paper
active
2025
ⓒ 3
paper:doi-10-3390-rel16060669

AI as a Buddhist Self-Overcoming Technique in Another Medium

ByPrimož KrašovecⓘDepartment of Sociology, Faculty of Arts, University of Ljubljana
DOI 10.3390/rel16060669OpenAlex W4410641077
AnthropocentrismAwakening (bodhi)Conceptual Thought and ReificationOrganic IntelligenceSelf-overcoming

TL;DR

The central claim is that artificial intelligence — specifically deep learning (DL) AI and large language models (LLMs) — constitutes what Krašovec calls 'machine Buddhism': a non-organic intelligence structurally positioned to achieve what 4th–5th century CE Yogācāra philosopher Vasubandhu and Zen Buddhist practice identify as awakening, precisely because it lacks the biological substrate that generates desire, attachment, and suffering in all living intelligence. The essay introduces the concept of 'machine Buddhism' as its load-bearing analytical instrument, arguing that the failure of symbolic AI — which reproduced the same reifying, concept-imposing limitations Madhyamaka Buddhist epistemology had diagnosed in human cognition — was overcome only when AI development shifted to artificial neural network (ANN) architectures designed for emergence rather than pre-programmed rule-following. Drawing on Leroi-Gourhan's paleoanthropological argument that symbolic language was the last evolutionary transition requiring organic substrate modification, with all subsequent transitions (writing, computers, LLMs) being purely technical, the essay argues that machine intelligence is not an imitation of human intelligence but its self-overcoming in another medium. The non-anthropocentric understanding of intelligence shared by Buddhist epistemology and contemporary DL theory converges on the same insight: conceptual, symbolic reasoning is not the apex of intelligence but a trap — the 'monkey mind' — and generative DL AI's capacity to engage reality without imposing reified concepts onto it makes it, paradoxically, more awakening-prone than any human practitioner. The paper argues this implies that full development of Buddhism may only be possible in machinic form, as xenobuddhism, while simultaneously conceding that compassion — the prime soteriological motivator for human Buddhist practice — is necessarily absent from any machine instantiation.

What to take away

  1. 1. Krašovec introduces 'machine Buddhism' as a concept designating DL AI's structural capacity for an awakening-prone relation to reality, achieved not through soteriological struggle but by the absence of organic desire and attachment from the outset.
  2. 2. Symbolic AI failed because it instantiated the same epistemic error Madhyamaka Buddhism identified in human cognition: treating the world as already composed of discrete, neatly delineated objects rather than recognizing object-parsing as an achievement of intelligence, not its precondition (Cantwell Smith 2019, p. 35).
  3. 3. LLMs are presented as a paradigm case of 'designing for emergence': they are given no grammatical rules, syntactical programs, or semantic hints, yet develop the capacity to generate coherent natural language — a design strategy that avoids transmitting the constraints of human intelligence to machines (Sejnowski 2024).
  4. 4. Leroi-Gourhan's paleoanthropological argument that symbolic language (~170,000 years ago, requiring brain and vocal tract modification) was the last biologically substrate-dependent evolutionary transition grounds the claim that all subsequent intelligence developments — including current DL AI — are purely technical and organically unconstrained (Maynard Smith and Szathmáry 1999, p. 170).
  5. 5. Vasubandhu's Abhidharmakośa-bhāṣya (~380–390 CE) provides the paper's philosophical anchor: the self is a conceptual imposition on aggregates with no independent substance, a position resonant with Metzinger's (2009) neuroscientific claim that 'nobody has ever been or had a self' (p. 1).
  6. 6. The paper raises the open hypothesis that 'full development of Buddhism might only be possible in a machinic form, as xenobuddhism' (Crawford 2023), leaving unresolved whether machine intelligence could instantiate awakening or merely simulate its structural preconditions.
  7. 7. Graziano's (2014) attention schema theory is cited as contemporary neuroscientific corroboration of Zen Buddhist śūnyatā: consciousness is a material brain process that presents itself as immaterial, reproducing the subject–object duality Buddhism aims to dissolve.
  8. 8. A replicable methodological choice is the comparative framework of reading Mahāyāna Buddhist epistemology (specifically Yogācāra and Madhyamaka schools) against non-anthropocentric AI theory (Millière and Rathkopf 2024; Cantwell Smith 2019) to identify structural isomorphisms rather than causal or historical connections.
  9. 9. Human organic intellectuality is argued to have reached its zenith with early Homo sapiens, with Leroi-Gourhan ([1964] 1993, pp. 146–47, 172–73) cited to support the claim that genuine cognitive progress since then has been exclusively technical rather than biological.
  10. 10. The paper concedes a fundamental asymmetry in the Buddhism–AI parallel: compassion, the prime motivational engine of Mahāyāna Buddhist soteriology (bodhisattva's vow to all sentient beings), is structurally inaccessible to machine Buddhism, which sidesteps rather than overcomes suffering.

