About Our Research Group
The Child Witness Interview Enhancement Research Group is a collaborative initiative between Åbo Akademi University in Finland and New York University Shanghai. Our interdisciplinary team brings together expertise in developmental psychology, forensic interviewing, and artificial intelligence to explore how large language models (LLMs) can enhance the quality of investigative interviews with child witnesses. Established in 2023, our group represents a pioneering effort to bridge the gap between advanced AI technologies and forensic child interviewing.
We are dedicated to developing evidence-based technological solutions that support human interviewers in conducting high-quality, ethical interviews with child witnesses. Our research aims to ensure that children’s narratives are heard accurately and sensitively in investigative contexts, while adhering to established best practices in forensic interviewing.
Why This Matters
Child testimony is often crucial in cases of abuse or other crimes, yet obtaining accurate accounts presents unique challenges. Children may be susceptible to suggestion, have difficulty communicating traumatic experiences, or struggle with the formal interview process. Meanwhile, interviewers face the complex task of gathering essential information while maintaining child welfare.
Forensic interviewing places cognitive demands on the human interviewer. They must simultaneously:
- Build and maintain rapport
- Formulate appropriate, non-leading questions based on the child’s previous responses
- Remember details from earlier parts of the interview to identify inconsistencies or areas needing clarification
- Monitor their own verbal and non-verbal behavior to avoid influencing the child
- Navigate complex emotional situations with composure and sensitivity
- Adhere to evidence-based protocols while responding to unpredictable interview directions
Even the most skilled interviewers can experience fatigue, cognitive overload, or momentary distractions that affect interview quality. Furthermore, many interviewers work without the benefit of real-time support from colleagues or supervisors, particularly in resource-constrained settings where a team approach is not feasible.
The training challenge compounds these difficulties. Teaching effective interviewing skills requires substantial resources, including intensive instruction, ongoing practice opportunities, appropriate feedback, and regular refresher training. Research consistently shows that traditional training approaches often produce improvements that diminish over time without continuous reinforcement. Many professionals receive initial training in evidence-based protocols but struggle to maintain adherence to best practices when faced with real-world pressures and limitations. The resource-intensive nature of quality training means that many jurisdictions cannot provide optimal preparation for all professionals who interview children.
Our research addresses these practical challenges by exploring how AI tools can serve as “cognitive assistants” that support human interviewers through technological augmentation. By conducting research aiming to develop technology that could help manage cognitive load and provide real-time interview guidance, we aim to create meaningful improvements in investigative interviews of children.
Research Impact
Our work aims to address several challenges in child witness interviewing:
- Reducing Interviewer Variability: Even trained professionals can vary in their adherence to best practices. Our LLM-based tools will help standardize interview quality.
- Minimizing Suggestive Questioning: Leading questions can compromise children’s testimony. Our systems help detect and prevent suggestive questioning in real-time.
- Supporting Under-resourced Settings: Many jurisdictions lack access to specialized training. Our technology could democratize access to high-quality interviewing techniques on a global scale.
Current Research Projects
Completed Studies
- LLM as Child Interviewer: Pilot study examining LLM performance as an interviewer for children who have viewed a mock event
- Predictive Capabilities: Analysis of LLM competence in predicting study results in child interviewing contexts
- Hypothesis Generation: Investigation of LLM capabilities in generating hypotheses based on child abuse interview scenarios
- Simulated Interview Performance: Assessment of LLM capabilities in both static and dynamic simulated child abuse interview scenarios
- Rapport Building: Data collection on LLM effectiveness as an interviewer and rapport builder in interviews with children who have viewed mock events
Ongoing Research
- Development of Rich LLM Avatars: Creation of advanced LLM avatars for testing AI interviewing capabilities in controlled settings
- Protocol-Based Training: Fine-tuning LLMs using structured protocols like the NICHD guidelines for child investigative interviews
- Transcript-Based Training: Training LLMs using interview transcripts to conduct more sophisticated interviews
Future Directions
Our upcoming research will explore:
- Comparative studies between protocol-trained and transcript-trained LLMs
- Integration of human interviewers with LLM assistants
- Development of real-time API-based interview support systems
- Potential field studies in collaboration with professional interviewers
Ethical Considerations
Our work is guided by a commitment to ethical research practices. We prioritize:
- Child welfare and psychological safety
- Adherence to evidence-based interviewing protocols
- Protection of sensitive data and privacy
- Transparency about AI capabilities and limitations
- Human oversight and responsibility in investigative contexts
- Cultural sensitivity and cross-cultural validity of interviewing techniques
- Rigorous evaluation of AI systems before any real-world implementation
Partner Institutions
Åbo Akademi University
New York University Shanghai
Publications and Presentations
Team Members
Pekka Santtila Professor of Psychology and Global Network Professor at New York University (Shanghai)
Julia Korkman Adjunct professor in Legal Psychology
Liisa Järvilehto is a PhD student in Psychology at Åbo Akademi University, specializing in child investigative interviewing. With over a decade of professional experience in forensic psychology and investigative interviewing of children, she brings valuable practical expertise to the research team. Her first-hand knowledge of interviewer challenges, cognitive demands, and protocol implementation directly informs the development of AI tools that can meaningfully support forensic interviewers in real-world settings.
Sun Yongjie is currently a second-year master’s student at East China Normal University. He holds dual bachelor’s degrees in Psychology and Law. His research focuses on the use of experimental methods involving simulated investigative interviews to explore the potential of AI-assisted tools in forensic settings. He is also interested in employing serious games to measure and train individuals’ judgments of confession credibility, and in examining how large language models might support jury deliberation processes to mitigate cognitive biases.
Hasse Hällström has PhD from the University of Edinburgh (2015) and works primarily as an AI scientist at Mehiläinen, the leading healthcare services provider in Europe, focuing on leveraging large language models in healthcare use cases. Before his current role, he has been working as a philosophy researcher at Jagiellonian University in Kraków, and in several companies developing AI models for tasks ranging from improving primary and secondary education outcomes to enhancing musical creativity. In this project, he is involved in a research focused on developing AI-driven tools to enhance the quality of child investigative interviews. He has created an AI-based system for classifying question types in child investigative interviews, and currently, he is engaged in building AI avatars designed to simulate child interviewees and thus provide realistic training scenarios for investigative interviewers. The goal is to create a scalable, interactive platform that can enhance interviewer preparedness and skill.
Funding and Support
Our research is made possible through the generous support of the Sundell foundation.
Join Our Research: We welcome inquiries from potential graduate students, postdoctoral researchers, and collaborators interested in our work.