The End Is Always Near
Eschatology in the Age of Artificial Intelligence
Eschatological claims are theological, philosophical, or ideological assertions regarding the ultimate destiny of humanity, history, and the universe, commonly referred to as the "end times" or the final state of existence. Derived from the Greek words eschatos (last) and logos (study), eschatology deals with both individual transitions (death and the afterlife) and cosmic resolutions (the climax of history, final judgment, and cosmic renewal).
Across religious and philosophical frameworks, these claims generally fall into several core structures and categories:
Core Categories of Eschatological Claims
Cosmic vs. Individual Eschatology: Cosmic eschatology focuses on the ultimate fate of the entire created order, nations, and the universe. Individual eschatology focuses on the destiny of the single human soul after death (e.g., heaven, hell, purgatory, or reincarnation).
Linear vs. Cyclical Time: Abrahamic traditions view history linearly, moving toward a definitive final conclusion. Eastern traditions often view time cyclically, where "ends" are merely transitions to new cosmic epochs or re-creations.
Futurist vs. Realized Eschatology: Some claims argue that the "end times" or kingdom of God is a future, literal event (futurist), while others argue it is an internal, spiritual, or present reality initiated in the current era (realized).
Major Traditions and Their Claims
1. Abrahamic Traditions (Linear & Messianic)
These traditions view history as a purposeful line directed by a creator, culminating in a dramatic divine intervention.
Judaism: Centers on the Olam Ha-Ba (the World to Come). Key claims include the arrival of a Messiah, the ingathering of the Jewish exiles to Israel, the rebuilding of the Temple, a universal era of peace, and the physical resurrection of the dead.
Christianity: Centers on the Parousia (the Second Coming of Jesus Christ). Mainstream claims emphasize a final judgment, the defeat of evil, the resurrection of the body, and the creation of a "New Heaven and New Earth." Interpretations vary widely regarding the nature and timing of the millennium of Revelation 20 (Premillennialism, Postmillennialism, and Amillennialism).
Islam: Centers on Yawm al-Qiyamah (the Day of Resurrection) and Yawm ad-Din (the Day of Judgment). Claims include a series of major and minor signs (such as the appearance of the Mahdi and the return of Isa/Jesus to defeat the Dajjal/Antichrist), followed by the destruction of the world, bodily resurrection, and the assignment of souls to Jannah (paradise) or Jahannam (hell).
2. Dharmic Traditions (Cyclical & Experiential)
Time is viewed as cyclical, and beings are bound to an endless cycle of birth, death, and rebirth (samsara), meaning "the end" operates on an entirely different philosophical plane.
Hinduism: Cosmic time is divided into four massive eras called Yugas. The current age, Kali Yuga (the age of darkness and decay), is prophesied to end with the arrival of Kalki, the tenth avatar of Vishnu, who will destroy wickedness, restore dharma (cosmic order), and initiate a new Satya Yuga (golden age).
Buddhism: Focuses on the eventual decline of the current Buddha's teachings (Dharma). Once the teachings are entirely forgotten, history transitions until the arrival of Maitreya, the future Buddha, who will achieve complete enlightenment and re-introduce the path to Nirvana.
3. Secular and Philosophical Eschatology
Modern secular thought has adapted eschatological structures to explain historical progress, existential risks, and technology without requiring a deity.
Political Ideologies: Marxism asserts a historical teleology where the internal contradictions of capitalism inevitably produce its collapse and the emergence of a classless, stateless communist society, essentially a secularized kingdom of peace.
Technological Eschatology: Proponents of the "Technological Singularity" claim that the exponential growth of artificial intelligence will reach a point where human history is fundamentally altered, transcending biological limitations through digital immortality or superintelligence.
