Doctoral thesis
To be defended at Aix-Marseille University, on December 13th 2025. Aix-en-Provence.
Defence committee:
- Dr. Daniel Andler. Emeritus Professor, Sorbonne University. Académie des Sciences morales et politiques.
- Dr. Jean-Philippe Narboux. Professor, Strasbourg University. CrePhAC (UR 2326).
- Dr. Juliet Floyd. Professor, Boston University. Center for the Humanities.
- Dr. Élise Marrou. Associate Professor, Sorbonne University. Métaphysique, histoires, transformations, actualités (UR 3552).
- Dr. Jérôme Dokic. Professor (Directeur d’études), École des hautes études en sciences sociales. Institut Jean Nicod.
- Dr. Michel Le Du. Professor, Aix-Marseille University. Centre Gilles Gaston Granger (UMR 7304).
Thesis summary:
In the field of AI and its related philosophy, there are a number of questions that stir the imagination and speculative minds of the researchers exploring them: Can a computer think? Could we build a conscious software? Should we be concerned about the welfare of robots? Are we witnessing an explosion in machine intelligence? etc.
Curiously, in contrast, there are certain questions that are seldom raised, but which nevertheless prove decisive in order to be able to address the former with confidence. My purpose here is to examine one of these neglected questions. Although seemingly very simple, this question (at least, as I hope to show) proves to be extraordinarily instructive: can a machine speak? In other words, can there be or could there be such a thing as ‘artificial speech’? The radical possibility being considered is that of machines that actually ‘speak’ (i.e. machines that assert, interrogate, command, question, advise, judge, promise, etc.).
In this work, I show that this putative possibility is widely accepted by theorists with very diverse positions and perspectives: the question of speech is a traditional focal point of speculation concerning the intellectual or mental capacities of machines, and the image of a talking machine is an important driving force in the field of AI (chapter 1). However, and here lies the thesis defended in this work, this image is misleading and rests on a series of confusions concerning both the nature of machines and the nature of speech. Drawing in particular on Aristotelian distinctions (chapter 2), I first show that we actually have very good reasons to doubt that what we immediately identify as artificial speech is an authentic form of speech. Following in the footsteps of Wittgenstein, Ryle and Austin, and more recently Peter Hacker
and Richard Moran, I argue – against unfounded accusations of chauvinism – that this ‘artificial speech’ is ultimately not a species of speech at all (chapters 3 and 4); authentic speech must indeed be motivated (it must have a point); it has a conventional dimension (and therefore requires that a convention be followed); it is most often expressive (and therefore requires desires and goals of one’s own); lastly, it is intentional (one cannot speak without being aware of it, and one always aims for something through speech) and binding. A useful way to understand machine text production (Chapter 5) is to elucidate the concepts of ‘imitation’ (for text production of artificial origin) and ‘simulation’ (for such production integrated into simulated conversations that may involve human beings). As a case study, I then examine an unfortunate conceptualisation of artificial speech in the epistemic context
of expertise (chapter 6). My investigation concludes with a more general reflection on what machines can truly do, in which it becomes clear once again that speech is clearly not the kind of ability that can be attributed to an artificial system (chapter 7). The idea of artificial speech can therefore be abandoned entirely, and the fantasies that accompany it (and which often fuel the exciting questions raised at the outset) can then evaporate without further ado.