The panel ‘AI Poetics’ aims to discuss contemporary literary works from a transnational viewpoint that engages with recent developments in the fields of Artificial Intelligence. If AI has been recurrent theme in science fiction, it is now directly implicated in the making of literary works – both fiction and poetry – across linguistic, cultural, and geographical borders. How is this developing within World Literature and/or individual literatures shaping new directions for literary studies? To what extend, if at all, the literary and critical language employed so far can still work to engage with works produced through AI? In what ways can AI rather than bringing forward innovation – as they do in the sciences – is leading literature backwards? How can AI shake and diversify literary canons? Does the advent of AI make the use of computer programming and other technologies used so far in literary works outdated (e.g. Python programming language)? What does the interaction with AI tell us about the nature of literature and poetry?

Capra's paper will open this panel by posing a provocative and legitimate question on the value of literature produced by AI, by focusing on the quality of natives it can produce. Literature seems to share ChatGPT bullshitting-status: literature too can do away with truthfulness while leveraging all its persuasive power. Large Language Models thus call for a renewed engagement with poetics. According to Capra, literature is not bullshit, because it does not primarily dwell in the space of truth-production. At the same time, literariness is deceptively yet frequently deployed beyond literature to bullshit an audience about AI and its potentialities.

Dimirouli's paper expands on what has been introduced by Capra by focusing on the problematic notion of AI personhood and creativity. While the New Critics did not have machine learning in mind when debating the ‘death of the author’ and the ‘intentional fallacy’, questions about the extent to which text and author should be linked return with new force at the dawn of the AI era. As ChatGPT and GPT-4 produce verse indistinguishable from human writing in a matter of seconds, who exactly is speaking does seem to matter. Dimirouli focuses on authoriality and the reader-writer relationship to argue for the need to reframe well-established categories of the critical tradition in the context of AI developments.

The voice of the author is at the very core also of Straw's paper. Straw's paper will engage with the key question on the role of silence in poetry and poetics. In so doing, the comparative analysis of Philip and Celan will allow Straw to discuss the limits of sayability in poetry drawing on recent studies on AI poetics. For both Philip and Celan, there is a charged relationship between historical catastrophe, responsibility, and the limits of the sayable.

The panel is closed by a case study: the project Graph Poem, in the framework of the decade-long work at the intersection of NLP, network-analysis, and community-based performance. The previous accomplishments included developing NLP classifiers for poetry analysis (mainly in English), releasing computationally assembled anthologies. More recently, the project has moved on to assembling multilingual (minority language) multiplex corpora while keeping on developing analytical-creative approaches to HCI poetry. Chris and Raluca Tanasescu will highlight the data(-based) poetics ranging from text writing/assemblage/generation to algorithmic translation to automated corpus expansion, as informed by multiplex assemblage and analysis, concepts such as “computational prosody, and an intermediality poetics.

The session will operate within the allotted time by featuring short presentations from each panelist, followed by a moderated discussion that will allow for at least 15 minutes of discussion. This format will allow for a lively and engaging conversation that will encourage interaction between the panelists and audience members.