Your second brain, built right.
~ / formula · blog
‘Ecology-as-a-Flow’ As an Excuse for Low Pay and No Fun At Work
Sat · Sep 2 · 2023 4 min read

‘Ecology-as-a-Flow’ As an Excuse for Low Pay and No Fun At Work

by Zhenia Vasiliev · bogotá

As Generative AI gets increasingly implicated in human decision-making, machine learning becomes a new sphere of pedagogy. The processes of teaching and learning, however, require a unified space of exchange between the counterparts. Such a space, created by the functioning of media objects, is explained as a media ecology by the cultural theorist Matthew Fuller. He notes its three general facets. First, it is the human-to-human network, which is mundanely institutional in that it describes the information flows between the departments and hierarchies. Second, it is a machinic ecology that describes the access protocols, the design of databases and archives, or the intellectual property governance. Third, there is an audit, - a practice of controlling the control itself, which links the first two characteristics and warrants ecological consistency.(Fuller, 2005: 3.) This last aspect is the most important for the understanding of the human-machine flow since this is where the human counterparts of AI pedagogy are held responsible for labelling and annotating the real world for the machinic apprentices. Simultaneously, here’s where the unbounded control makes their work alarmingly exploitative. The work of annotating here takes form of the cognitive microtasking where the value is no longer associated with an outcome but instead is turned into teaching as performative engagement with the machine. The ecology is therefore powered by the audit, or a verification process which benefits both human and non-human counterparts of the relation. To describe how such an audit works, we should associate the flow in biology and computation, and, once that link is sorted, trace the implications back to the two former assets theorised by Fuller, the institutional-biological and the computational.

The Flow and the invisible annotators

Biology, as philosophers of science Daniel J. Nicholson and John Dupré explain, predominantly uses verbs to discuss its research as ‘breathing, growing, flowing, cooling, contracting, singing, and orbiting.’ (Simons in Nicholson and Dupré, 2018:50.) In a similar move, the analytically-savvy technology, - Generative AI, or in this case more precisely, Ecology as a Flow, - does not deal with discrete events, but with processes that should be captured in their temporal character. In the observation of complexity scientist Brian Arthur, computation deals with verbs, such as ‘identifying a problem, constructing solutions to it …, testing and experimenting…, and solving the subproblems these bring up’ (Arthur, 2021: 6.) The Ecology as a Flow is supported by human labour of cognitive micro-tasking. The work of annotation and labelling, such as discussed by Josh Dzieza in his piece for the Verge, had not turned out to be a one-off event of helping the machine to learn about humanity. Rather, it increasingly becomes something of an attempt to formalise a new long-term bond with the non-human. As Dzieza observes, the work of annotation is often perceived as a slightly awkward and bothersome precondition for a more exciting work of building models: ‘you collect as much labeled data as you can get as cheaply as possible to train your model, and if it works, at least in theory, you no longer need the annotators. But annotation is never really finished.’ (Dzieza, 2023.) Quite on the contrary, the more proximal human and media ecologies become in their practices of learning about one another, the more nuanced their negotiations of matters of their affiliation have to become. The annotators deal with non-trivial problems of defining the difference between the white shirt with red stripes and the red shirt with white stripes, - and whether the items of clothing seen in the mirror are still clothes should still be considered reals clothes. For AI to persist, the cognitive labour of humans has to continue the Flow.

‘You collect as much labeled data as you can get as cheaply as possible to train your model, and if it works, at least in theory, you no longer need the annotators. But annotation is never really finished.’ - Dzieza, 2023.

The benefits of the Flow

Turning to what humans get from their participation in the Flow, we return to the question we started from. Are the benefits of the Flow big enough to justify the low pay and poor working conditions of the anonymous annotators?

Institutional and computational facets of media ecology provide two different ways of downplaying the machine’s significance. First, institutionally, the benefits of the machine are great, but they primarily aim at reducing the accidental complexity of production - such as helping to speed up novices, as the study by Stanford Digital Economy Lab found out. As the director of the Lab, economist Erik Brynjolfsson notes, ‘high-skilled workers may have less to gain from AI assistance precisely because AI recommendations capture the knowledge embodied in their own behaviors.’ (Lynch, 2023.) Second, from the computational perspective, the capitalist mode of production appropriates Ecology as a Flow and excuses itself for the sloppy organisation of cognitive micro-tasking due to the increase in surplus value. In its role as the agent of the connective tissue of the media ecology, Generative AI brings profit to capital by delivering the outcomes of future production to the present, essentially treating time as infrastructure. This, in turn, enables the capital to derive profits from the distribution of the ready product earlier than the competition, as explained by the political theorist Timothy Mitchell. (Mitchell, 2020.) The conversion of time into space happens when the inference of the relationships between the inputs and outputs can be dispatched algorithmically, as a property of the epistemic infrastructure. In other words, since AI creates the future as always already here, the role of the media ecology is to create a Flow that would enable the participants to negotiate the present.

@benjaminbardou. Imaginary Landscape #6 on objkt

References

← The dinosaurs in the tar pit, or why the software is so hard to produce

// SUBSCRIBE

Get the Daily by email

Every daily note and weekly guide, in your inbox. No spam.