Chatbots and the Initial Account Interview for Gathering Eyewitness Information: Development, Research and Efficacy of ‘ChatCharlie’

This thesis presents a programme of empirical research centred on promptly eliciting information from witnesses and victims of crime to consolidate memory for later full retrieval. The College of Policing sometimes advocates Initial Account Interviews, yet more research should focus on the prevailing Initial Account interview approach. This novel research considers the nature of episodic memory and highlights a dearth of literature in the domain of initial accounts. In doing so, the potential of a chatbot approach for remotely gathering initial accounts is considered when logistical and personnel challenges preclude eliciting an initial account in person.

This body of work underscores the significance of in-person Initial Account Interviews and suggests the potential of chatbots as a novel interviewing modality referred to as ChatCharlie. The thesis reports the development and first evaluation of ChatCharlie, aligning its design with best interviewing practices and psychological insights into human-computer interaction. Specifically, it uses a mock witness paradigm to explore the impact of typed interviews and technology usage across age groups. The acceptability of chatbots in an eyewitness context was examined, potentially contributing to the ‘toolbox’ of investigative techniques for frontline investigators.

Study 1 presents the first in a series of empirical investigations of eyewitness memory performance as a function of reporting modality and considers typed modality interviews as one method for potentially expediting the collection of witness information. This study provides insight into the potential costs and benefits of typed versus spoken responses to investigative questioning. Results revealed no difference between conditions for memory performance (correct, incorrect and percentage accuracy), apart from confabulations. Non-significant differences were consistent globally and as a function of the recall phase

Study 2 focused on the typed recall of eyewitness information as a function of age. Again, this study uses the mock-witness paradigm to examine episodic recall performance across three different age groups when providing typed information. Results revealed that participants in the 55+ group recalled more correct items than the two younger age groups, with no significant differences across age groups for confabulated or erroneous recall. Despite lower confidence in technology in the 55+ group, here familiarity with technology did not appear to impede the retrieval and reporting process. These findings further suggest that text-based retrieval may have the potential for promptly and accurately capturing witness accounts in some instances and that older witnesses may also benefit.

Study 3 examined the efficacy of ChatCharlie for consolidating eyewitness memory versus an in-person Initial Account Interview. At Time 1, participants were randomly assigned to ChatCharlie, In-person initial account or the no initial account control condition. There were significantly fewer confabulations in the ChatCharlie condition At Time 1, with non-significant differences in correct and incorrect memory performance across the ChatCharlie and In-person conditions. At Time 2, participants in the ChatCharlie and In-person Time 1 conditions were more accurate. They recalled significantly more correct items globally and during both interview phases than the Control group without a concomitant increase in errors and confabulations. No significant differences emerged at Time 2 between ChatCharlie and the In-Person Time 1 conditions.