What if humans are more like machines than we are willing to admit? This is the most basic question posed by Megan Ward's book. She boldly declares that the line between being human and seeming human is less important, and more tenuous, than we have determined, at least in literary studies (though her epilogue clearly shows how accepting mechanistic forms of seeming human could produce a more nuanced and constructive humanism).
While this book highlights the conversation between literature and technology, it reaches far beyond that conversation. For one, Ward's book joins a debate about character in realist fiction. "The scholarly story of Victorian character has long been a story of interiority," begins Jill Galvan in her entry on the keyword "character" for the latest issue of Victorian Literature and Culture. Surveying recent scholarship on new ways of understanding character, Galvan identifies this shift in the field as a "phenomenologically posthuman shift: a serious consideration of how we might read characters and their shared embodiment in light of the fallacies of liberal humanism" (615). A few years earlier, Daniel Hack seized the crux of the problem of character with a question: what is the distinction "between fictional characters and actual persons"? ("Fictional Character: Response," Victorian Studies 59.3: 420).
Ward reconstructs this question. If fictional characters are nonreferential, why do we need theories from psychology and anthropology to understand how they behave like real people? For a non-literary model of how fictional characters function, Ward turns instead to the twentieth-century concept of Artificial Intelligence. Taking four aspects of character--development, predictability, flatness, and mind-- that are traditionally linked to interiority, Ward shows how they can be freshly examined by means of routinized, superficial, and mechanistic characters. Though such characters ubiquitously populate realist literature, we have not yet adequately theorized them. By formulating a new theory of character, Ward's book reconstitutes them as major components of Victorian realist fiction.
On the surface, Ward simply reads Victorian literary character through mid-twentieth-century forms of artificial intelligence. But as her other writing on the V21 website demonstrates, her way of using what she calls this "historical middle" complicates the line between history and theory in debates about presentism in Victorian studies. Reading literature through the lens of AI, she writes,
challenges us to rethink our categories of history and theory, to reclassify what constitutes a historicist and a presentist approach to studying the past. By reading the nineteenth century through the mid-twentieth century, we both theorize and historicize in a way that challenges our definitions of both of those terms. I hope that doing so can shift critical debate away from the well-trodden ground of history vs. theory and re-imagine the relationship between the Victorian past and its multiple futures." (V21 website)
The book thus asks us to do more than just reconsider how to interpret characters. It rethinks the entire relationship between literary criticism, history, and theory. As Ward herself claims, her book "calls into question the uses of history in literary criticism" (6). The historical middle offers a moment when there were multiple models for understanding consciousness, exemplifying the versatility that realism itself models-- if we know how to look for it. Before 1960, when AI began limiting itself to a single theory of intelligence as dematerialized thought, it posited various forms of artificial intelligence: feedback loop, stochastic system, imitation, perceptron, and physical symbol system. Ward examines each of these in separate chapters.
First, Ward shows how the feedback loop is exemplified by repetitious domestic routines in three novels by Margaret Oliphant, Elizabeth Gaskell, and Charlotte Yonge. In Yonge's The Daisy Chain (1856), for instance, Ethel May "remakes herself" when she must assume the new role of household management by repeatedly doing daily household chores, "those little tiresome things" (qtd. from The Daisy Chain 54). But according to Ward, this kind of character development does not mean change. In a feedback loop, she contends, the goal of development could also be homeostasis or self-regulation. That is, a character such as Ethel could adapt to new responsibilities in something like the way a thermostat maintains a steady temperature. Likewise, in Margaret Oliphant's Miss Marjoribanks (1856), Lucilla keeps her name even after marriage and returns to her original suitor despite courting many others over 14 years.
Turning next to network models in sensation and detective fiction, Ward examines Charles Dickens's Bleak House (1853), Wilkie Collins's No Name (1862), and Mary Elizabeth Braddon's Lady Audley's Secret (1862). In AI, Ward explains, information from distinct signals depends on their relation to one other, a relation used to measure predictability. Though predictability in fiction indicates how characters are connected, Ward argues that critics have neglected this aspect of character. "[B]y prioritizing individual interiority in the development of realism," she writes, "we have overlooked the ways that fictional characters also form systemically across a text" (44). In Bleak House, for example, the circulation of Lady Dedlock's portrait upsets the equation between intimacy, knowability, and a character's lifelikeness. Other characters and the reader feel as though they know her, but not by means of her thoughts or her interiority, but rather through "superficial portable networks of characters' circulation" (53). In a system that looks to networks rather than individuals, that no longer reads character only in terms of surfaces and depth but as circulated information, predictability helps to make a literary character realistic.
Ward next challenges the widespread notion that flatness usually deflates a character's importance. Taking for example the fiction of Anthony Trollope, Ward argues that our inability to account for flat characters in realist novels shows how "the critical emphasis on interiority as the defining feature of realism has limited our notion of what constitutes modes of characterization" (95). Even in recent discussions of flat characters, Ward contends, the explanation and justification of these characters still positions and defines them solely in relation to more realistic "round" characters (79). Against this practice, Ward questions one of the basic tenets of realist character, that verisimilitude depends on interiority.
To contest this notion, Ward invokes the imitation game or "Turing test," as it is now known, invented by Alan Turing in 1950. Rather than trying to prove that machines can think like humans, Ward notes, Turing aimed to show that they could think in ways that seem human, as when a machine answers the following questions:
Q: Please write me a sonnet on the subject of the Forth Bridge.
