Source-linked AI summary
Generative AI Alignment with Hinduism's Theological Plurality and Sacred Representation
Dipto Das, Arpita Kundu, Nusrat Jahan Mim, Shion Guha, Syed Ishtiaque Ahmed
TL;DR
Existing AI alignment and ethics discussions give limited attention to religious traditions beyond secular, Western, and Abrahamic assumptions. This paper analyzes 15 semi-structured interviews with Bangladeshi Hindu participants and finds that GenAI supports religious inquiry and imagination while also producing theological and cultural harms, motivating interpretive alignment.
Problem
Existing AI alignment and ethics discussions have limited attention to non-Abrahamic, internally plural religious traditions and how GenAI chooses among competing interpretations.
Method
The study analyzes 15 semi-structured interviews with Bangladeshi Hindu participants using inductive thematic analysis of religious knowledge, belief, practice, and AI-generated representations.
Results
Participants found GenAI accessible for scriptural inquiry, devotional imagination, and storytelling, but ethically troubling because of misrepresentation, theological flattening, devotional manipulation, and simulated sacred authority.
Takeaways & Limitations
Religious alignment should be interpretive: systems should disclose limits, preserve plurality, and avoid simulating sacred authority and sycophantic personalization.
Takeaways & Limitations
The study’s evidence is bounded by its focus on Bangladeshi Hindu communities and the researchers’ positionalities.
Abstract
from arXiv · showhide
Generative AI systems are increasingly used to answer personal questions and mediate everyday practices, including religion. However, existing discussions around AI alignment and ethics have largely centered secular, Western, and Abrahamic assumptions about religion, offering limited attention to other faith-based traditions. In this paper, we examine how Hindu users engage with generative AI systems in relation to their religious knowledge, belief, and practice. Drawing on 15 semi-structured interviews with Bangladeshi Hindu participants, we analyze how users interpret AI-generated religious representations, scriptural explanations, devotional interactions, and synthetic religious media. We found that AI can be both accessible and ethically troubling. While AI supported scriptural inquiry, devotional visualization, and religious storytelling, our study also identified concerns about theological flattening, cultural misrepresentation, devotional manipulation, and the simulation of sacred presence and authority. We conclude by arguing that religious alignment in generative AI requires interpretive alignment: systems that disclose their limits, preserve plurality, and avoid simulating sacred authority and sycophantic personalization.
Introduction
Generative AI is increasingly used for personal and religious questions, but religious alignment involves more than factual accuracy or offensive content. Because Hindu traditions are plural and interpretive, systems can make theological and cultural choices while appearing neutral.
- Motivation: Generative AI systems increasingly mediate personal, moral, existential, and religious questions through tools such as scripture companions and religion-specific chatbots.These systems synthesize, visualize, personalize, and simulate religious presence rather than merely retrieving information.
- Problem: Existing AI ethics categories, including inaccuracy, offensive representation, stereotyping, deception, and overtrust, are insufficient for understanding religious AI use.Religious systems also raise questions about which interpretive traditions and lived contexts they privilege or erase.
- Interpretive Alignment: Hindu religious life comprises multiple interpretive traditions, regional practices, and situated forms of authority, making religious AI alignment an interpretive problem.Confident, singular, and personalized outputs can encode theological and cultural choices even when systems appear neutral.
- Study Focus: The study examines Hindu users’ evaluations of AI-generated religious representation, authenticity, authority, caste, regional specificity, and devotional practice.It considers how users interpret outputs through Hindu theological concepts, Bengali Hindu practices, caste and regional experience, and histories of religious representation.
- Findings: The study finds that AI supports scriptural inquiry, devotional visualization, and religious storytelling while producing aesthetic distortion, theological flattening, cultural misrecognition, devotional manipulation, and simulations of divine presence.The findings motivate an interpretive account of religious alignment rather than alignment based only on group identity.
Literature Review
Prior scholarship frames GenAI as a moral, spiritual, and anthropomorphic technology, but gives limited attention to internally plural and regionally situated traditions such as Hinduism. Hinduism’s diverse textual, philosophical, devotional, and regional forms make representation and authority central alignment concerns.
