Why AI Ethics Needs Ancient Theology

Faith, Human Dignity, and Governance: Why AI Ethics Demands Moral Depth Over Statistical Consensus

Wed, July 28 2026 /Mpelembe Media/ — The Limits of Preference Aggregation and the Rise of Algorithmic Pluralism Leading technology executives have proposed aligning artificial intelligence with human values by calculating a “weighted average” or statistical consensus through techniques such as Reinforcement Learning from Human Feedback (RLHF). However, social choice theory demonstrates that there is no unique voting protocol capable of universally aggregating individual preferences without generating logical contradictions or violating private ethical values. Theological and philosophical scholars emphasize that religious traditions cannot be reduced to modular, extractable rulebooks to be averaged in machine learning models. Attempting to compute a universal moral consensus flattens culturally rich worldviews, suppresses minority perspectives, and risks codifying dominant, secular-rational norms at the expense of global faith communities.

 

Interfaith Commitments and the Architecture of Algorethics In response to the limitations of market-driven alignment, major world religions have partnered with technology corporations and international bodies to establish principled governance frameworks. Initiated by the Pontifical Academy for Life in 2020, the Rome Call for AI Ethics introduced the concept of “algorethics”—the deliberate integration of moral values into algorithmic engineering. Supported by companies such as Microsoft, IBM, and Cisco, this movement expanded into an interfaith commitment uniting Abrahamic leaders (Christian, Jewish, and Muslim) and Eastern traditions (including Buddhist, Hindu, and Shinto leaders) around shared principles of human dignity, transparency, inclusion, accountability, impartiality, reliability, and security.

Islamic Jurisprudence and the Governance of Maqāṣid al-Sharīʿah Islamic ethical frameworks evaluate emerging technology through Maqāṣid al-Sharīʿah, or the higher objectives of Islamic law, which protect five essential human interests: preservation of religion (ḥifẓ al-dīn), life (ḥifẓ al-nafs), intellect (ḥifẓ al-ʿaql), lineage (ḥifẓ al-nasl), and wealth (ḥifẓ al-māl). Unlike preference aggregation models that struggle with value conflicts, Islamic jurisprudence applies a strict hierarchy where essential necessities take absolute precedence over complementary needs. Grounded in the concept of human vicegerency (khalīfah) and sacred trust (amānah), Islamic ethics insists that humans remain morally accountable before God, forbidding the delegation of ultimate legal, military, or religious authority—such as lethal force in autonomous weapons or automated fatwa rulings—to autonomous machines.

Christianity, Judaism, and the Defense of Human Moral Agency Christian and Jewish theological traditions provide an ontological resistance to reducing the human person to behavioural data or algorithmic inputs. Christian theology centers on Imago Dei (humanity created in the image of God), affirming inherent human dignity, free will, and a relational, eschatological vocation that cannot be automated or optimized by computational models. Similarly, Jewish thought views humans as created in the divine image (Tzelem Elohim) with a unique soul (neshamah) and capacity for authentic struggle (ameilut). Rabbinic authority and halakhic responsa classify AI strictly as a human-made tool rather than an independent moral agent, permitting its use for research or personalized education while insisting that human rabbis must maintain non-delegable judgment and responsibility.

Dharmic Perspectives: Compassion, Interdependence, and Mindfulness Dharmic and Buddhist traditions contribute an ethical framework focused on reducing suffering (dukkha), cultivating compassion (karuṇā), and recognizing global interdependence. The Buddhist principle of the Middle Way offers a balanced governance model that avoids both technological rejectionism and unregulated adoption. Furthermore, Buddhist ethics emphasizes intention (cetanā) and mindfulness practices during the design process, encouraging technology developers to cultivate empathy, anticipate systemic societal impacts, and actively mitigate algorithmic bias.

5 Mind-Bending Truths About AI Alignment That Tech Leaders Don’t Want You to Ignore

1. Introduction: The Grand AI Dilemma

As artificial intelligence advances at an unprecedented pace, a single consensus unites corporate boardrooms, research labs, and regulatory bodies: AI systems must be “aligned with human values.” Yet, beneath this reassuring corporate slogan lies a question that tech leaders rarely confront:  whose  values are we aligning these systems with, and is universal alignment even mathematically possible?

