20 Machine Learning and the Digital Archiving of Death

Learning with Archivists

The human (and humanist) practice of archivism involves keeping things for future learning. The archivist sorts through data, organizing, classifying, thinking, and deciding what to discard and what to keep in a collection. In turn, the archivist determines what the representative objects and or materials of an era or period are, the style and design of its making or production, and the category of object, as well as its position in an existing collection, or identification of the need for a new one, or dismantling of a previous holding. Such decisions about what to collect and how to categorize must be made in terms of a considered set of data—the sets of which are then available as selected items, or for evaluation. The subsequent data set stands for what is representative about the knowledge foundations of different cultures, and each of their geographically specific materials and conceptual meanings. The archivist gathers evidence of significance—whether artifactual materials, ephemera, or conceptual—in the form of the material information and its documentation, in order to ask the infrastructural informing question: “What is this?” (Diamond 1994, 139).

Designing the archival infrastructure for future learning is crucial because ML learns through practices of data set retrieval—as a multiple not singular system. Describing the process of moving from the one-shot learning ML algorithms to big data to solve the perception knowledge or visual learning of ML, Li (2023) comments, “Without data, what does ‘learning’ in ‘machine learning’ refer to?” (144). Li influenced the field of AI to think with big data, altering the course of ML when she developed ImageNet in 2012. This milestone work generated the first neural network image archive for ML. As such, what are the implications for the use of big dataset of human life, and what is in the archive that provides meaning to think about “an individual” death? Can humans still consider one brain, body, or multiple artifacts from a synaptic life to be representative of life and/or death? Using data sets removes any organic residue, resulting in the state imagined by Deleuze and Guattari (1980) as the ultimate model of death; the body without organs (149ff), which would become a set of bodies without organs. The question of learning moves from the subjective to the scientific technicalities of various algorithms, contingent upon the task(s) they are programmed to execute: Which data set is used? Is the data set then disposed of or retained? This radical change in standpoints concerning the value of a human, alters the field of learning and decision-making in terms of a neurological, psychological, socioeconomic, cultural function, and requires that we develop ethical methods for joining the scientific with the existential aspects of sentient life.

What Is Death?

Death can be many things, but it ultimately is a state that serves as a reminder for the finiteness of states of being/s. Death can be an abstract concept, a human-centric experience, an environmental event, or used as a data point for algorithms in the service of information governance. For humans, biological death is commonly understood as the cessation of biological functions that sustain a living organism, leading to the irreversible end of life. After the end of life, some culturally specific narratives and objects frame the previous “life” in terms of a philosophical or theological “afterlife.” These narratives and objects are often archived. In biological terms, the organism then transforms into another state, where the natural processes such as decay and decomposition change the matter of the body into materials used by other bodies.

In machinic terms, sorting, classification, and probabilistic algorithms collect examples of the different human-knowledge states of death as text, audio/visual, or human and synthetic data sets. Through the verification processes that AI knowledge-based systems use, and with the increasing use of AI, it is likely that the “human” cultural, theological, and social issues around death may well be stripped out—the subject of many a science-fiction narrative that speculates on the future state of human bodies. AI developments directly concern all aspects of death for living organisms, including data around healthcare, security, economic and environmental arenas. Geoffrey Hinton has suggested that digital intelligences, being “immortal,” might behave kindlier since they don’t fear death (Radical Ventures 2023, 1.03). Simanowski (2018) explores how algorithms mediate human experiences with death, through socializing frameworks like digital memorialization, raising ethical concerns in areas like automated vehicles, health care, and mortality risk assessment. He argues that by reducing human existence to data points, algorithms risk dehumanizing and objectifying individuals. The social consequences of AI objectification have been highlighted by scientists like Li (2023), who has noted that while academics focus on its negative aspects, the corporate and military interest in AI has made the technology both discriminatory and life threatening. Li (2023) refers us to not just the infamous philosophical “trolly” conundrum—about making an ethical choice over who lives and who dies in a complex scenario, but to the first known death caused by a self-driving car system in 2018, and the need for the AI industry to develop an “ethical framework” (Li 2023, 286–88, 291–92).