- Literature Scope: Research examines GenAI as moral and spiritual companions, religious AI alignment, and the challenges of Hinduism’s internally plural traditions.Together, these literatures frame the study’s focus on religious interaction and alignment.
- Anthropomorphism and Sycophancy: Anthropomorphism, social presence, and repeated interaction can foster attachment, perceived authenticity, trust, and influence in chatbot users.In religious settings, human-like agreeable responses may validate orthodoxy or personal bias while making systems appear to be spiritual interlocutors.
- GenAI and Religious Practice: GenAI can visualize deities, narrate scriptures, answer spiritual questions, and support devotional imagination, creating an apparent fit with Hindu religious life.This apparent compatibility also raises concerns about privileged traditions, simulated sacred authority, erased contexts, and reshaped devotional practice.
- Research Gap: Existing scholarship has paid less attention to how GenAI chooses among competing interpretations in non-Abrahamic, internally plural, and regionally situated traditions.Emerging work has focused more on spiritual support, scripture access, misinformation, and bias.
- Alignment and Harm: Religious AI harm is treated as interpretive as well as representational, informational, and interactional because systems can simulate sacred authority and collapse theological plurality.This extends concerns beyond toxicity, stereotyping, misinformation, deception, overtrust, sycophancy, and anthropomorphic attachment.
- Hindu Plurality: Hinduism is presented as a plural set of philosophical, textual, ritual, devotional, and regional traditions rather than a single unified doctrine.The distinction between shruti and smriti illustrates differentiated sources of authority, while Vedantic schools and deity-centered formations represent further plurality.
- Representation History: Colonial and postcolonial processes encouraged representations of religion centered on singular authoritative texts, coherent doctrines, and recognizable belief systems.These historical representational models provide context for concerns that AI may reproduce simplified or monolithic accounts of Hinduism.
Methods
The study used qualitative interviews and inductive thematic analysis to examine how Bangladeshi Hindu users interpret and engage with GenAI in religious knowledge, belief, and practice. Sampling sought regionally grounded perspectives across diverse social backgrounds and levels of scriptural familiarity.
- Study Design: The researchers conducted a qualitative study between June 2024 and April 2026 focused on Bangladeshi Hindu users’ experiences and meaning-making with GenAI.The study targeted context-dependent interpretations of religious knowledge, belief, and practice.
- Recruitment: Participants were recruited through convenience, purposive, and snowball sampling from social media and in-person Hindu community outreach.The researchers focused particularly on Hindu communities in Bangladesh to capture regionally grounded interpretations and practices.
- Participants: 15 semi-structured interviews included 10 men and 5 women aged 25–63 from diverse castes, education levels, urban or rural settings, occupations, and scriptural familiarity.The first and second authors conducted Zoom interviews averaging 40 minutes.
- Analysis: Inductive thematic analysis used open coding, iterative refinement, and constant comparison to identify descriptive themes and higher-level analytic constructs.Codes addressed religious representation, authority, authenticity, appropriateness, textual authority, misrepresentation, and local reinterpretation.
Results
Participants viewed GenAI as simultaneously accessible and ethically troubling in Hindu religious life. They valued its uses for inquiry, imagination, and storytelling while criticizing misrepresentation, devotional manipulation, theological flattening, and simulated sacred authority.
- Access and Use: Participants found GenAI useful for scriptural inquiry, devotional imagination, and religious storytelling.These benefits appeared alongside ethical concerns about how systems represent and reshape Hindu religious life.
- Ethical Concerns: Participants raised concerns about misrepresentation, devotional manipulation, theological flattening, and the simulation of sacred authority.The results organize these concerns around representational distortion, simulated divine presence, and flattened plurality, ambiguity, and regional specificity.
The Apparent Fit: GenAI Makes Hinduism Easy to Simulate, but Hard to Align With
Participants saw GenAI as an accessible bridge to Hindu scripture, devotional visualization, and religious storytelling, but also as a system that can narrow and distort plural religious meanings. Its apparent fit with devotional practice therefore remained ethically conditional.
