ESRA 2027 sessions by theme
Back to Overview of Sessions
Human expertise in human-AI collaborative research workflows |
| Coordinator 1 | Dr Yongwei Yang (Google) |
| Coordinator 2 | Dr Rich Timpone (Protopian Works) |
| Coordinator 3 | Dr Joe Paxton (Google) |
Session Details
The survey research field has tested and applied (LLMs) across all aspects of the workflow (von der Heyde et al., 2026). AAPOR’s Responsible AI Integration in Survey Research report examines the use and best practices for integrating AI, particularly LLMs, into survey research (Rothschild et al., 2026), arguing that AI should augment human capabilities rather than enable end-to-end automation.
Building on models of human-machine collaboration and task automation (Dougherty & Wilson, 2018; Autor et al., 2003; Autor, 2015), Timpone and Yang (2025) discussed how different forms of human-AI collaboration can be intentionally selected based on the volatility, uncertainty, complexity, and ambiguity (VUCA; Barber, 1992) of specific decision-making tasks.
This session will advance theories and practices for designing human-AI collaboration (HAIC) in survey research, with particular emphasis on leveraging human expertise. Relevant topics include:
1. Critical human expertise when LLMs augment research, including high-order functions (e.g., planning, goal-/value-setting) and domain-specific knowledge and skills.
2. Human-AI collaboration skills that provide value that LLMs cannot deliver alone. These may involve delegation (identifying suitable tasks), interaction (eliciting effective output), evaluation (assessing accuracy, appropriateness, and usefulness), customization (enhancing output), and integration (incorporating AI-generated content into research).
3. Quality assurance and control (QA/QC) practices in AI-in-the-loop workflows, focusing on human and AI roles under different VUCA conditions and mechanisms for triggering human intervention to ensure effectiveness, consistency, and scalability.
4. Analyses of the impact of human expertise in HAIC workflows.
5. Creating and structuring skills for LLM agents, evaluating their effectiveness, and ensuring that human contributions are recognized and rewarded.
6. HAIC curricula to train and upskill survey professionals in task-based choices about whether AI should supplement, replace, or override researchers, while avoiding cognitive surrender (Shaw & Nave, 2026).
