MindShift: Leveraging Large Language Models for Mental-States-Based Problematic Smartphone Use Intervention

TL;DR

MindShift addresses problematic smartphone use by enabling context-aware, mental-state-informed intervention content generated by LLMs. The method integrates in-the-moment usage, physical context, goals, and habits with four persuasion strategies; evaluated in a 5-week within-subject field study (N=25) against Baseline and MindShift-Simple. Results show higher intervention acceptance, reduced usage duration, SAS decreases, and boosted self-efficacy, demonstrating the potential of LLM-powered adaptive persuasion for behavior change. The work offers design implications for LLM-based interventions in other domains and discusses ethical considerations and limitations.

Abstract

Problematic smartphone use negatively affects physical and mental health. Despite the wide range of prior research, existing persuasive techniques are not flexible enough to provide dynamic persuasion content based on users' physical contexts and mental states. We first conducted a Wizard-of-Oz study (N=12) and an interview study (N=10) to summarize the mental states behind problematic smartphone use: boredom, stress, and inertia. This informs our design of four persuasion strategies: understanding, comforting, evoking, and scaffolding habits. We leveraged large language models (LLMs) to enable the automatic and dynamic generation of effective persuasion content. We developed MindShift, a novel LLM-powered problematic smartphone use intervention technique. MindShift takes users' in-the-moment app usage behaviors, physical contexts, mental states, goals \& habits as input, and generates personalized and dynamic persuasive content with appropriate persuasion strategies. We conducted a 5-week field experiment (N=25) to compare MindShift with its simplified version (remove mental states) and baseline techniques (fixed reminder). The results show that MindShift improves intervention acceptance rates by 4.7-22.5% and reduces smartphone usage duration by 7.4-9.8%. Moreover, users have a significant drop in smartphone addiction scale scores and a rise in self-efficacy scale scores. Our study sheds light on the potential of leveraging LLMs for context-aware persuasion in other behavior change domains.

Figures (12)

• Figure 1: Overview of MindShift. When users exhibit problematic phone usage, MindShift actively collects data on phone usage behavior, physical contexts, and mental states, and uses the customized persuasion strategies we designed along with users' goals and habits, to generate prompts. Then, the large language model will generate a persuasion message. Finally, the persuasion will show up in users' phones to encourage more mindful usage.
• Figure 2: Mapping of Persuasion Strategies to the ERG Theory (Existence, Relatedness, and Growth) and the Dual Systems Theory. According to the Dual Systems Theory, there is a competitive relationship between System 1 (habitual smartphone use) and System 2 (meaningful activity), much like being placed on a scale. To make System 2 heavier than System 1, weights are added—Relatedness needs and Growth needs (the second and third levels of ERG Theory). To support users' Relatedness needs, we design two persuasion strategies: Understanding and Comforting. To motivate Growth needs, we design two other persuasion strategies: Evoking and Scaffolding Habits.
• Figure 3: Summary of Persuasion Strategies under Different Mental States. In the table, each color denotes the corresponding persuasion strategy applicable to this scenario.
• Figure 4: Prompt Templates Used to Generate Persuasive Content with GPT-3.5. The word slots (in the top box) represent different categories of information to be filled in based on the user's current situation. The color indicates the mapping between the slots and the input prompt. The small words under each slot explain the source of the content. The input prompt consists of four parts: (1) Task Setup; (2) Description of the Current Contexts; (3) Prompt Optimization (to improve language quality and reduce harmful content); and (4) Description of Persuasion Strategy (as introduced in Sec. \ref{['sec:formative_studies']} and Figure \ref{['fig:strategy']}). The LLM output shows the persuasion example of four strategies when the user's mental state is "stressed, engaging in an activity".
• Figure 5: Interaction Flow. Users first complete the global settings for their value and app list (Step 0). When opening a black-listed app, users need to first self-report their phone usage intent (Step 1). If the intent is habitual use, the app asks them to report their current mental state (Step 2). After that, a corresponding persuasion shows (Step 3).