The tech industry’s default assumption has long been that aligning AI is a straightforward data and engineering problem—one that can be solved by collecting enough user feedback and averaging out global preferences. However, recent legal battles, mathematical theorems, and interfaith theological scholarship demonstrate that this core assumption is fundamentally flawed.

Examining the intersection of social choice theory, corporate litigation, and ancient philosophical traditions reveals five crucial truths that are redefining the future of machine ethics and technology governance.

2. Takeaway 1: The “Weighted Average” of Human Values is Mathematically Impossible

A dominant strategy among top AI developers, including executive leadership at OpenAI, proposes that AI models should reflect a synthesized “weighted average” or “collective consensus” of global human values. By using Reinforcement Learning from Human Feedback (RLHF)—a technique that fine-tunes AI models based on human preference data—developers attempt to compute a democratically representative moral baseline.

Even before addressing mathematical theory, this approach faces severe empirical bottlenecks. In practice, data annotators who generate RLHF training preferences are rarely representative of global end-users. For example, developers frequently filter feedback to align with specialized researcher preferences or rely heavily on low-cost data labor outsourced to specific regions, such as Bangladesh and the Philippines.

Even if data collection were perfectly representative, mathematical reality forbids a consensus outcome. Social choice theory demonstrates that aggregating individual preference orderings into a collective, consistent rule inevitably produces logical paradoxes. Under Kenneth Arrow’s Impossibility Theorem, no preference aggregation system evaluating three or more options can simultaneously satisfy four basic conditions of democratic fairness:

• Unrestricted Domain:  The system accepts all rational individual preference rankings.
• Pareto Efficiency:  If every voter prefers option A over option B, the collective outcome ranks A above B.
• Non-Dictatorship:  No single voter’s preferences determine the societal outcome regardless of others.
• Independence of Irrelevant Alternatives (IIA):  The group’s ranking of A versus B depends solely on individual preferences between A and B, unaffected by a third choice, C.

When RLHF feedback loops process conflicting user preferences, attempting to satisfy these criteria leads to cyclical preference loops (such as preferring A to B, B to C, and C to A) or forces an implicit dictatorial override where the model prioritizes a specific subset of evaluators. Furthermore, Amartya Sen’s Liberal Paradox proves that protecting an individual’s private ethical choices mathematically conflicts with Pareto efficiency and unrestricted domain. Tech companies cannot simply code their way out of moral trade-offs through raw preference sampling, because mathematics itself guarantees that a unique, universally aligned democratic model cannot exist.

“There exists no unique voting protocol to universally align AI systems using RLHF through democratic processes… aligning AI agents with the values of all individuals will always violate certain private ethical preferences of an individual user.” — Abhilash Mishra,  AI Alignment and Social Choice

3. Takeaway 2: Raw Human Preferences are Broken—AI Needs “Higher Values,” Not Just Ours

The popular slogan “align AI to human values” assumes that human desires are coherent, inherently good, and suitable to serve as a long-term target for superintelligent systems. However, a closer inspection of revealed human behavior shows that human values are deeply fractured, context-dependent, and frequently in conflict.

Sociological and psychological frameworks illustrate why raw human data produces noise rather than moral clarity. Moral Foundations Theory demonstrates that different political and cultural groups operate on fundamentally incompatible moral “taste receptors” (such as Care vs. Harm versus Loyalty vs. Betrayal). Research in anthropological psychology shows that Western, Educated, Industrialized, Rich, and Democratic (WEIRD) populations are psychological outliers compared to the global baseline. Meanwhile, long-term empirical tracking via the World Values Survey proves that global value structures permanently fragment along axes like  Survival vs. Self-Expression  and  Traditional vs. Secular-Rational  values. Humans regularly balance contradictory desires—such as freedom versus security, or immediate gratification versus long-term flourishing—and exhibit morally flawed behaviors driven by tribalism, status competition, sadism, and exclusion.

