How to Make AGI Execute Reasoning Tasks in a Time-budget-sensitive Manner?

Studies in Logic, Vol. 18, No. 6 (2025): 62–87                    PII: 1674-3202(2025)-06-0062-26

Yingjin Xu, Bojie Feng

Abstract. An idealized artificial general intelligence (AGI) system is expected to be smart enough to balance competing needs for caution and urgency via the management of limited time resources. However, mainstream Large Language Models (LLMs), due to its working principle requiring operating resources of a tremendous magnitude, cannot be sensitive to the limitation of time resources, hence, it cannot adjust its time-distributions vis-à-vis the combinations of different internal and external conditions. Hence, a new AGI approach to the time-management is needed. By integrating the psychological notion of Need-For-Closure (NFC) and Nagel’s (2008) intellectual invariantism as well as Stanley’s (2005) stake-based epistemology, we will present a unified and computable framework for time-management, namely, a framework intended to handle both the system’s own estimation of the complexity of the task and its perception of the degree of stakes triggered by the task. In addition, in ourmodel, the computable notion of “budget deficit” plays a pivotal role in explaining how the NFC-value is fine-tuned to guide epistemic shifts.