Existential and Cosmic Risk: Secular scientific models map out the literal end of the world via anthropogenic catastrophes (climate collapse, nuclear war) or astrophysical timelines (the heat death of the universe, also called the "Big Freeze"). Across these traditions, a recurring assumption is that history moves toward an identifiable culmination. Whether that culmination is understood as divine judgment, political transformation, or technological transcendence, the future acquires a narrative structure. Modern information technology makes such narratives easier to construct, disseminate, and reinforce. Large language models add another dimension: the ability to synthesize disparate information into remarkably coherent explanations, even when the underlying evidence remains incomplete.
Across these traditions, a recurring assumption is that history moves toward an identifiable culmination. Whether that culmination is understood as divine judgment, political transformation, or technological transcendence, the future acquires a narrative structure. Modern information technology makes such narratives easier to construct, disseminate, and reinforce. Large language models add another dimension: the ability to synthesize disparate information into remarkably coherent explanations, even when the underlying evidence remains incomplete.
Philosophical and Psychological Functions
Eschatological claims are rarely just about predicting the future. They serve profound and immediate purposes for the communities and thinkers who hold them. The first is theodicy: by promising a final reckoning in which systemic injustices are corrected and the righteous are vindicated, they answer the problem of evil, the question of how suffering and injustice can persist in a world said to be governed by justice. Closely related is teleology. Eschatology gives history a telos, an ultimate purpose or goal, so that human suffering and historical events appear to be moving toward a meaningful resolution rather than unfolding at random.
These two functions converge on a third, which is ethical. If the present is a prelude to an ultimate evaluation, then how one lives now carries lasting weight. Eschatological claims therefore establish powerful frameworks for orthopraxy (correct conduct) and lend the present day a distinctive urgency. The end, in other words, is never only about the end; it reshapes how believers understand justice, meaning, and obligation in the here and now.
Twenty-First-Century Context
The transition from traditional, geographically isolated media to the exponential growth of the internet, social media algorithms, and Large Language Models (LLMs) has fundamentally altered how eschatological claims are generated, distributed, and internalized.
Historically, eschatological frameworks, whether biblical apocalypses, political teleologies, or cultural doomsday prophecies, relied on centralized authorities, slow-moving textual transmission, and physical communities. Today, digital infrastructure has hyper-accelerated these claims, shifting them from distant theological speculations into immediate, omnipresent, and highly personalized realities.
The comparison below summarizes this shift; the sections that follow examine the three engines driving it.
Direct Comparison: Pre-Digital vs. Hyper-Digital Eschatology
| Dimension | Pre-Digital Eras | The Internet, Social Media, & LLM Era |
|---|---|---|
| Velocity & Reach | Restricted by physical text, geographical proximity, and institutional gatekeepers. | Instantaneous, global, and frictionless. Claims scale from a single account to millions in hours. |
| Feedback Loops | Slow, localized social consensus; limited to church pews, specialized journals, or small cells. | Algorithmic confirmation bias. Metric-driven platforms reward high-arousal (fear/outrage) content. |
| Epistemic Authority | Vetted by formal clergy, academics, or established political theorists. | Decentralized and synthetic. Authority is simulated through algorithmic prestige, virality, and AI synthesis. |
| Nature of "The End" | Dominated by cosmic or divine intervention (e.g., God, the Yugas) or defined state-level shifts. | Complex, fragmented, and anthropocentric (e.g., AI singularity, engineered pandemics, algorithmic collapse). |
The Engines of Modern Amplification
1. The Algorithmic Optimization of Doom (Social Media)
In eras prior to social media, maintaining an intense state of eschatological urgency required deliberate communal assembly or continuous reading of specific literature. Today, social media algorithms operate on an engagement-maximizing model that functions as an unintentional catalyst for eschatological thinking.
The Economy of Apocalypse: Algorithms favor high-arousal emotional states: fear, anxiety, and tribal outrage. Eschatological claims inherently possess these traits. A narrative proclaiming that "the current system is collapsing" naturally outperforms a narrative of gradual progress.