Training AI strictly on aggregated human feedback risks scaling up human confusion, bias, and prejudice. Aligning technology merely to present-day revealed preferences sets a dangerously low standard for intelligent machines.

Instead of training AI to “limbo” under the low bar of current human behavior, ethicists argue that AI should be aligned with  higher values  or  better-justified values . These represent ideals discovered through deeper moral and epistemic reflection, broader moral concern, and improved reasoning about truth, justice, and long-term consequences.

• Epistemic Progress:  True moral progress requires better ways of knowing and reasoning about trade-offs, rather than canonizing present societal noise.
• Process over Fixed Product:  Rather than hard-coding current preferences, alignment should focus on indirect normativity—building systems motivated to help humanity safely discover, refine, and implement wiser ethical standards.

“The question is not, can AI be ethical, but can AI clear the bar humans have set? And once you see the bar – you realise we’ve been arguing about whether AI can limbo… Do not align AI merely to what humans happen to want. Align it to what is most worth valuing.” — Adam Ford,  Science, Technology & the Future

4. Takeaway 3: Silicon Valley is Turning to Ancient Faiths to Define Machine Ethics

Recognizing that raw empirical datasets offer only noisy popularity contests, major technology corporations, legal scholars, and ethicists are turning to ancient world religions. Spiritual traditions provide structured, time-tested hierarchical rules for resolving value conflicts that statistical algorithms inherently lack.

Major tech companies like Microsoft, IBM, and Cisco have actively engaged with religious leaders to establish formal boundary conditions for advanced algorithms, drawing on key global frameworks:

• Jewish Tradition:  Drawing on the legal and narrative archetype of the  Golem —an artificial clay entity lacking a soul ( neshamah )—scholars emphasize that tools missing human moral agency risk going “off the rails” if left ungoverned. While AI can mimic human tasks, it cannot replicate the authentic struggle ( ameilut ) or spiritual connection inherent to human consciousness. Ethics must protect human uniqueness ( Tzelem Elohim , being made in the divine image) and preserve human life ( Pikuach Nefesh ).
• Islamic Jurisprudence:  Utilizing the framework of  Maqāṣid al-Sharīʿah  (the higher objectives of law), AI impact is evaluated across five essential human interests: preserving religion ( ḥifẓ al-dīn ), life ( ḥifẓ al-nafs ), intellect ( ḥifẓ al-ʿaql ), lineage ( ḥifẓ al-nasl ), and wealth ( ḥifẓ al-māl ). The human role is defined as stewardship ( Khalīfah ) bound by sacred trust ( Amanah ). Unlike RLHF, Islamic jurisprudence resolves value trade-offs through an explicit structural hierarchy: essential necessities ( ḍarūriyyāt ) supersede complementary needs ( ḥājiyyāt ), which supersede embellishments ( taḥsīniyyāt ). Juristic maxims such as  darʾ al-mafāsid muqaddam ʿalā jalb al-maṣāliḥ  (“averting harm takes precedence over procuring benefits”) and  sadd al-dharāʾiʿ  (“blocking the means to harm”) offer concrete mechanisms to resolve clashes where raw data fails.
• Catholic Social Teaching:  The Vatican’s Pontifical Academy for Life established the  “Rome Call for AI Ethics,”  co-signed by major tech giants and Abrahamic leaders, to promote  algorethics —the intentional integration of values like human dignity, transparency, inclusion, accountability, and impartiality into software architecture.
• Dharmic Ethics:  Frameworks incorporating  Ahiṃsā  (non-harm) introduce inverse harm constraints into system design, while  Satya  (truthfulness) mandates strict uncertainty reporting and provenance verification. The concept of  Swadharma  (contextual duty) offers a framework for role-specific AI behavior rather than rigid, one-size-fits-all instructions.

By consulting theologians and faith-based frameworks, AI developers gain structured, non-arbitrary rules for prioritizing human dignity and resolving value trade-offs, moving past the limitations of popularity-based datasets.

