Performativity without theatricality: experiments at the limit of staging AI

ABSTRACT

Referencing participant observation in a research-creation lab devoted to performance and artificial intelligence (AI), this article summarizes and intervenes within two discourses surrounding the performativity of computation. I first summarize the media-theoretical debate over whether or not electronic computation counts as what J. L. Austin and Jacques Derrida defined as ‘performative’. This turns out to be a divide over the politics of theoretical analysis, and as such these positions can be synthesized together. Relying on Samuel Weber’s concept of ‘theatricality’, I set out a novel proposal for understanding computation as representing a limit of performativity without theatricality. Secondly, I review the experiments conducted with staging recent machine-learning models within the University of Toronto’s BMO Lab. A scholarly tradition distinct from the above has turned to a ‘metaphysical performativity’, describing all reality as performatively animate rather than representational and inert; some have pointed to recent AI developments as a demonstration of the truth of this view. I dissent, with evidence from the aesthetic experience of watching AI performance. Finally, I critique the ideology implicit in theories that take the appearance of AI animacy as a model for social reality.

Disclosure statement

No potential conflict of interest was reported by the author.

Notes

1 The ‘Mechanical Turk’ automaton, first presented by inventor Wolfgang von Kempelen at the court of Empress Maria Theresa in 1770, professed to be a machine in the form of a life-sized doll ‘Oriental’ that could beat a human contestant at chess. In fact the device depended on a human chess-master concealed within a trick table; though early audiences (notably Napoleon) were reportedly convinced of the automaton’s inanimacy, subsequent tours in American led many (notably Edgar Allen Poe) to deduce and publish its secret. Since 2005 Amazon’s low-paying online microwork platform, cheekily and cruelly, has been called Mechanical Turk. At once a neat distillation of claims to false automation, a profound reminder of the centrality of racialization to imaginations of computing, and a clear link between European imperialism and contemporary global labor distribution, the ‘Mechanical Turk’ has prompted recent theoretical reflection (Geoghegan 2020; Dhaliwal 2022).

2 On performatives and promises, see Felman (2003, 1–47).

3 This is not to say that machine learning does not anticipate – one could argue that this is precisely what machine learning models do. They interpellate and project, and from predictive typing to television frame-insertion, actively interject with their anticipations; much recent media theory has emphasized the radical consequences of such ‘protention’ (Hansen 2015; Denson 2020). However, these protentions shape and extend the subjectivities of users: to the extent that they ‘work’, they do not present themselves as dialogic. To construct a call-and-response interaction demands far more computing power. Though this article focuses on the un-theatricality of contemporary AI, this disjuncture suggests a further non-dramatic aspect to computation worth further thought.

4 For reasons not entirely clear, a moderately high temperature can turn into an angry stream of modern profanity, frequently with offensive language: an outcome sure to spur lengthy discussion if demonstrated in the classroom.

5 BMO Lab internal video, 3 May 2021, timestamp 7:31. I have resisted the impulse to include more quotations from our models: to cherry-pick excerpts misrepresents the whole.

6 The particular boredom of AI spectatorship was attested to by audience members of Annie Dorsen’s algorithmically parsed Hamlet, A Piece of Work, as discussed in Jucan (158–159).

7 Though I here lean on his newer book, which adopts a more radical position, the term ‘metaphysics' was used by Pickering to frame his 'performative idiom' in his early work (1995b, 5–9).

8 Latour uses a theatrical metaphor to make the point: ‘To use the word ‘actor’ [in actor-network theory] means that it’s never clear who and what is acting when we act since an actor on stage is never alone in acting. Play-acting puts us immediately into a thick imbroglio where the question of who is carrying out the action has become unfathomable’ (2005, 46). In truth, it is only after the various Latourian dicta – one should not look for representation, nor citation, nor compulsion, nor desire, and certainly not for coherent social groups or classes – that any theatregoer could begin to find it so mysterious that actors work together.

9 Derrida might note, as he did with J. L. Austin, that this ontology of non-representational force resembles that of Nietzsche.