Rethinking Intelligence Power and Epistemic Authority in the Age of Superhuman AI

As artificial intelligence (AI) systems increasingly claim “human-level” or “superhuman” performance, foundational assumptions about intelligence remain under-theorized. Existing AI safety and alignment discourses often frame intelligence as disembodied, universal, and measurable-reinforcing Western-centric benchmarks such as logical reasoning, linguistic proficiency, and computational speed. This narrow framing risks legitimizing a technocratic model of alignment that excludes diverse cultural and epistemic perspectives. This paper investigates how dominant imaginaries of intelligence are constructed and contested within contemporary AI governance. Drawing on a mixed-methods design, the study combines critical discourse analysis of foundational texts from leading AI labs (OpenAI, DeepMind, Anthropic) with twenty semi-structured interviews involving AI researchers, ethicists, and interdisciplinary scholars. The analysis reveals that the prevailing conception of “superintelligence” privileges optimization, prediction, and control-while marginalizing moral reasoning, relational understanding, and embodied or situated forms of knowledge. In response, the paper proposes a pluralistic reframing of machine intelligence grounded in cognitive diversity, cultural epistemology, and participatory governance. It argues for expanding benchmarks, safety protocols, and alignment frameworks to reflect diverse values and cognitive styles. Concrete pathways for operationalizing this shift include care-aligned safety audits, epistemically diverse evaluation metrics, and polycentric governance structures. By decentering dominant techno-scientific paradigms, the paper contributes to a more inclusive and socially grounded vision of AI governance. This reframing holds significance for increasing public trust, mitigating epistemic injustice, and developing alignment strategies that are not only safe, but also equitable and contextually relevant on a global scale.