Machine Learning: The Difference Between Knowing and Understanding Death

What can human societies learn about death in the algorithmic condition? Conversely, how does the mechanization of ML consider human existentialisms, including its own death? This question becomes pertinent when current AI models have yet to demonstrate a breadth of “understanding” of different models of the human world. Discussing the recent claims of “intelligence” in the large language model ChatGPT 4.0, Russell (OECD.AI 2023) suggests that while the model may demonstrate some knowledge of the world, it remains partial in its understanding of the material and conceptual items it claims to know. At this time, learning is limited to the algorithmic goals set by humans in a largely unregulated and unconnected global curriculum space, and those goals are set by profit and territory driven market factors. We can find some instances in health care where data sets collected are put to a nonprofit and human-centric beneficial use; however, the larger connected infrastructure for delivery and sharing of understanding is absent. For example, during the global Covid-19 pandemic ML algorithms were used in mortality statistical modeling of human data sets (Alballa and Al-Turaiki 2021), also raising issues related to privacy, autonomy, and social justice.

Weighting the Ethics of Death with Technology

Reflecting on aspects of the mechanization of modern life through the nineteenth and twentieth centuries, including the industrial consumption of animals by humans, Giedion (1948) comments that with “the greater degree of mechanization, the further does contact with death become banished from life” (242). Giedion’s observation about the distance created by the machinic executioner and processor of animals as human food on an industrial scale, at greater speeds, is a reminder that machines do not approach the concept of death in the same way humans do, as they lack consciousness or the subjective weight that experiences contain, affecting individuals and communities in different ways. Death data in ML exists in various algorithmic numerical dimensions, encompassing event detection, prediction, sentiment analysis, representation learning, population health analytics, and natural language generation. In the context of military operations, which can lead to loss of life, data bundles can play a crucial role in intelligence gathering, surveillance, reconnaissance, decision-making, and operational planning. The thus far unregulated development and use of Lethal Autonomous Weapon Systems (LAWS) by the military use trained neural networks that currently rely on incomplete data sets, or poor generalized perception, for “codified” algorithmic decision-making (Longpre et al. 2022, 51).

Concerns about ML algorithms have led to calls for global ethical guidelines in the use of ML algorithms. Currently, AI guidance varies across nation-states, and one of the leading ANN scientists, Yoshua Bengio (2023), is calling for the development of a global AI ethics policy, outlining the potential catastrophic risks to humanity if AI continues its unregulated development. Such a policy would need to address the issue of the ethical requirement of beneficence, and by implication, respect for life—and the ways in which death is managed—in terms of the use of AI across all aspects of human society and the entire planetary ecology. In 2025, the ethical guidelines for AI use and development vary across nation-states, for example, the European Union (2023), the United States (White House Office of Science and Technology Policy 2022), and the United Nations (Perrin 2025).

Knowing and Understanding Death: Toward a Dignity Economy of AGI

Consideration of the large existential concepts that factor in every human body’s life can provide the ethos-making guidelines for deciding on what areas of artificial general intelligence (AGI) to pursue and develop. Li argues that in the conceptualization of AGI, we should be directing digital agents through algorithms that work on sub-groupings, such as transformers, toward consideration of issues around health, digital authentication, energy, policies to regulate AI, and the question of human dignity (Radical Ventures 2023). She sees the need to consider the future of ML as a major risk factor for human societies now, requiring international regulations to avoid “catastrophic” outcomes to avoid continued weaponization and human destructive tendencies (Radical Ventures 2003, 1.07ff). Li advocates the use of a human-centered AI learning as offering a potential move away from a labor economy, to one of a dignity economy, if carefully managed (Radical Ventures 2003, 1.08ff).

The question of dignity in relation to death of course is not a new one. How to have a good death is something many cultures and healthcare systems discuss and teach. ML modeling improvements in the understanding of bodies, their capacity for quality of life, could be improved for the future so that dignified death choices are readily available (Francis 2019).

How does the synthetic neural archivist determine what to keep of the human data set? What information or experiences are deemed appropriate to archive within a neural network or discard? Some work toward a new ethics of archiving automated paradata is emerging but remains inadequate for addressing the larger algorithmic condition driving this new way of learning with big data (e.g., see Davet et al. 2023).

AI has the potential to enhance end-of-life care, support grieving individuals, and assist with various practical and emotional aspects of managing death. However, it is essential to approach the development and implementation of AI in this context with sensitivity, empathy, and a deep understanding of the ethical, cultural, and humanistic dimensions of death and dying. As Russell (OECD.AI 2023) recently commented, “We’re going to make systems that are more powerful than us—how do we retain power over entities more powerful than us forever? . . . right, that’s the question we need to ask” (11:45).

References