I’ve become a bit fascinated with GPT-3’s predictive text abilities. Beyond the potential to replace an entire wage tier of freelance journalists (dare I say good riddance?), GPT-3’s ability to unintentionally create art is probably the most successful thing about it. The ever-reliable Gwern tested this to great effect, with the AI returning the following poem when trained on acrostics:
…H A M M E R
H ands | Hands you must hold the hammer with to pound the nail
A rm | Arm you must raise to strike the nail to make it fall
M ind | Mind you must think with to know when to strike the nail
M achine | Machine you can make a hammer from a rock
E ngineer | Engineer you can be to make a better machine
H A M M E R
H uman | Human life is the most important thing we have
A ction | Action we must take to preserve life for all
M oment | Moment of time is what life is all about
E nd | End is something we all must face one day
R eason | Reason is why we should preserve life on Earth
This poem immediately reminded me of Russian poet and socialist organizer Aleksei Kapitonovich Gastev. A metalworker by trade, Gastev codified Soviet technology fetishism into the school of machinist poetry in the 1920s. Notable for its observationalist style and severe restriction of vocabulary and form, machinism repackaged poetry as technical instructions for life in an industrial age. In Gastev’s view, human cognition and anatomy could be honed through repetition to become ever more precise, with the ultimate goal of biomechanical perfection: man-as-machine in transcendent symbiosis with his tools, able to swing a hammer with profound, replicable accuracy or operate a typewriter with divine aplomb.
Per Gastev, poems were to be recited in “uniform batches…as if one were operating a machine”, that the listeners should perceive them as “a libretto of crucial processes.” Machinism provided instructions for technical mastery. See below, excerpted from Gastev’s 1921 publication A Packet of Orders:
To the machines.English translation via Flesh to Metal: Soviet Literature and the Alchemy of Revolution by Rolf Hellebust
Charge of attention.
Half-minute interval. Switch over.
Method two, method four.
Series twenty, begin work.
Gastev is notable for predicting the datafication of the body at work about a century early: the optimized, measured, and surveilled body of Gastev’s mechanized laborers has a parallel in the modern employee subject to productivity tracking. Yet farther afield, the training of an AI such as GPT-3 represents a human-machine interface that Gastev could hardly have predicted. Gastev’s machines were tactile, iron and steel, blood and muscle. An AI, especially one tucked behind a proprietary API like GPT-3, is a black box. The user and the AI communicate in a sort of disjointed language, acting in tandem but without full knowledge of the mechanism that drives their mutual work. In this sense, AI is incompatible with Gastev’s vision, lacking in physicality and explicable processes. Yet if the goal of machinist art is to render the human more mechanical, is there not also potential for making the machine more humanlike? GPT-3, with its accidental artworks, might be an inverse machinist: a machine that aspires to reciprocally humanize its behavior.