Turret Theory
‘Demystification’, the namesake of ‘dmstfctn’, however, is an intriguing alternative to these demands for explanation or clarity. It’s a tempered request: neither for the neural network that underlies machine learning to somehow be able to show its internal working, nor necessarily for the exposure of dataset composition. Rather, it sheds light on the impetus for running a project like this: that there are different analytic levels upon which we understand a complex system like machine learning – different degrees to which it is ‘explained’ (or made accountable) to different audiences.
Godmode Epochs is perhaps a didactic explanation, giving the audience an impression of how the algorithmic systems which have influence over our lives are trained and operated. It doesn’t open up the black box of a neural network, but rather the system as a whole: the agent, its environs, and the environs’ content. This stands whether we see the neural network’s internal state as constituted by its ‘experience’ (read: data inputs) or if there is an internal and obscured logic beyond what can be accounted for. I would suggest that ‘demystification’ as an alternative metric for explicability brings the audience towards a halfway-knowing point, as proposed by Nora Khan and Peli Grietzer. If we accept their proposition that often our understanding of a phenomenon is really ‘half-understood’, then Godmode Epochs is containerful of suitably unstable machine learning exploration, packaged as a sad robot assistance game which imparts, alongside the player, an opportunity for reciprocal learning.
—
Alasdair Milne is a PhD researcher with Serpentine’s Creative AI Lab and King’s College London. His work focuses on the collaborative systems that emerge around new technologies, synthesising critical and analytic philosophical approaches to assess them through ‘cultural systems analysis’.
Notes
Godmode Epochs and the earlier performance GOD MODE (ep. 1) are part of a long-term research project, prototyped and incrementally evolved into episodic outputs – video, performance, simulation, participatory multiuser game – and both provoke increasingly important questions about how to communicate holistic systems to an audience. This essay primarily refers to the narrative which holds across both the earlier performance GOD MODE (ep. 1) and now the game Godmode Epochs as presented by Serpentine. R&D for the project was conducted in collaboration with Kevin Walker, Coventry University, and Eva Jäger and Mercedes Bunz, Creative AI Lab, Serpentine and King’s College London. Soundtrack for the work by Hero Image.