16 Machining Coloniality and Learning Otherwise

Coloniality of Being

In recent decades, a wide range of thinkers1 have argued that any consideration of contemporary epistemology or scientific reason must run through Sylvia Wynter’s work. One of her core arguments helps us make sense of machine learning. According to Wynter, the Western philosophical project that Heidegger invokes in his cybernetic lament was never concerned with being in general and, instead, was always conditioned on counting and accounting (Wynter 1984). From its humanist inception in the overturning of the scholastic-theological model of philosophy, the West has articulated its ontoepistemology with white manhood as its ground. Wynter argues that the myth of the unmarked Western Man is recursively constituted in a binary division of the world between order and disorder. Order, marked as white, is (to borrow from Ferreira da Silva’s [2006] framework) transparently understandable and knowable. Disorder (or chaos), marked as Black, is intrinsically unknowable and therefore must be made to be understandable through counting and accounting (Wynter 1984, 1992, 2003). If Heidegger imagines that Western philosophy ends with cybernetics because cybernetics reduces being to calculation, and in so doing inaugurates a world system that is “the beginning of the world civilization that is based upon Western European thinking,” then Wynter responds that “Western European thinking” is, and has been, the episteme of the world system ever since it was inaugurated in the accounting practices of the Middle Passage (Keeling 2019).

Kara Keeling’s work, informed by scholars like Ian Baucom (2005), whose close reading of the Zong massacre’s insurance and criminal cases, shows that contemporary capital’s “death dealing abstractions” (Gilmore 2023) are not a break from but are a remediated intensification of the logics of genocidal eighteenth-century Atlantic capitalism. Drawing on Stephanie Smallwood (2007), she argues that in the context of eighteenth-century capitalism, Blackness is posited as outside of the human and is intelligible only through the brutal calculating remediation of property and value. Extrapolating from this insight, Keeling continues that to think contemporary digitality’s algorithmic condition requires an encounter with the intrinsic anti-Blackness of contemporary rationalizing and calculative tendencies. This dovetails with (and expands on) what Wynter (2001) calls the sociogenic principle, whereby the racist and racializing narrative and calculative infrastructure of Western humanism becomes enfleshed in living bodies.

These are not new problems to the post-metaphysical cybernetic world, nor are they universal to the totality of the world. But they are old problems for the Atlantic world of learning machines. The exclusion of the “African” from the realm of spirit, reason, and the knowable is a common feature of many major European post-Enlightenment metaphysical systems (Jackson 2020; Amaro 2023). This exclusion, along with a concurrent genocidal debasement and dismissal of Indigenous knowledge and indigeneity globally, set the stage for the sociogenic mathematics that inheres in racial capitalism’s learning apparatuses. This inheritance might not be a problem proper to mathematics in general, but it is a central and recurring problem of how “the determination of man as an acting social being” is realized in the accounting practices of US hegemony. As Wynter (1993), riffing on the LAPD, put it—this ontoepistemology looks at negatively racialized and poor people and sees No Humans Involved. Wynter (2003) named this structure of thought the coloniality of being and argued that “the social order proper to this world” is caught up in its constitutive mythopoesis.

Machines, Learning

The question of whether machine learning is, in fact, an example of machines learning is a question of thought. Specifically, it is a matter of determining whether machines think anything new (as Beatrice Fazi [2019] has put it). Here a distinction must be made between the capacities of computational media to conduct alien (novel) thought, and the “real world” contexts in which machine learning is imagined, developed, and deployed against the social world. In the first case, it seems clear that computational media’s potentialities for alien thought are enormous and—in fact—effect a quantitative dimension that puts them qualitatively outside of the realm of human reason (Parisi 2022). But machine learning’s ding an sich and alien potentiality sometimes seems of little moment to a social context in which what machines learn under algorithmic conditions appears to be little more than a recuperation of the sociogenic principle (Wynter 2001), quantified and re-presented as recursively automated Bayesian tools.

Parisi and Dixon-Román (2020) articulate the stakes of this problem in their consideration of predictive intelligence as an accrual strategy of data capitalism. Looking at predictive policing, they point out that the algorithmic modeling that underlies such systems depends on a basic uninterpretability. Contrasting what they name “computational modeling” against “data modeling” (where statistical techniques are used to make sense of existing sets of data), they show that what predictive policing systems “learn” is predictive accuracy, whereby that which is predicted (“crime”), and the degree of success (“accuracy”) are both nebulous and determined by sociogenic conditions. Such systems search for “crime” as determined by police forces, rather than some other problem like “threats to democracy” (Parisi and Dixon-Román 2020, 54). Whether or not these systems “work” is determined by an ineffable: If crime prediction software is effective, then there will be no data to memorialize and trace its impact (because no crime). If it is not effective, then there is no way to make it accountable, or isolate its impacts vis other factors because its data is already folded into a datafied social world, constantly in flux. Given that the one constant in predictive policing software appears to be that it exacerbates racial disparities in enforcement, what this particular type of machine learning leaves us with is little more than a reified quantitative grip on sociogenic violence work (Seigel 2018). In the case of predictive policing, that which algorithms are supposed to “learn” is a seamless labor management strategy for violently maintaining racial capitalist hierarchy (Scannell 2019).

