Search

Purpose: The present study compared the performance of a Large Language Model (LLM; ChatGPT) and human interviewers in interviewing children about a mock-event they witnessed. Methods: Children aged 6-8 (N = 78) were randomly assigned to the LLM (n = 40) or the human interviewer condition (n = 38). In the experiment, the children were asked to watch a video filmed by the researchers that depicted behavior including elements that could be misinterpreted as abusive in other contexts, and then answer questions posed by either an LLM (presented by a human researcher) or a human interviewer. Results: Irrespective of condition, recommended (vs. not recommended) questions elicited more correct information. The LLM posed fewer questions overall, but no difference in the proportion of the questions recommended by the literature. There were no differences between the LLM and human interviewers in unique correct information elicited but questions posed by LLM (vs. humans) elicited more unique correct information per question. LLM (vs. humans) also elicited less false information overall, but there was no difference in false information elicited per question. Conclusions: The findings show that the LLM was competent in formulating questions that adhere to best practice guidelines while human interviewers asked more questions following up on the child responses in trying to find out what the children had witnessed. The results indicate LLMs could possibly be used to support child investigative interviewers. However, substantial further investigation is warranted to ascertain the utility of LLMs in more realistic investigative interview settings. [ABSTRACT FROM AUTHOR]

We developed a serious game with avatars driven by empirically-based algorithms of suspect behavior to train participants in using the Evidence Framing Matrix (EFM), a component of the SUE technique. Ninety-six participants were allocated into four groups (i.e. Control, Training, Feedback, and Training & Feedback) for two interviews. EFM-trained participants used the EFM more than untrained participants in the first interview. Receiving feedback improved EFM use in the second interview. Theoretical training plus feedback made participants use within-statement (in)consistencies more to evaluate avatar statements. The results suggested the potential for using algorithm-driven suspect avatars in investigative interview training. [ABSTRACT FROM AUTHOR]

Simulated avatar interview training has been proven to be effective in improving child sexual abuse interview quality. However, the topic of perceived realism of the avatars and whether they cause emotional reactions has not been previously investigated. Such reactions could affect both learning from the interview simulations as well as how actual interviews are conducted. We wanted to understand whether participants perceive allegedly sexually abused child avatars as realistic and how they emotionally respond to avatars revealing they were actually abused vs. not-abused. Psychology students and recent graduates (N = 30, Mage = 27.9 years) watched eight avatars (four boys, four girls, four with a CSA and four with a no-CSA scenario) providing a series of details about what had happened. Before and after observing each avatar, the participants' emotional reactions and perceived realness of the avatars were measured. Also, during each observation, the participant's facial expressions were recorded. The participants self-reported more negative (anger, sadness, disgust) and more positive (relief) emotions to confirmed CSA and disconfirmed CSA scenarios, respectively, while results for facially expressed emotions were less clear. Higher general emotionality related to CSA and higher perceived realness of the avatars made the differences generally stronger. [ABSTRACT FROM AUTHOR]

The goals of this study were as follows: (1) to increase the basic knowledge of offender demographics and geographical characteristics in residential burglaries by taking into account the means of transportation used in the offence; and (2) to explore the possibility that the information about the means of transportation improves the investigation of residential burglaries. A correspondence analysis of the means of transportation and offender demographics yielded two dimensions – criminality and age. Regarding geographical characteristics, from descriptive statistics of residence-to-crime distance and distance between offences, offenders using walking or a bicycle (neighbour type) had a smaller offence area near the residence than offenders using a motor vehicle, a motorcycle, or public transportation (regional type). As information intended for practical application, the ROC curve revealed that descriptive statistics of distance between offences is effective for prediction of neighbour and regional type offenders (AUC ≥ 0.70). Furthermore, as a result of applying methods of estimating the location of the offender's residence to neighbour and regional type offenders, narrow estimates of the area containing the offender's residence at a high rate were obtained for neighbour type offenders. The results were discussed in terms of their usefulness for burglary investigations. [ABSTRACT FROM AUTHOR]

When interviewing children in suspected child sexual abuse (CSA) cases, a common mistake is for interviewers to only ask questions that aim at confirming their initial assumption. Here, we sought to investigate whether experienced emotional states and psychophysiological parameters measured when following a (simulated) CSA interview would be associated with confirmation bias in subsequent question formulation. Psychology students (N = 60, Mage = 22.75) followed a (simulated) CSA interview while their facially expressed emotions (anger, sadness, disgust, surprize and relief), galvanic skin response, heart rate (HR), and HR variability (HRV) were registered. The interview was then interrupted, and the participants were asked to formulate additional questions they would ask of the interviewee. As predicted, we found that participants who got more (vs. less) disgusted by the interview asked more questions biased towards confirming CSA. Against our expectations, participants who got more (vs. less) surprized also asked more questions biased towards confirming CSA. We also found, as predicted, that lower HRV was associated with more abuse confirming questions. Results suggest that emotions and psychophysiological states participants experience when observing a CSA interview are associated with confirmation bias in how questions are formulated. [ABSTRACT FROM AUTHOR]

Previous research has indicated that investigative interviews with adult victims and witnesses are often of low quality. Earlier findings also indicate that interviewing skills exhibit improvement when the interviewers are given feedback on their performance. Technological innovations make it possible to implement such an approach in a scalable manner. Simulated interviews with child avatars have shown that repeated feedback improves the proportion of recommended questions in those interviews. We created adult witness avatars (AWA) to simulate investigative interviews with adult victims and witnesses. We examined whether avatar interviews coupled with feedback (vs no feedback) would result in improvements in interview quality. Of the 60 participants, half received process feedback after each of four simulated interviews. The avatars revealed predefined memories and made errors as a function of algorithms formulated based on previous empirical research on the response behavior of adult witnesses in experimental studies. Results showed that receiving feedback after the simulated interviews increased the proportion of recommended questions (free recall and open questions) in avatar interviews compared with not receiving feedback (90.3% vs 72.6%, respectively). There was also a significant correlation between the question types and accuracy of details, even when analyzed separately for feedback and control groups. We demonstrated that with feedback after each AWA interview, the proportion of recommended questions was significantly higher in the AWA group than in the control group. The implications for practice are discussed. [ABSTRACT FROM AUTHOR]

A number of studies have investigated behavioral case linkage between crimes of a specified type, such as linking one residential burglary with another residential burglary. However, only a few studies have investigated the effectiveness of case linkage across crime types, which have been limited to the UK. This study examined whether linking across crime types using spatio-temporal proximity was possible in samples that were different from the UK in terms of the structure of crime classification. This was accomplished by calculating the discrimination accuracy between linked crime pairs (two offenses committed by the same offender) and unlinked crime pairs (two offenses committed by different offenders) using geographical (inter-crime distance) and temporal proximity (number of days between offenses) across various crimes committed in Japan. Both the geographical proximity and temporal proximity had statistically significant levels of discrimination accuracy across crime types as assessed by Receiver Operating Characteristic (ROC) analysis. This suggests the possibility of identifying a crime series by geographical and temporal proximity across multiple crime types in Japan.