, “postprint.” Yet, as our title reveals, it is this work and this concept that we think may hold especially exciting resources for rethinking book history in relation to digital media history. In the way of other “posts”—the postmodern, the postcolonial—might it also be true of the postprint condition that it resides latent in the prior era it names, and carries that era along with it as the residual element whose reemergence it periodically stages?

Here are some ideas from Hayles that we find provocative in this regard “Interpretations and meaning-making practices,” Hayles writes, “circulate through transindividual collectivities created by fluctuating and dynamic interconnections between humans and computational media,” the interconnections that she calls “cognitive assemblages” (8). In what Hayles calls “cybersymbiosis,” “Intention is preempted before it can begin to operate on a conscious level,” and “human being is so entangled with computational media that human intentions and desires literally cannot be thought without them.” This entanglement, Hayles avers, is so intense that while “the changes wrought by the invention of print were also immense … Nevertheless, if the focus is specifically on the cognitive dimension of the relevant changes … the differences between print and postprint are not merely quantitative but qualitative” (168). We find this model compelling. Yet we can’t help but want to extend Hayles’s postprint arguments back to reincorporate print culture and better explain it as well.

What if, we ask—most controversially, we imagine—Interpretations and meaning-making practices also circulate through transindividual collectivities created by fluctuating and dynamic interconnections between humans and linguistic media, which are likewise cognitive assemblages? What if “cybersymbiosis” is preceded by, and then accompanied by, quixoticism, where intention is preempted before it can begin to operate on a conscious level, when human being is so entangled with reading matter that human intentions and desires literally cannot be thought without them? What if the changes wrought by the invention of writing were immense because they contain, immanently, the qualitative shift made apparent by computational media: the possibility of dependence on distributed cognition with external, transpersonal, encoded storage, i.e. autonomy, or even commodity?

Finally, then, to our case studies of books dreaming robots. On such grounds that we propose a further, mutual imbrication of print and postprint culture than has been imagined. Leaving aside for now these problems of historical modeling, we want in conclusion to suggest what this inter-imbrication of print and postprint could mean for understanding how books continue dreaming robots. Here, as mentioned, we have three case studies in view: two novels--Autonomous, by Annalee Newitz, and Kazuo Ishiguro’s Klara and the Sun--as well as the supertext encompassing Philip K. Dick’s short story “Autofac” and its recent adaptation for Amazon Prime’s Bladerunner-lavored science fiction series “Electric Dreams.” Talking briefly about these works in turn, I will now go into literary critic mode and cease flashing any new slides at you all.

Of these works, it is Autonomous that most clearly charts out the conceptual terrain that a postprint understanding of book/robot relations will need to traverse—unsurprisingly, since Newitz, in addition to being a literary artist, is an accomplished cultural critic and futurist. Newitz imagines a dystopian near future in which Jack, a piratical merchant who deals smart drugs, strives to free herself from the constraints of her trade and her past political embroilments, while the policeman and police robot team chasing her slowly lose their relative autonomy by—spoiler alert!—falling in love with one another. Literature here again finds its time-honored correlates, sex and drugs, but this time, with robots fully in the mix in ways that reveal how what we dream about, when when we dream about robot autonomy, might have more to do than we think with love, altered states, and literary experience.

In Klara and the Sun, Kazuo Isiguro takes a different, formal approach to thematizing autonomy, by crafting a first-person narration from the point of view of Klara, a companion bot designed to teach and comfort children. The tragic arc along which Klara loves and loses Josie, the child who chooses her, recalls those of fantasies about transitional objects achieving autonomy, from The Velveteen Rabbit to the Toy Story films: but when Isiguro affords his reader the robotic object’s first person vantage-point on this experience, he opens a window onto the struggle for autonomy that turns out to also uncover the coeval human struggles of young people in a near future where, again, smart drugs are threatening to constrict human autonomy within an extreme cyberbiopolitical meritocracy. Once again, what the robot achieves, in Klara and the sun, is becoming the commodity that returns to us the gift of humanity brought by novelistic experience.

Where do novels come from, anyhow? Mostly, these days, the answer is Amazon, which is why it was with grim, if playful, irony, that Amazon Prime recently adapted Philip K. Dick’s “Autofac,” a tale of automated factories run amok, still delivering packages well after a nuclear holocaust has decimated their consumer base. As Colin Koopman delineates in his recent study How We Became Our Data: A Genealogy of the Informational Person, what the would-be resistance heroes who confront the Autofac in Dick’s original story fail to understand is that the problem they face is not one of *communication*, of letting the Autofac’s robot representatives know that they no longer want it to operate; it is one of *information*, of having become enmeshed in a world-making system no longer respectful of their wishes for human autonomy. In Amazon’s version, which, befitting a series of Dick adaptations entitled “Electric Dreams,” infuses Bladerunner aesthetics into the “Autofac” story, not only is the Autofac representative portrayed by Janelle Monae as a sympathetic android, the humans who face down the Autofac discover that they, too, are androids. They were always already androids, we might think, even when they were only getting novels, and not everything, from the Amazon/Autofac: cognitive assemblages, for worse or, the episode suggests, perhaps for better. If we, too, were not always already androids, perhaps our books have always been, after a fashion, automata, which is why they insist on reading us, going forward, as robots.