A: Count me out on this one. I never could write poetry.
Q: Add 34,957 to 70,764.
A: (pause about 30 seconds and then give as answer): 105, 621.
These answers imitate human thinking because the machine apologizes for its inability to create original work and downplays computational skills by pausing and then by providing an answer that is incorrect (the right answer is 105,721). Though the machine has not learned to think like a human, it has learned to seem like a thinking human (71). Applying this formula to Trollope's flat characters, to characters whose interiorities we never see, Ward says that "being natural means faking imperfection, but only slightly" (81). Characters such as Lizzie and Lopez, then, who fake their status through superficial appearances and material means, can now help to show how realism achieves the effect of seeming human. Though they lack interiority, their "seeming real while being fake" (89) makes them realistic characters. According to Ward, they can even reveal the limits of interiorized characters. The solidity of surfaces in a character like Lizzie actually exposes by contrast the feebleness of attempts to represent interiority when the interiorized self is valued for its inscrutability, for the very fact that it is "almost impossible to represent" (77).
Pursuing this point in her final chapter, Ward argues that we should not construe the history of literary realism as a progress toward ever more interiority. Rather than ending with Henry James, therefore, she juxtaposes his fiction with that of Thomas Hardy in order to compare the realism of consciousness with the realism of characters who act like automatons.
To demonstrate that even Hardy's would-be automatons are human, Ward takes from AI the perceptron model, which helps us to see how Hardy's characters "offer surfaces as a kind of deep knowing" (105). By contrast, she tries to show that even James's most humanized characters act like machines, that his model of interiority resembles the AI physical symbol system "that reads like pure information" (110). Without privileging one method of characterization over the other, Ward argues that both are viable models of consciousness. Just as we can begin to see interiority represented as "another kind of machine," we can also see embodied and automated actions as ways of seeming human (111).
Since Ward applies AI to each novel differently, it can sometimes be difficult to grasp the analogy she draws in each chapter between a literary character and a particular form of AI. In chapter two, for example, where she reads character predictability through the networked model of the stochastic information system, the realist effect is said to be generated by the proliferation of information in No Name, but the reduction of information in Lady Audley's Secret. And in chapter one, we learn that while interiority amounts to an unethical secret in Gaskell, those "hidden interiors" become necessary for realist character in Oliphant (39). These contrasts do not undermine Ward's overall project, but they sometimes limit the cumulative power of her examples.
Ward makes good use of previous scholarship. She cites not only studies of literature and machines, but also conversations about literary character, network theory, and historicism as well as debates over the "rise" of theories about realism. While acknowledging the importance of this scholarship, she also shows how much its way of theorizing character privileges interiority. For example, Gerard Genette theorized repetition in Narrative Discourse (1979), but not, Ward claims, in a way that accounts for the "recursive temporal structures inherent to domestic realism" (19). Likewise, Nancy Armstrong's work on the development of selfhood in Desire and Domestic Fiction (1987) makes interiority necessary for such development. In Narrating Reality (1999), Harry Shaw notes the importance of surfaces to realism but still contends that realist fiction must represent inner life above all else. Disappointingly, Ward barely explains the relation between her argument and that of John Frow's Character and Person (2014), which also tries to undo the binary between purely fictional constructs and characters who seem just like humans. And while she skillfully uses Alex Woloch's The One vs. the Many (2003) and Franco Morretti's Graphs, Maps, Trees (2005), whose theories of character networks seem crucial for Ward, she still claims that they rest on "a bedrock assumption of fictional character as individualistic" (49).
Ward could also have made more room for the historical debates about the "realness" of realism that include the critique of "copyism" and the arguments made by realist artists for the value of seeing life reflected back to us. Though these debates and arguments are well known, might they not include points that may have escaped our narrowed definition of what counts as "real" characters? Though Ward's book draws flat characters and repetitious acts out of our blind spots, she could have explained how critics formulated those blind spots in the first place. Without recycling the very kind of historicism she overtly resists, she might have recovered for us terms we may have lost as we narrowed the conversation about literary realism.
In fact, Ward does make brief and always refreshing references to nineteenth-century historical moments. She recalls how Victorians' own understanding of information shifted from embodied knowledge to abstract knowledge, exemplified in the census, the office, and the library. She reminds us that Byron's daughter Ada Lovelace was not only a mathematician but a forerunner in the field of artificial intelligence: her work on the "problem of origination" would form the basis for the Turing test, which treats the ability to copy, not originate, as a sign of intelligence in machines.
Ward's specific interpretations are not always convincing. I question, for instance, the claim that Molly's stitching in Gaskell's Wives and Daughters (1864) becomes consciousness itself as the feedback of miscounted cross-stitches develops a new kind of self in her. Nonetheless, Ward provocatively challenges settled ways of reading and interpreting character. With patience, carefully selected examples, detailed close readings, and broad strokes of theory and science, she rewards the reader's willingness to follow her to the end. And in the end, I found myself already thinking differently about the literary characters I have revisited since reading Ward's book. Perhaps her own book generates the sort of recursive, feedback loop that she finds working in AI and the Victorian novel.
Kristen Pond is Assistant Professor of English at Baylor University.