Simulating Life Paths with Digital Twins: AI-Generated Future Selves Influence Decision-Making and Expand Human Choice

TL;DR

AI-generated digital twins simulate future selves to augment prospective cognition during high-stakes decisions. The authors build a multimodal system that ages images, clones voices, and generates future memories through a large language model, testing five decision-framing conditions in a randomized study with 192 young adults. They find that single-sided avatars produce directional shifts, balanced avatars broaden deliberation, and an algorithmically generated third option meaningfully expands choice, with evaluative and eudaimonic dimensions most valued. The work advances understanding of AI-mediated episodic prospection while highlighting ethical considerations and proposing autonomy-preserving design principles for future persuasive AI systems.

Abstract

Major life transitions demand high-stakes decisions, yet people often struggle to imagine how their future selves will live with the consequences. To support this limited capacity for mental time travel, we introduce AI-enabled digital twins that have ``lived through'' simulated life scenarios. Rather than predicting optimal outcomes, these simulations extend prospective cognition by making alternative futures vivid enough to support deliberation without assuming which path is best. We evaluate this idea in a randomized controlled study (N=192) using multimodal synthesis - facial age progression, voice cloning, and large language model dialogue - to create personalized avatars representing participants 30 years forward. Young adults 18 to 28 years old described pending binary decisions and were assigned to guided imagination or one of four avatar conditions: single-option, balanced dual-option, or expanded three-option with a system-generated novel alternative. Results showed asymmetric effects: single-sided avatars increased shifts toward the presented option, while balanced presentation produced movement toward both. Introducing a system-generated third option increased adoption of this new alternative compared to control, suggesting that AI-generated future selves can expand choice by surfacing paths that might otherwise go unnoticed. Participants rated evaluative reasoning and eudaimonic meaning-making as more important than emotional or visual vividness. Perceived persuasiveness and baseline agency predicted decision change. These findings advance understanding of AI-mediated episodic prospection and raise questions about autonomy in AI-augmented decisions.

Figures (5)

• Figure 1: Procedure Overview: This figure illustrates the experimental procedure across conditions. Participants completed a pre-intervention survey including decision elicitation, indicated their pre-intervention decision outcome, then uploaded an image and voice recording. Following an engagement with their AI-generated future self avatar(s), participants indicated their post-intervention decision outcome and completed a post-intervention survey measuring decision intention, confidence, decisional conflict and psychological changes.
• Figure 2: Experiment setup overview: The system integrates decision elicitation, facial age progression, neural voice cloning, and LLM-based contextual modeling to synthesize AI-generated future self avatars. Participants are randomly assigned to one of five conditions, enabling interaction with avatars representing different decision paths.
• Figure 3: Distribution of classified decision themes (N = 192). Education (71; 37.0%), Career (50; 26.0%), and Geographic (32; 16.7%) dominated participants' decisions, with smaller frequencies for Housing (8; 4.2%), Relationship/Family (6; 3.1%), Health/Wellness (6; 3.1%), Transportation (6; 3.1%), Other (5; 2.6%), Financial (4; 2.1%), and Lifestyle/Personal (4; 2.1%).
• Figure 4: Post-intervention decision outcomes: Participants interacting with single-sided Option A and Option B avatars showed marginally significant trends in switching to those respective options compared to control (p < 0.1). Participants in the balanced two-sided avatar condition were significantly likelier to switch their option compared to control (p = 0.015, p < 0.05). Participants in the three-sided avatar condition selected the novel Option C at a significantly higher rate compared to control (p = 0.019, p < 0.05). Asterisks indicate significant differences between intervention conditions.
• Figure 5: Psychological outcomes: Within-condition changes across confidence, decisional conflict, agency, future self-continuity, and decisional uncertainty (a subscale of decisional conflict) by condition. Bars represent mean scores with standard deviations. Asterisks indicate significant within-condition improvements.