ABSTRACT
Artificial Intelligence as a research project has its founding myth on the notion that mimicry and intelligence are indistinguishable. The limitations of imitation have recently been exposed by large language models, prompting computer scientists to move beyond Turing’s ‘Imitation Game’ as the arbiter of progress in the field. Taking interaction as a guiding theme, I aim to summarize three noteworthy critiques of AI—from phenomenology (Dreyfus), pragmatism (Brandom), and inferentialism (Sellars)—in an attempt to consider the aesthetics of deep learning models and the notion of intelligibility that they (re)present to us. The epistemic capacities of AI will be explored with a focus on interaction, agency and performativity, emphasizing the re-cognitive character of intelligence. I will discuss two challenges to the Chomskian view that automata are moored to generative grammars, firstly from advocates of interactive computation and their proposal of ‘interaction grammars’, and secondly from Brandom’s elaboration of expressive bootstrapping. I will then present a challenge to the central position of denotational semantics and Boolean logic in theories of computation, as a means of rendering richer semantics for computational states. As a case study, I will analyse the family of models known as latent diffusion, examining how notions of conceptualisation and creation are present in our interactions with these multi-modal models, in the context of art production specifically.