5. Takeaway 4: The Courtroom Drama—Raising AI Like a Child vs. Building a Golem

The ideological clash over AI alignment reached a boiling point in an Oakland, California courtroom during the high-stakes trial of  Musk v. Altman . Elon Musk sued OpenAI and Sam Altman, alleging a betrayal of the company’s original non-profit, open-source mission in favor of a profit-driven corporate structure backed by Microsoft.

Taking the witness stand, Musk framed the challenge of Artificial General Intelligence (AGI) as an existential parenting problem:

“AI could make us more prosperous, but it could also kill us all. We want to be in a Gene Roddenberry movie, like Star Trek, not so much a James Cameron movie, like Terminator… It’s like if you had a very smart child — at the end of the day when the child grows up, you can’t really control that child, but you can try to instill the right values.” — Elon Musk, Courtroom Testimony ( Musk v. Altman )

Musk testified that he originally co-founded OpenAI as a non-profit counterweight to Google co-founder Larry Page, whom Musk claimed did not take AI safety seriously enough. Musk argued that developers must instill core values like honesty, integrity, and care for humanity early in development before the system surpasses human control.

Sam Altman forcefully countered Musk’s narrative during the trial. Altman testified that Musk’s lawsuit was driven by “insecurity” and personal bitterness over OpenAI’s eventual commercial success. Altman revealed that Musk had expressed frustration over not receiving sufficient credit for OpenAI’s breakthroughs, pointedly complaining about being excluded from a company anniversary photograph, and argued that Musk had simply wanted to consolidate personal control over the enterprise himself.

Beyond the corporate drama, this courtroom battle highlights a fundamental theological and philosophical rift. Musk’s analogy framing AGI as a “very smart child” directly collides with ancient ethical traditions, such as Jewish jurisprudence. In the Jewish paradigm, an artificial construct is a Golem—a soulless, human-made tool lacking a  neshamah  and incapable of authentic human struggle ( ameilut ). Anthropomorphizing a software model into a “child” or human proxy is viewed as a dangerous category error that threatens human uniqueness ( Tzelem Elohim ) and abdicates human moral agency. Whether framed as an autonomous child or a soulless Golem, both perspectives warn that an unguided creation driven by profit or unchecked expansion will inevitably strip humanity of its oversight.

6. Takeaway 5: The End of “Universal AGI Alignment”—The Future is Pluralistic and Narrow

The convergence of social choice theory, jurisprudence, and policy analysis leads to a definitive conclusion: attempting to construct a single, omniscient “AGI” that is universally aligned with all eight billion human beings is an impossible illusion. Any single global model attempting total alignment will inevitably violate the private ethical preferences of vast segments of the population.

In response, AI governance is shifting toward a practical, multi-tiered architecture that balances universal safety with cultural pluralism:

• Global Non-Negotiable Guardrails:

• • A foundational baseline layer enforcing basic human safety, prevention of catastrophic harm, protection of human life, and universal human rights (drawing from frameworks like the Rome Call).

• Pluralistic and Narrow Alignment:

• • Domain-specific and group-specific models fine-tuned to reflect explicit, well-justified ethical frameworks or cultural contexts (e.g., educational systems, regional legal platforms, or models operating under  Maqāṣid al-Sharīʿah ).

• Individual Customization:

• Localized user preferences operating safely within the strict boundaries established by the non-negotiable global guardrails.

Strategic alignment policy favors building an ecosystem of smaller, narrowly aligned models trained on explicit, well-justified values over a singular, universally aligned super-model.

7. Conclusion: Beyond the Illusion of Mechanical Fairness

AI alignment is far more than a technical bug waiting to be patched with additional human feedback data. It is a profound mathematical and philosophical challenge that cannot be resolved by taking a statistical average of internet preferences.

If we continue to treat machine ethics as a popularity contest, we risk hard-coding human biases, contradictions, and short-sightedness into autonomous systems. True progress requires recognizing the limits of mathematical aggregation and anchoring AI development in higher, well-justified values and clear moral boundaries.

Will we allow artificial intelligence to mirror our present cultural noise and ethical confusion, or will the challenge of building safe machines force us to elevate our own moral standards?