Synthesized Signs and Wonders: Historically, believers looked for "signs of the times" in astronomical events or geopolitics, interpreted slowly over months. Social media provides a real-time, curation-ready stream of global crises (wildfires, financial volatility, wars). The algorithm bundles these disparate events into a cohesive timeline, creating an illusion of accelerating cosmic convergence for the user.
2. The Fragmentation and Hyper-Personalization of "The End"
Before the internet, eschatological claims were generally monolithic, bound to major institutional religions or clear secular ideologies (like Cold War nuclear anxieties). The internet has democratized prophecy, creating what can be called open-source eschatology.
Micro-Apocalypses: Individuals no longer need to adopt an entire religious apparatus to participate in an end-times narrative. They can engage with bespoke, hyper-specific secular eschatologies: the technological singularity, collapsitarianism1, hyperinflationary financial doomsday, or niche algorithmic conspiracy theories.
Echo Chambers as Prophets: The frictionless formation of digital subcultures allows fringe apocalyptic groups to achieve critical mass. Isolated individuals who historically would have lacked a community to validate their eschatological anxieties now find global reinforcement, accelerating radicalization and the perceived certainty of the coming climax.
3. LLMs and the Accumulation of Knowledge: Semantic Inevitability
The emergence of Large Language Models and vast semantic databases introduces an entirely new vector for eschatological amplification. Rather than just distributing human claims, technology now synthesizes and recombines them.
The Illusion of Epistemic Inevitability: LLMs train on the massive, historical accumulation of human thought, much of which is structurally teleological or anxiety-driven. When users prompt models to map out future trajectories, synthesize complex existential risks, or analyze geopolitical trends, the model compiles these patterns. Because AI presents information with authoritative, objective prose, it can inadvertently reify speculative anxieties into seemingly "logical" and inevitable historical conclusions.
Secularizing the Eschaton: LLMs have democratized the language of complex systems (climate data, biosecurity, artificial superintelligence). This allows traditional eschatological structures, the transition from absolute corruption to a final, perfected state, to be translated into highly sophisticated, technical, and scientific jargon. The psychological function remains identical to ancient apocalyptic literature, but it carries the prestige of data science and technical forecasting.
On-Demand Exegesis: If a user wants to find connections between an ancient prophecy, a modern corporate move, and a recent geopolitical shift, an LLM can map those semantic relationships instantly. It acts as an infinitely patient hermeneutic tool, instantly creating the complex cross-references that historical apocalyptic writers took lifetimes to construct.
The Psychological Result: Living in the "Permanent Present"
The ultimate consequence of this exponential growth is the compression of time. Traditional eschatology treated the "end" as a distant destination toward which history was steadily marching.
By contrast, the hyper-networked world creates a psychological state of imminent, ongoing crisis. Because information is delivered continuously, personalized by preference, and validated by algorithmic structures, modern individuals do not just look forward to an eschatological event; they experience the sensation of living inside an unfolding one every single day.
Dismantling the Illusion
To dismantle the Illusion of Epistemic Inevitability when interacting with highly cohesive, AI-generated narratives, users must move from passive consumption to active epistemic friction2. Epistemic friction is an intentional intervention in an information environment that introduces critical resistance through contradictory evidence, competing interpretations, and challenges to underlying assumptions. Its purpose is to disrupt unexamined certainties, expose weaknesses in prevailing explanations, and compel renewed examination of how knowledge claims are formed, justified, and defended.
Because LLMs naturally iron out contradictions and smooth over chaos into elegant, linear stories, breaking the illusion requires intentionally re-introducing structural messiness, contingency, and foundational skepticism into the prompt architecture.
If epistemic friction is to be more than a disposition, it must be applied where AI narratives are actually built: in the prompt. A language model largely mirrors the premises it is given. Frame a future outcome as inevitable, and a language model may construct a persuasive explanation around that assumption, even when the available evidence supports competing possibilities. The same responsiveness that produces the illusion can also be used to dismantle it. Each of the strategies below targets a different source of false inevitability: the assumed endpoint, the erasure of contingency, the concealment of uncertainty, and the unexamined assumptions on which the whole narrative rests.