Predictive policing is critical low-hanging fruit because it is intuitively dystopian—even as its futural orientation obscures actually existing violence (Scannell 2019). But what makes it particularly interesting as a case study in the coloniality of machine learning is precisely how bizarre these systems tend to be. Crime prediction software is built on computational models of everything from earthquake aftershocks to predatory animals’ hunting patterns to neolithic migration flows and the competitive behaviors of grass species (Scannell forthcoming). In other words, this software “learns” about “crime”—a fuzzy sociopolitical concept—by way of the attempted capture and calculation of inhuman forces. In a very real sense, computational modeling vastly expands Wynter’s analysis and introjects planetary computation (Bratton 2015) into the sociogenic principle. Rather than “learning” in terms of tool adaptation or humans systematizing knowledge and experience, the “machine learning” in this context recruits human-agnostic dynamics to testify in service of the continued hegemony of racial capitalist power dynamics. This harnesses ecological learning to a specific end: a sociogenic conclusion that must be matched to the planetary syllogism. Rather than make it an outlier, this strangeness tells us something central about sociogenic machine learning: In addition to the epistemological questions raised by the idea of learning machines, predictive policing points us back toward algorithmic coloniality’s curious and unstable ontologies.

Coloniality Machines

Western Man has not historically maintained clear distinctions between humans, tools, and machines, a problem that we call “race.” Reading Wynter into the machine, as it were, allows us to see something basic about sociogenic machine learning: that it does not threaten “the human” as such, because it is not “human” thinking and learning that is being modeled and there is no agreed-upon figure of “the human” as a learning target. Instead, sociogenic machine learning is a rationalization and axiomatization of the coloniality of being. Machine Learning’s fundamental operation is to determine a chaotic outside to knowledge, to harness that chaos and then wrangle it into creative order. Under US conditions where ordered accounting subtend claims to personhood, this ordering scans less as a sociological racializing project than an ontoepistemological working of sociogeny (Fanon 2008). The outside can be made to be known because it can be made affectable. It can be made a ground (Silva 2006, 2015). Rather than the end of Western metaphysics, sociogenic Machine Learning’s ordering of automatic thought points to our specific genealogy’s beginning.

Calvin Warren (2019), reflecting on Katherine McKittrick’s canonical “Mathematics Black Life,” names mathematical thinking as “[carrying] a certain phenomenological violence with regard to blackness” (361). He points out that, far from the ontological purity with which it is normally associated, mathematics’ centrality to the history of anti-Black violence is located “precisely [in] the imbrication of phenomenology and ontology, and it is difficult to disentangle pure form from a violent situation” (Warren 2019, 360). That is to say that, in the context of McKittrick’s archive in the hold, the supposed neutrality of mathematics “is complicit in reproducing antiblack violence. To ‘do the math’ within McKittrick’s theory acknowledges that epidermalization contaminates every ‘pure form’ and thinking mathematically requires protocols for addressing this contamination” (361). This is a critique of the narrative (in Wynter’s sense) of what mathematics as “Western European Thinking” is and can do, rather than of mathematics itself. Warren ends his essay with a gesture toward the possibilities of chaos theory, itself a “Western” mathematics that holds open the untoward and otherwise. But what this leaves us with in the context of machine learning in “the social order proper to this world” is ambiguous. If metaphysics subsumed under the sign of cybernetic reason reduces the world to violent accounting practices, as Wynter suggests it might, then the calculative basis of the contemporary appears as little more than the recursive annihilation of possibility. As a set of sociogenic practices, Wynterian analysis suggests that sociogenic machine learning is not only “contaminated” (Warren’s word) but constituted by coloniality. This suggests the radical parochialism of the Heideggerian “Western.” That “Western European thinking” in all its bloodied calculus dominates planetary horizons might be the case. But it also rests on a skein of absurdities that are not, ultimately, tenable. The planet will not tolerate the hegemony of Western European thinking for much longer, as increasingly dire warming models remind us. And so the project must be to find the way out from the trap of machined coloniality. Wynter’s work was not, after all, merely to diagnose the violence of Western reason, but to undo it. Her effort was to discover the ceremony that could overturn the coloniality of being, not to surrender to its hegemony.

Such an effort requires, of course, a new type of learning how to be in the world that is untethered to the hegemony of Western Man. Or, pace Heidegger, a way of staging an encounter with a world that is not “based on Western European thinking.” There is no reason to assume that such an encounter will have nothing to do with cybernetics and its inheritors in the world of learning machines. But it might, at least, require a relinquishing of sociogenic cybernetic reason from its grasp on the doings of the world. Perhaps, against the ordering of the world, a radical filiation with its claimed opacity (Glissant 1997), or a disordered embrace of algorithmic thought’s incompressible remainder (Parisi 2013)? Machines are strange learners, and more capacious than the parochializing foreclosures of the world that sociogeny demands. What might a world look like that embraces their alienness, rather than looks to them for the way forward in maintaining the sociogenic principle?

Note

1. 1. See, for example, McKittrick (2015).

References