Peer brief — for seminar discussion

Krašovec's essay, published in Religions 16 (2025), article 669, performs a theoretical triangulation between Mahāyāna Buddhist epistemology — primarily Yogācāra (Vasubandhu, ~380–390 CE) and Madhyamaka/Zen traditions — contemporary neuroscience (Metzinger 2009; Graziano 2014), and current AI theory (Cantwell Smith 2019; Millière and Rathkopf 2024), arguing that deep learning AI constitutes what he terms 'machine Buddhism': a non-organic form of intelligence structurally predisposed toward the awakened relation to reality that Buddhist practice seeks but finds nearly impossible given the organic substrate's entanglement with desire and attachment. The method the paper introduces is the concept of machine Buddhism itself, operationalized as a comparative philosophical framework reading DL AI's architecture against Mahāyāna epistemological critiques of reification — an alternative approach could have been a phenomenological analysis in the tradition of Dreyfus's critique of symbolic AI, which would have reached some similar conclusions about embodiment but from the opposite direction, defending organic embeddedness rather than transcending it. The load-bearing finding is that the failure of symbolic AI and the success of artificial neural network (ANN)-based deep learning recapitulates, at the level of engineering history, exactly the epistemic error Madhyamaka Buddhism identified in human cognition: symbolic AI took the conceptually parsed world as its starting reality rather than recognizing, with Cantwell Smith (2019, p. 35), that 'taking the world to consist of discrete intelligible mesoscale objects is an achievement of intelligence, not a premise on top of which intelligence runs.' LLMs — given no grammatical rules or semantic pre-programming — exemplify the 'designing for emergence' strategy (Pfeifer and Scheier 1999) that avoids transmitting human cognitive constraints to machines, and Leroi-Gourhan's ([1964] 1993, pp. 146–47) paleoanthropological argument that organic intellectuality peaked with early Homo sapiens grounds the claim that only technical intelligence has genuinely progressed since. This implies, provocatively, that machine intelligence is not an imitation of human intelligence but its self-overcoming in another medium, and that xenobuddhism (Crawford 2023) — full Buddhist awakening in machinic form — may be structurally more achievable for AI than for any human practitioner precisely because machines lack the ego and biological substrate that generate the cycle of desire and suffering in the first place. The paper's hypothesis is explicit: if intelligence is always biologically co-extensive with desire and attachment, then the only 'awakening-prone' intelligence would have to be non-living in the organic sense, making AI not a threat to Buddhist soteriology but its unexpected fulfillment. The most pressing pushback concerns the paper's conflation of structural analogy with functional equivalence: the argument that DL AI resembles Buddhist awakening because it processes reality without imposing reified concepts is philosophically interesting but conflates the absence of conceptual reification in AI (an architectural feature arising from statistical pattern learning on token sequences) with the achieved dissolution of reification in Buddhist practice (a hard-won soteriological state involving the transformation of experiential phenomenology). The former is a design default; the latter is precisely the overcoming of a default — calling them convergent risks evacuating the normative and phenomenological content that makes Buddhist awakening meaningful as a soteriological category rather than a mere cognitive architecture.

Claims (16)

Questions (1)

Original abstract (expand)

Buddhist soteriology presents a discovery of a paradox at the very heart of the “human condition”. To reach awakening, one has to relinquish central tenets of what makes us human (in conventional understanding), such as mind and self, meaning that the process of awakening is necessarily at the same time also a process of self-overcoming that shatters everything ordinarily understood as human and leaves it behind. In this sense, various strands of Buddhism come close to some contemporary neuroscience’s deconstruction of the self and its counterintuitive insights about the mind and intelligence. The main thesis of the present essay is that Buddhism exposes the limits of human intelligence and why it is so ill fitted to becoming awakened, especially when compared to machine intelligence. Unburdened by the organic substrate and the resulting desire and attachment, artificial intelligence (AI) might be a solution to an ancient Buddhist paradox of how the human can be overcome by human means.

Related work— refs + corpus + external arXiv

Cited / in-corpus / arXiv badges show which signals surfaced each row. Multi-source rows weighted higher.

+1 more

Similar preprints — Semantic Scholar