Shakespeare and Artificial Intelligence
Since the Enlightenment, Shakespeare’s plays have been served as heuristics for exploring human behaviour and personality. The uncanny capacity of Shakespeare’s characters to make us infer an inner mental life from exterior behaviour and words is, for the first time, being matched by a similarly uncanny ability in machines. Shakespeare’s plays do not merely reflect human behaviour but create plausible prognostic simulacra of human minds not unlike modern Artificial Intelligences (AIs). If, as some believe, the machines are starting to actually understand us, then they do so in ways that drama has understood people (and predicted their behaviour) for centuries.
This panel will explore how Shakespeare’s creative explorations of human society, culture and interaction can contribute to the comprehension of the increasingly sophisticated AIs that are being developed, while also honouring the fundamental difference between machine and human intelligence. The precise inner workings of artificial neural networks are too complex to be fully understood even by their creators. We account for them by how we train them and by what they subsequently do. So it is, to a large extent, with people. As machines start exhibiting human characteristics, the methods by which we have long made sense of human behaviour will increasingly be needed to understand machines, while, in turn, the inscrutability of AI might have a bearing on the way we think about plays.
The problem of inscrutability affects our understanding of AIs and Shakespeare’s characters in similar ways. In both, ‘inner workings‘ can only be hypothesized from outward manifestations; characters are black boxes to each other and spectators. Even theatrical strategies that serve to reveal characters‘ minds at work (soliloquies, asides) ultimately affirm the inscrutability they ostensibly transcend by sourcing it for plot devices such as dramatic irony. When plot construction is foregrounded in this way, plays illustrate that they are not about human desires, intentions and plans, but, like AIs, goal-driven systems following a theatrical rather than a psychological logic.
In English studies we have specialists who discriminate between existing writings (literary critics and linguists) and those who create new writings. AIs likewise divide between the discriminative (for example those detecting spam in email and cancers in x-ray images) and the generative (those creating new texts and images). If techniques from English studies help us make sense of AI, it may be worth exploring whether this likeness is significant or trivial.
The research questions the panel will seek to explore are:
* How can the linguistic and structural properties of Shakespeare’s plays help model the inscrutable workings of AI?
* In turn, how can what we know about AI contribute to a more sophisticated understanding of these dramatic properties (for instance, does this knowledge help shed light on how plays lead spectators to develop a sense of character)?
* What are the limitations of comparing dramatic art and AI? In how far will inscrutability also increasingly differentiate art and AIs, especially generative AIs whose communicative transparency can be developed and improved by, among other things, prompt engineering.
Gabriel Egan, De Montfort University, Leicester, England
How are the speeches of Shakespeare’s characters individuated?
The transformational generative grammar introduced by Noam Chomsky in ‘Syntactic Structures‘ (1957) offered the first model that could plausibly account for how the human mind generates original sentences. Chomsky thought he had proved that finite state automata were inadequate to the task, but the new Large Language Models that so impressively mimic human speech are built on those derided finite state automata. This talk will compare how these two ways of modelling speech help us understand Shakespeare’s techniques for individuating his characters‘ speaking styles.
Anja Muller-Wood, Johannes Gutenberg-Universitat, Mainz, Germany
Inscrutable Plots
Anthropomorphism is a problem in the discussion of plays and of AI. Neither dramatic characters nor machines have minds of their own, and yet intuitively we seem to think about both in this way. Strongly plotted plays such as ‘The Comedy of Errors‘ and ‘Othello‘, however, defy such psychologizing and instead foreground that character is subjected to, even created by the plot and its integral logic. In this paper I will consider how such plays demand to be thought of as “systems” not unlike AI and how this take on plays may also shed light on the inscrutability of AI.
Heejin Kim, Kyungpook National University, Daegu, South Korea
Computational Dramaturgy: Deep Learning for Character Type Analysis in Early Modern Plays
This study explores the integration of Artificial Intelligence with literary analysis to examine character archetypes in early modern dramas. By leveraging a dataset comprising more than 500 plot summaries linked to full dramatic texts, a neural network model is used to provide quantifiable insights into the prevalent character archetypes of the period and their narrative functions. The model, trained on both summaries and full texts, aims to classify complex character roles, such as heroes, villains, and sidekicks, based on their narrative contributions and developmental arcs.