Does Sequencing Matter? Evaluating AI and Human Simulations for High-Stakes Communication Training in Law Enforcement

Paper Title

Does Sequencing Matter? Evaluating AI and Human Simulations for High-Stakes Communication Training in Law Enforcement

Publication Info

• Topic area: High-stakes communication training using AI and human simulations.
• Keywords: AI training, human role-play, law enforcement, trauma-informed communication, hybrid training, simulation sequencing, psychological distance, experiential learning, self-efficacy, generative AI.

Background and Problem

• Problem / challenge: Traditional role-play simulations for high-stakes communication training are resource-intensive, emotionally taxing, and not scalable. AI-driven training tools offer scalability but lack emotional nuance, nonverbal expressiveness, and human realism. The optimal sequencing of AI and human simulations remains unexplored.
• Significance: Effective training in trauma-informed communication is critical in law enforcement to minimize victim trauma and improve investigative outcomes. Understanding how to combine AI and human simulations can enhance training scalability and effectiveness.
• Motivation and related work: Prior research has shown the benefits of role-play and AI-based training in isolation but has not explored their integration. Existing studies focus on "AI versus human" comparisons rather than their complementary use. This paper investigates how sequencing these modalities affects learning outcomes and trainee perceptions.

Solution

• Proposed approach: A hybrid training system combining AI-based and human role-play simulations, with an emphasis on sequencing to optimize learning outcomes.
• Novelty:
  1. Introduction of a conceptual design framework based on social-emotional, temporal, and embodied distance to guide hybrid training design.
  2. Empirical investigation of how sequencing AI and human simulations affects learning trajectories and trainee perceptions.
  3. Demonstration of how psychological distance can be modulated across training stages to scaffold preparation, performance, and reflection.
• Procedure and key techniques:
  • Developed an AI-powered training system featuring three victim personas with distinct trauma profiles.
  • Conducted a mixed-methods study with 35 police recruits using a counterbalanced crossover design (AI-first vs. human-first).
  • Collected quantitative data on self-efficacy and technology acceptance, and qualitative data through interviews to analyze learning experiences and sequence effects.

Results

• Concrete findings:
  • Both AI and human simulations significantly improved self-efficacy (p = 0.01 for AI; p < 0.05 for human role-play).
  • Participants’ perceptions of the AI system improved significantly after use (p = 0.00 for ease of use, usefulness, and intention to use).
  • AI-first sequencing was preferred by over 80% of participants for its preparatory value.
• Advantage over baselines:
  • AI simulations provided a safer emotional environment for experimentation, reflective space for thoughtful communication, and focused practice on core skills.
  • Human role-play offered emotional immediacy, real-time pressure, and multimodal interaction, complementing AI’s strengths.
• Experiments / evaluation:
  • Participants completed both AI and human simulations in counterbalanced order.
  • Quantitative measures included self-efficacy and technology acceptance; qualitative insights were derived from 15 semi-structured interviews.
  • The study highlighted how sequencing shaped emotional engagement, skill transfer, and learning trajectories.
• Limitations and future work:
  • Small, gender-imbalanced sample (31 males, 4 females) from a single cohort limits generalizability.
  • Lack of control conditions (AI-only or human-only) and reliance on self-reported measures.
  • Future research should validate the distance framework, assess long-term skill retention, and expand to other domains and populations.

Summary

This study demonstrates that AI-based simulations can complement human role-play in high-stakes communication training by offering scalable, low-stakes environments for skill development. Sequencing AI before human simulations was found to scaffold preparation and reduce anxiety, while human role-play provided emotional realism and multimodal interaction. The proposed conceptual design framework, based on social-emotional, temporal, and embodied distance, offers a structured approach to hybrid training design. These findings have implications for law enforcement and other domains requiring sensitive communication skills, emphasizing the value of intentionally orchestrating AI and human modalities.