Prompt Engineering Strategies for Epistemic Friction
1. Invert the Premise via "Premortem" and Alternative Axioms
LLMs are designed to look forward and build momentum on whatever premise you provide. To break this, force the model to look backward from a position of failure or divergence. If an AI generates a highly cohesive narrative about an inevitable outcome (e.g., an economic collapse or AI singularity), issue a follow-up prompt that establishes a completely different endpoint as historical fact.
Assume, as a counterfactual exercise, that fifty years have passed and your predicted outcome did not occur. Identify three plausible alternative developments and the assumptions in your original analysis that would have to be mistaken. Distinguish presently available evidence from hypothetical future events. Explain why each alternative is plausible without presenting imagined history as fact.
2. Introduce "Radical Contingency" Adjustments
AI models excel at macro-level trends but struggle with micro-level human chaos. You can break a linear narrative by forcing the model to account for random, irrational, or highly local human variables. Inject erratic human behavior, institutional inertia, or resource bottlenecks into the equation to disrupt the flawless flow of the AI's system-theory logic.
Reexamine your prediction for its dependence on uninterrupted trends. Identify three plausible sources of disruption, including institutional, technological, economic, or human factors. For each, explain whether evidence of the constraint already exists, how it might alter the outcome, and what developments could overcome it. Distinguish plausible contingencies from speculative possibilities.
3. Epistemic Status and Evidence Disclosure
By default, AI outputs often sound fully certain because they adopt an encyclopedic, authoritative register. You must explicitly demand that the model peel back its own confidence layer. Force the model to assign speculative weights to its assertions and call out its own gaps in knowledge.
Reexamine every major claim in your previous answer. Classify each as an established fact, an interpretation of evidence, an extrapolation, or speculation. Identify the evidence supporting it, any significant contradictory evidence, and the principal uncertainties. Explain which claims cannot presently be verified. Do not assign numerical probabilities or imply statistical confidence unless a defensible quantitative model supports them.
4. The "Deconstruct the Axioms" Drill
Because a narrative's cohesive flow can blind us to its weak foundations, you must isolate the unspoken assumptions the AI used to build the story. Strip away the beautiful prose and look strictly at the logical dependencies.
Identify the foundational assumptions necessary for your conclusion to be valid. Separate empirically supported assumptions from contested or speculative ones. For each assumption, identify the evidence that supports it, the strongest available counterargument, and the observations that would demonstrate that it is false. Determine which assumption is most critical to the overall conclusion.
5. The Missing Strategy: Independent Evidentiary Challenge
The previous four prompts challenge a model to reconsider its own conclusions, assumptions, and interpretations. Advances in model training have improved the ability of some LLMs to recognize and correct their own errors, reducing the need for routine instructions to double-check an answer. Yet improved self-correction leaves a fundamental epistemological problem unresolved: a model can produce internally consistent reasoning while relying on false premises, incomplete evidence, or unrecognized assumptions.3
An effective method of epistemic friction must therefore go beyond the model's internal capacity for correction. It should introduce independently verifiable evidence, competing interpretations, and observations capable of falsifying the original conclusion. The objective is to test the narrative against reality rather than merely ask its author to reconsider it.
Examine your conclusion against independently verifiable evidence. Identify the strongest available primary sources that support it and the strongest credible sources that contradict it. Distinguish agreement among secondary commentators from independent empirical confirmation. Explain what evidence would cause you to reject your original conclusion. If reliable sources cannot be accessed or verified, identify that limitation explicitly rather than inventing citations or claiming verification.
Cognitive Frameworks for the User (Internal Checks)
Beyond prompt engineering, breaking the illusion requires shifting your own psychological relationship with the screen.
