Temporal Preferences in Language Models for Long-Horizon Assistance

2025-09-05

AI Summary

This study investigates whether large language models (LLMs) exhibit future- or present-oriented temporal preferences in intertemporal choice tasks, and whether such preferences are controllable. Method: Leveraging a standardized time-tradeoff paradigm adapted from human behavioral experiments, we systematically manipulate temporal orientation via prompt engineering and introduce the Manipulability of Temporal Orientation (MTO) metric to quantify controllability. Results: LLMs demonstrate significant plasticity in temporal preference: reasoning-oriented models consistently exhibit delayed gratification under strong future-oriented prompts and adapt decisions to identity- and geography-specific contexts. Notably, certain models internalize future orientation as a stable decision strategy rather than merely reacting to surface-level cues. This work establishes the first reproducible evaluation framework for assessing temporal preferences in LLMs, offering novel methodological pathways for AI value alignment and modeling long-horizon planning capabilities.

Technology Category

Humans and AI: Learning Human Values and PreferencesMachine Learning: Large Multimodal Models (LMMs)Knowledge Representation and Reasoning: Preferences

Application Category

User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendationEconomics, Online Markets and Human Computation: LLM based quality controls for crowd workSemantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

We study whether language models (LMs) exhibit future- versus present-oriented preferences in intertemporal choice and whether those preferences can be systematically manipulated. Using adapted human experimental protocols, we evaluate multiple LMs on time-tradeoff tasks and benchmark them against a sample of human decision makers. We introduce an operational metric, the Manipulability of Time Orientation (MTO), defined as the change in an LM's revealed time preference between future- and present-oriented prompts. In our tests, reasoning-focused models (e.g., DeepSeek-Reasoner and grok-3-mini) choose later options under future-oriented prompts but only partially personalize decisions across identities or geographies. Moreover, models that correctly reason about time orientation internalize a future orientation for themselves as AI decision makers. We discuss design implications for AI assistants that should align with heterogeneous, long-horizon goals and outline a research agenda on personalized contextual calibration and socially aware deployment.

Problem

• Study LM time orientation preferences manipulation
• Evaluate models on intertemporal choice tasks
• Measure Manipulability of Time Orientation metric

Innovation

• Adapted human experimental protocols evaluation
• Introduced Manipulability of Time Orientation metric
• Reasoning-focused models for time preference manipulation