We have two aims in this paper: (1) to taxonimise the literature on machine speech, (2) to evaluate whether the question of machine speech is worth asking.
Firstly, we suggest that responses fall into two camps: superficialists and deepists.
Superficialists think that we can discern whether a machine speaks by considering its outward properties. Any system rightly viewed under the intentional stance (Dennett 1989), or 'meets all the a priori constraints' on the concept of speaking (Chalmers 2023) thereby counts as speaking.
Conversely, deepists think we should look “under the hood” to details like algorithms and computational processes. Deepists may think that speaking requires intentionality, which requires 'internal causal powers equivalent to those of brains' (Seale 1980), or that current LLM-based systems are mere stochastic parrots (Bender and Koller 2020).
We then argue that the question of machine speech is not worth asking. Our guide is Turing, who thinks that the question of machine intelligence is "too meaningless to deserve discussion" (1950:442). We concur: despite progress in our understanding of meaning and communication, these key notions are still too undetermined. What we should ask, instead, is how to go on using ‘speaks’.
