PRACTICAL THEOLOGY IN THE AGE OF POST-SCALING AI
Enrico Beltramini
PRACTICAL THEOLOGY IN THE AGE OF POST-SCALING AI
Enrico Beltramini
In the age of post-scaling AI, theology is called to shift from abstract speculation to practical engagement with how AI shapes everyday ecclesial life and ministry.
Not long ago, philosopher Nick Bostrom warned of the existential risks posed by a superintelligent AI (Superintelligence, 2014), while the theoretical physicist Max Tegmark envisioned the possibility of interstellar travel enabled by AI-driven cyborgs (Life 3.0, 2017). Today, mainstream media continues to fuel narratives of either existential risk or techno-messianism. Yet, quietly and steadily, signs are accumulating that the so-called Singularity remains a distant prospect. These signs come from various domains: some are technical in nature, others economic and financial, and still others pertain to issues of sustainability.
Until 2022, it was widely believed that the primary technical limitation to training large AI models was computing power—specifically, the ability to process massive datasets through increasingly complex neural networks. The dominant assumption was: the more computational resources (i.e., GPUs/TPUs), the more parameters one could train, leading to more powerful models. This belief drove massive investments in high-performance computing infrastructure and cloud-based AI systems.
However, after 2022, the landscape began to shift. Researchers and engineers began recognizing that computational power is only one bottleneck among several others. New constraints emerged or became more evident. The publication of the so-called Chinchilla Paper (2022) highlighted the fact that high-quality, diverse, and ethically sourced training data is limited. Even with more computing power, models cannot improve if they are trained on redundant or low-quality data. A second limitation lies in the inefficiency of scale: as models grow larger, the improvements in performance become increasingly marginal. This reveals an inherent imbalance—substantially greater computational resources yield only modest gains, calling into question the long-term viability of scaling as a strategy for advancing AI capabilities.
A third limitation is related to the fact that training large AI models demands vast amounts of energy, often equivalent to powering hundreds of homes for weeks. As model sizes grow, energy requirements rise exponentially, straining power grids and increasing carbon emissions. These escalating energy demands pose significant environmental and economic challenges, raising concerns about AI’s long-term sustainability. A fourth limitation involves the anticipated economic returns. Although the dominant narrative suggests that AI will replace human labor to boost efficiency and profits, Turkish-American Nobel laureate economist Daron Acemoglu projects only a modest increase in GDP—between 1.1 and 1.6 percent over the next decade—with an annual productivity gain of approximately 0.05 percent (“The Simple Macroeconomics of AI,” 2024).
In short, while computing power was once seen as the main limiter, by 2022 it became clear that AI scalability faces a constellation of limits—technical, economic, and energetic. The hypothesis of the Singularity is grounded in the assumption of unlimited AI scalability; if scalability encounters limits, the Singularity ceases to be a plausible scenario.
While these limitations may not entirely rule out the prospect of a future Singularity, they substantially delay its anticipated arrival. In the meantime, AI is increasingly being deployed as a tool for localized, context-specific applications—enhancing medical diagnostics in rural clinics, optimizing resource distribution in underserved communities, supporting education through adaptive learning platforms, and assisting with language translation in multicultural settings. Rather than ushering in a universal technological revolution, AI is gradually becoming a means of addressing concrete human needs, shaped by the particularities of place, culture, and available infrastructure.
The reality of post-scaling AI calls for ethical and theological reflection rooted not in speculation about distant futures, but in the concrete conditions of contemporary human life. Public theologians are uniquely positioned to engage with the ethical, pastoral, and communal implications of AI in concrete settings. By focusing on how AI intersects with daily life—such as caregiving, education, justice, and spiritual formation—public theologians can offer critical discernment, shape human-centered uses of technology, and contribute to the development of frameworks that prioritize dignity, solidarity, and the common good. Rather than reacting to AI with fear or utopian optimism, they can guide communities in interpreting its presence through the lens of faith, service, and the lived experience of the Gospel.
A telling example can be found in pastoral settings, where priests and preachers are increasingly relying on AI-powered applications to aid in the preparation of their homilies. These tools provide scriptural commentary, historical background, theological interpretation, and even suggested outlines shaped to fit the liturgical calendar or specific congregational settings. The use of such applications raises significant theological questions. It touches on the issue of authority in scriptural interpretation: does the interpretive weight lie with tradition and ecclesial teaching, or can it be guided by algorithmic correlations drawn from large datasets? It also challenges preachers to exercise discernment in distinguishing between meaningful insights and reductions of theological depth into surface-level associations. Moreover, it raises concerns about the formation of clergy, particularly when reliance on technology might displace the disciplines of prayer, study, and communal reflection (Beltramini, “The Democratization of Theology and the Death of the Theologian,” forthcoming).
A constructive theological response would not reject these tools outright, but would seek to integrate them critically—ensuring they serve, rather than supplant, the preacher’s engagement with the Word. This includes developing clear criteria for their use, forming users to approach them thoughtfully, and reaffirming that preaching remains a sacramental and ecclesial act that cannot be fully delegated to a machine. AI, in this setting, becomes a catalyst for reflection on the nature and vocation of preaching in a digital age.
Enrico Beltramini specializes in political theology and historical theology, drawing on the intellectual and spiritual resources of Roman Catholicism. He is the author of four monographs and over seventy peer-reviewed articles published in academic journals. He currently serves as a retired professor of theology with an affiliation to Notre Dame de Namur University in California, where he taught for 15 years. He holds doctoral degrees in theology (Vidyajyoti College of Theology), history (University of London), and social theory (University of Manchester).