Separate Coherence from Probability: Remind yourself that in a pattern-prediction system, grammatical elegance does not equal historical accuracy. A perfectly structured story about the future feels true to our brains because we crave narrative resolution, but it remains a mathematical simulation of human writing, not a calculation of reality. Models can perform substantive reasoning, but eloquence alone provides no assurance that a conclusion is correct.
Identify the "Laundering" of Tone: When reading an objective-sounding, passive-voice AI summary of an existential crisis, mentally translate the text into the first-person perspective of a biased human writer. Recognize that the AI is often just wrapping the anxious blog posts, speculative sci-fi, and opinion pieces it trained on in the prestigious attire of an academic briefing.
Look for the Absence of "Nothing": History is full of periods where absolutely nothing of cosmic significance happens: years of gridlock, plateauing technologies, and boring maintenance. AI models rarely generate "boring plateau" narratives unless explicitly asked, because human training data favors the dramatic. If the AI output feels like a movie script, that is often because it is copying human storytelling templates.
The enduring appeal of eschatology lies in its promise that history has a discernible destination. Religious prophecy, political ideology, and technological forecasting can each transform uncertainty into the expectation of an inevitable outcome. Artificial intelligence introduces a powerful new means of constructing and communicating such narratives, but its fluency cannot establish their truth.
Epistemic friction offers a practical response. By deliberately introducing contradictory evidence, alternative explanations, and challenges to foundational assumptions, we can test the conclusions that an AI presents with apparent certainty. The same scrutiny must apply to our own convictions. A method that challenges only conclusions we dislike simply substitutes one form of intellectual conformity for another.
The purpose of epistemic friction is to make knowledge claims answerable to evidence, logic, and the possibility of being wrong. The future remains uncertain, and the intellectual discipline required to understand it begins with recognizing that uncertainty.
Notes and Sources
- The word collapsitarian (and the related collapsitarianism) appears in Ben McGrath’s “The Dystopians”, published online January 18, 2009, and in the January 26 print issue of The New Yorker. McGrath quotes James Howard Kunstler using the term while discussing a subculture concerned with the possibility of societal and economic collapse. The article documents the term’s use, but does not establish that McGrath coined it. ↩
- Gila Sher developed an influential account of “epistemic friction” in “Epistemic Friction: Reflections on Knowledge, Truth, and Logic” (Erkenntnis 72, 2010, pp. 151–176) and her 2016 book of the same name (Oxford University Press). Her account concerns the constraints that keep knowledge grounded in reality, alongside the complementary principle of “epistemic freedom.” Sher traces aspects of the idea to Kant, Wittgenstein, and John McDowell, whose Mind and World appeared in 1994. A closer antecedent for the active, critical sense adopted here is José Medina’s treatment of “beneficial epistemic friction” in The Epistemology of Resistance (2013). Sher emphasizes epistemic grounding; Medina emphasizes productive resistance through encounters with differing perspectives. This essay applies those insights as an intentional method of inquiry, using contradiction, criticism, and independent evidence to challenge unexamined claims, especially those produced by artificial intelligence. ↩
- Research on LLM self-correction has evolved. Jie Huang et al., “Large Language Models Cannot Self-Correct Reasoning Yet” (ICLR, 2024), documented limitations of correction through prompting alone. Ryo Kamoi et al., “When Can LLMs Actually Correct Their Own Mistakes? A Critical Survey of Self-Correction of LLMs”, Transactions of the Association for Computational Linguistics 12 (2024): 1417–1440, distinguish unreliable prompted self-correction from improvements obtained through external feedback and fine-tuning. More recently, Yuxin Chen et al., “Learning to Self-Verify Makes Language Models Better Reasoners”, Proceedings of the 43rd International Conference on Machine Learning, PMLR 306 (2026): 17291–17307, found improvements from explicitly training verification capabilities. These findings concern the ability to detect or correct errors; they do not establish that an internally consistent account of an uncertain future has sound premises or independently verified evidence. ↩