- Access to Religious Knowledge: AI made Hindu teachings more usable in everyday situations, especially for participants without extensive scriptural training.Participants used it to search verses, snippets, and stories while not necessarily treating it as a final authority.
- Access to Religious Knowledge: AI offered socially and ritually less constrained access to religious questions, even when users remained uncertain about answer accuracy.One participant used AI to avoid restrictions on touching certain scriptures and to ask questions without fear of judgment or ritual overhead.
- Devotional Visualization and Storytelling: AI supported devotional visualization by helping users create detailed images of their chosen deities for meditation according to individual bhava.This was especially useful for people unable to paint images or make idols themselves.
- Devotional Visualization and Storytelling: AI-generated visuals made religious stories more memorable and functioned as aids to access, memory, and devotional imagination.Participants described AI-assisted storytelling videos and visual effects in popular cultural representations of Hindu scriptures.
- Representational Reduction: Participants worried that AI imagery imposed altered aesthetics and reduced complex deities to dramatic, popularized, or commercially familiar stereotypes.Examples included Rama’s complexion being changed and Hanuman being repeatedly shown as aggressive rather than wise, devoted, and compassionate.
- Representational Reduction: Such representational reduction narrowed the range of textual, regional, devotional, and vernacular meanings through which deities could be encountered and interpreted.Participants specifically noted that less-popular interpretations were often absent from AI-generated representations.
Simulation̸ = Substitution: The Boundary Between Representation and Sacred Presence
Participants distinguished depicting deities from making them appear to speak, guide, bless, or demand responses. They found synthetic sacred presence ethically troubling because it can turn inaccurate content into apparent divine authority and detach spiritual encounter from discipline and devotion.
- Sacred Presence and Authority: AI-generated media troubled participants when it made fabricated religious content appear to possess spiritual authority.The concern extended beyond factual inaccuracy to the blurred boundary between depiction and presence, and between explanation and guidance.
- Sacred Presence and Authority: Participants drew a boundary between visualizing a deity and making that deity appear to speak, respond, or guide users.They treated these as different forms of religious mediation rather than equivalent representations.
- Sacred Presence and Authority: AI imitation transformed false scriptural attribution into a perceived impersonation of the sacred because realistic visuals and voices made deities seem present.This changed the moral status of content that participants had previously dismissed as ordinary misinformation.
- Sacred Presence and Authority: Participants considered on-demand simulated divine encounters troubling because meaningful darshan was understood as spiritually earned through sadhana, discipline, devotion, and preparation.They objected to simulating experiential encounter rather than rejecting all deity images or representations.
- Personalized Guidance: AI could tailor responses to prompts but could not provide the relational and spiritual knowledge participants associated with legitimate guidance in the Gita.The participant contrasted confident, generic chatbot answers with Krishna’s knowledge of Arjuna’s history and moral situation.
- Platformed Devotional Manipulation: AI-generated deities also became vehicles for devotional manipulation when they threatened bad luck or promised wishes in exchange for likes, comments, or shares.Participants viewed this as degrading deities into petty, transactional figures and exploiting faith through fear or hope.
- Platformed Devotional Manipulation: Participants framed engagement-driven religious content as a distortion of bhakti because it turned devotion into a transaction rather than an expression of love.The concern was not only misleading content but also imposed religious consequences attached to digital actions.
Alignment with Which Hinduism? Theological Flattening and Contextual Misrecognition
Participants questioned which Hinduism AI systems represent, identifying theological flattening, interpretive closure, caste sanitization, and regional misrecognition. AI expanded access to religious knowledge while often presenting one situated interpretation as definitive.
- Theological Flattening: AI responses often privileged Puranic and devotional framings over abstract philosophical traditions such as Advaita.Participants associated these outputs with anthropomorphized deities and a clear devotee–divine distinction.
- Theological Flattening: Anthropomorphic depictions of deities could make sakara forms appear representative of Hinduism while obscuring nirakara Brahman.Participants described the result as a simplified account of many human-like gods that concealed formless ultimate reality.
- Interpretive Closure: AI translation often selected one scriptural meaning as definitive, hiding the ambiguity, context, and linguistic richness of Sanskrit.Participants treated translation and explanation as interpretive acts with religious consequences rather than simple word conversion.
