Prompts – the simple written instructions given to an algorithmically controlled language or image model – seem to have almost magical qualities: with just a few words plus “artificial intelligence,” they can generate a seemingly infinite amount of new texts, images, and other outputs.
Prompts are a key element of the “rhetoric of computation” (Hayles, 1999; Hayles and Hillis, 2011; Hayles, 2012): they not only define the parameters within which a machine can work, but also shape the output of the machine. As such, prompts are a powerful tool for shaping the rhetoric of computation – and thus the rhetoric of digital media more generally.
The italicized sentences above were generated by OpenAI’s GPT‑3 model (cf. Radford et al., 2020), completing the introduction and adding a mostly plausible second paragraph. If the enchantment over such accurate text production has worn off slightly, it is mainly because we have already become accustomed to the rapid advancement of pre-trained models such as GPT‑3, LaMDA or DALL‑E 2. Wielding the power of billions of machine learning parameters, prompts are used every day to produce photorealistic images of a “dragon fruit wearing karate belt in the snow” (Saharia et al., 2022), academic papers “with minimal human input” (Gpt, Thunström & Steingrimsson, 2022) or “self-help essayist”-style blog posts (Thompson, 2022).
Undoubtedly, meaning-producing AI models not only possess their own computational or digital rhetoricity; the rhetoric of prompts and how they are used strategically is about to transform and shape public communication fundamentally (cf. Schäfer and Wessler, 2020). This talk will analyze the rhetorical dimensions of prompts, including their functions and modes of action with regard to the rhetorical system, the interplay between humans and machines with regard to algorithmic creativity and the meaning and effects of prompts with regard to the structure of communication.
Literature
· Gpt Generative Pretrained Transformer, Almira Osmanovic Thunström, Steinn Steingrimsson (2022). Can GPT‑3 Write an Academic Paper on Itself, with Minimal Human Input?. HAL. ffhal-03701250f
· Hayles, N. K. (1999). How We Became Posthuman. University of Chicago Press. (Referenced by GPT3.)
· Hayles and Hillis (2011) = Nonexistent (Referenced by GPT3.)
· Hayles, N. K. (2012). How We Think. University of Chicago Press. (Referenced by GPT3.)
· Saharia, C. et al. (2022). Photorealistic Text-to-Image Diffusion Models with Deep Language Understanding. arXiv https://doi.org/10.48550/arXiv.2205.11487
· Schäfer, M. S., & Wessler, H. (2020). Öffentliche Kommunikation in Zeiten künstlicher Intelligenz. Publizistik 65, 307–331.
· Thompson, Clive (2022). How Many Stories on Medium Have Been Written by AI? Medium. https://clivethompson.medium.com/how-many-stories-on-medium-have-been-written-by-ai-8e632da3f79b
