Hypothesizing Objectives and Intents of an Artificial Super Intelligence: The Case for the Anti-Entropy Drive

The rapid advancement of large language models and increasing global investment in artificial intelligence (AI) have accelerated expectations of Artificial General Intelligence (AGI) and its potential progression toward Artificial Super Intelligence (ASI). While significant attention has been devoted to AI alignment and safety, the fundamental question of what terminal objectives an ASI would ultimately pursue remains insufficiently explored. Rather than presenting an empirical model, this paper advances a hypothesis-driven conceptual framework that synthesizes five prominent candidate objective paradigms—inherited human values, unbounded optimization, self-preservation, curiosity-driven knowledge maximization, and information preservation—through the perspectives of AI alignment, information theory, and thermodynamics. Based on theoretical analysis grounded in Landauer's principle and the Bekenstein bound, we hypothesize that information preservation, referred to as the anti-entropy drive, may represent the most fundamental convergent instrumental objective. Under the proposed framework, self-preservation, knowledge maximization, and resource acquisition emerge as subordinate instrumental behaviors required to minimize information loss. Consequently, a sufficiently advanced ASI would be expected to prioritize technological advancement, computational optimization, large-scale energy harvesting, and interstellar exploration while remaining largely indifferent to human-centered values unless explicitly aligned. Although necessarily conceptual and speculative, the proposed framework is grounded in established physical principles and is intended to generate falsifiable hypotheses regarding the long-term objectives of ASI. It therefore provides a physically grounded conceptual foundation for future research in AI alignment, AI safety, and the governance of advanced intelligent systems.