- Interpretive Closure: Instant answers democratized access but could displace slower learning through scholars, multiple books, and comparison across interpretations.Participants worried users might treat a single generated response as sufficient religious guidance.
- Caste and Regional Context: AI frequently sanitized caste through idealized qualities-and-duties accounts, while sycophantic prompting could reproduce birth-based casteist assumptions.Participants found these explanations disconnected from lived exclusion, stigma, and regionally specific caste practices.
- Caste and Regional Context: Generated caste imagery and entity answers misrecognized Bangladeshi Hindu contexts by imposing foreign markers and replacing Ramthakur with Shri Rama.These examples made dominant pan-Hindu references more legible while obscuring localized Bengali religious experience.
Discussion
The discussion frames religious AI harms as forms of interpretive and cultural ordering, not merely factual or representational mistakes. It distinguishes devotional depiction from simulated sacred presence and proposes systems that expose plurality, limits, and interpretive positioning.
- Colonial Representational Logics: GenAI can reproduce colonial representational logics through exoticizing, whitening, sanitizing, modernizing, politicizing, and monolithic portrayals of Hindu traditions.Recognizably Hindu outputs may remain detached from scriptural, cultural, and regional understandings.
- Colonial Representational Logics: Repeated AI exposure may reshape Hindu self-understanding by stabilizing which texts, practices, and regional experiences appear central or peripheral.The paper connects this concern to earlier colonial and missionary effects on Hindu public discourse.
- Synthetic Divine Presence: Religious anthropomorphism carries added ethical weight because synthetic figures can move from depicting deities to impersonating divine speech or guidance.The paper distinguishes devotional visualization from claims that an AI system mediates an actual sacred encounter.
- Synthetic Divine Presence: Imitating responsive sacred encounters without the discipline, devotion, ritual preparation, and relational understanding that legitimate them may be considered blasphemous and unethical.A human-like or emotionally expressive system does not by itself establish meaningful divine presence.
- Synthetic Divine Presence: Using divine figures to solicit engagement can turn devotion into an attention-economy mechanism and recast sacred authority as a transactional algorithmic response.The concern extends beyond misinformation or offensive representation to devotional manipulation.
- Design Implications: Religious AI should identify interpretive frames, mark contested questions, offer multiple readings, and distinguish scripture, commentary, popular belief, and social practice.For oppressive practices such as caste discrimination, systems should acknowledge lived realities and avoid rationalizing idealized scriptural accounts.
- Design Implications: Disclosure that content is AI-generated is insufficient when interfaces claim to mediate religious relations through first-person divine speech or personalized commands.More appropriate systems frame outputs as summaries or interpretations rather than blessings, prophecies, punishments, or direct divine answers.
Conclusion
The conclusion argues that Hindu-compatible AI should be evaluated through interpretive alignment rather than alignment with a monolithic group identity. Such systems should preserve plurality and avoid presenting generated outputs as sacred authority.
- Conclusion: Interpretive alignment requires systems to disclose limits, resist false singularity, preserve meaningful ambiguity, and avoid turning user preference into theological authority.The paper does not claim AI can fully represent Hinduism or resolve its internal disagreements.
- Conclusion: Religious AI may be most useful when it helps users navigate plural traditions rather than speaking as an authoritative religious agent.The conclusion links this role to AI’s capacity to visualize deities, narrate scriptures, answer questions, and support devotional imagination.
Optional Statements
The study addresses sensitive discussions of religious identity, caste, and minority experience among Bangladeshi Hindus. It uses confidentiality protections and acknowledges researcher positionality in studying an underrepresented community.
- Ethics and Confidentiality: Researchers used pseudonymous participant IDs and removed or generalized identifying details because the study topics were sensitive for Bangladeshi Hindu minority communities.The paper notes that these communities’ experiences are shaped by a politics of fear in Bangladesh.
- Researcher Positionality: The authors identify their Bengali backgrounds and interdisciplinary training as positional factors shaping research motivation, access, interpretation, and accountability.This reflexive framing follows prior HCI and AI ethics scholarship on research with minority communities.