Hugo Idarraga franco

Myths about the becoming of all modes of being from precosmological times are present in many Amerindian myths. In Amerindian thinking, these mythical times spread a common condition shared by all actants before their current metamorphosis. Eduardo Viveiros de Castro’s perspectivism, drawing upon these Amerindian cosmologies, suggests conceptualizing these mythical times as the moment of indifferentiation -or pure potency- where all ‘actants’ should be considered originally as humans. In this regard, perspectivism appears to challenge a classical perspective, which was rejuvenated in the 20th century by certain cybernetic and artificial intelligence theories, positing that all beings can and should be understood and explained as complex, discrete, rational, and mechanical machines.

My primary argument is that by comparing Amerindian and machine learning mythologies, perspectivism challenges the constrained conceptualization of actants and matter imposed by the latter, as it seeks to encompass the entirety of the world as its fundamental justification for existence and applicability. This speculative dialogue raises a crucial question regarding the distinction between what is computable and what is not.

In machine learning classification models, an individual is simply and arbitrarily defined as an array of ones and zeros tethered to a semantic label. This semantic arbitrariness is the mythological consequence of a time before algorithmic entities’ distinction and differentiation, as in the Amerindian humanness mythic time, but in this case a machineness mythical time. Like in the times where a shared-person condition of all actants is predicated, in machine learning classification models there is an indiscernible amount of potential entities before the onset of data processing. This myth is indispensable to the planetary applicability of machine learning and is predicated upon another myth regarding the universal aspirations of computing, wherein the idea that everything on this planet is computable is embraced. By examining Amerindian creation myths and confronting them with the myth underlying machine learning's ambition to classify the entire world, the question about what is computable helps elucidate certain epistemological bias that reinforce colonialism and contribute to the erasure of non-hegemonic knowledge regarding the ontology of the world.

Arising at the intersection of interdisciplinary scientific approaches, contemporary artificial intelligence (AI) techniques, such as machine learning and deep learning using neural networks, are often perceived as embodying orderly and impartial rational processes through complex mathematical formulations. Popular claims about the super-intelligence of these models point out to this ability to apply rational procedures to vast amounts of data that may be beyond human capacity. However, taken to their logical extremes, this form of rationality, which I term radical rationalism, can yield machine learning outputs that are incomprehensible to humans, appearing as if they are generated by an alien form of rationality rather than a human one.

In such cases, machine learning outputs may seem irrational to human understanding, but in reality, they are the unexpected outcomes of a radical rationalism that arises from algorithmic functioning within artificial neural networks. This paper posits two hypotheses: firstly, that human rationality could become an alien form of rationality when radical rationalism is demanded by algorithmic processes within AI neural networks; and secondly, that this alien form of rationality, or radical rationalism, could be utilized as a potential strategy in the context of emergent hacking practices within the field of machine learning. Thus, the notion of radical rationalism in machine learning raises intriguing questions about the potential alien nature of AI-generated outputs and their implications for the evolving field of adversarial machine learning.

If we can speak of ANN cultures that are both shaped by the spread of ML applications and by the cultural processes that these learning algorithms undergo, characterized by a privatization of process, I propose the term "Neural Networks Visual Counter-Culture" (NVC) to describe the emerging set of visual practices that oppose the dominant values of the current ANN culture, which is predominantly controlled by corporations and security forces. These artistic and political expressions are creating a new visual culture of evasion, camouflage, and invisibility, where visual data becomes an operational image, as it challenges the military and economic interests embedded in its functioning. If culture is a site of identity formation, then counter-culture becomes a space where people can subvert and run away from their algorithmic identities.

This paper presentation focuses on visual anti-hegemonic data practices and proposes an analysis of these counter-culture manifestations through the lens of visibility/invisibility dialectic within the field of computer vision. It aims to explore possible strategies to deceive the classificatory duties of algorithms over historically marginalized people. These practices are examined through a theoretical inquiry into the sociogenic replicator code that perpetuates a colonial epistemology based on classificatory practices from the 17th and 18th centuries in Europe, which were imposed as universal truths determining the hierarchical allocation of humans and non-humans in the modern chain of being.

In her search for a definition of humanity that goes beyond the sociogenic replicator code, where a second set of instructions defines what it means to be human according to Man1's Renaissance homo politicus and Man2's homo oeconomicus, Sylvia Wynter proposes a wide horizon in which a purportedly ecumenical condition of all humans obliges us to return critically to the origin myths that formed these Man1 and Man2 narratives. Through this critical return, different ways of how this replicator code is reproduced in an autopoietic fashion emerge. The hypothesis proposed for this presentation is that this sociogenic replicator code, drawing on feedback and autopoiesis cybernetics concepts, is inherited into machine learning classification models as a foundational myth, leveraging the monohumanist ideology of the colonial project.

This article proposes to analyze the definition of the human being through the ‘eyes’ of machine learning models dedicated to the recognition and classification of people. For this, the image of a dialectic of surveillance and social control is offered based on the constant visibility of those affected and the invisibility of those who benefit from this system. The algorithmic vision emerges here as a privileged place to exercise this power, but at the same time as the place of vulnerability, from where it is possible to transform this dialectic and the power relations that it determines. The article then proposes a reflection on the different counter-surveillance strategies that take advantage of creativity and aesthetic experience to think about other power relations, that is, about other possibilities to think about what a human being is today.