情感计算技术如何推动犯罪风险评估工具的发展？

摘 要 随着近年来人工智能深度学习技术的发展,情感计算与人格计算技术日渐成熟,在许多实际应用场景中取得了良好的表现,当前人工智能情感计算技术应用于犯罪风险评估领域,能够有效解决目前主流的风险评估工具难以解决的个体内差异性的预测因子评估问题以及被测评参与者因社会赞许性而导致结果失真的问题。本文在详细阐述目前主流评估工具的局限性基础上,详细阐述了以情感计算技术为支撑的动态风险评估工具的设计思路、目前已有的技术方案以及设计细节的理论依据,在此基础上最后提出以人工智能技术为支撑的新型评估工具的未来发展方向。     

Social status and friendship in peer victimization trajectories

Background/Objective: Most studies have evaluated victimization at a single time point, making it difficult to determine the impact of the time during which an individual is victimized. This longitudinal study aims to examine the differences in the levels of social status (social preference and perceived popularity) and friendship in peer victimization trajectories, and to analyse if there were changes over time in the levels of social status and friendship in each trajectory. Method: The final sample was composed of 1,239 students (49% girls) with ages between 9 and 18 (M = 12.23, SD = 1.73), from 22 schools in southern Spain. Peer nominations were collected. Results: The General Linear Model results associated the highest levels of social preference, perceived popularity and friendship with the sporadic victimization profile and the lowest levels of these dimensions with the stable profile. Conclusions:The results are discussed based on important personal aspects of stable victimization that confirms social rejection, unpopularity, and the low social support that victimization causes. This contribution is discussed in terms of health and social welfare in adolescence.     

On the evaluation of one-sided evidence

We examine predictions and judgments of confidence based on one-sided evidence. Some subjects saw arguments for only one side of a legal dispute while other subjects (called ‘jurors’) saw arguments for both sides. Subjects predicted the number of jurors who favored the plaintiff in each case. Subjects who saw only one side made predictions that were biased in favor of that side. Furthermore, they were more confident but generally less accurate than subjects who saw both sides. The results indicate that people do not compensate sufficiently for missing information even when it is painfully obvious that the information available to them is incomplete. A simple manipulation that required subjects to evaluate the relative strength of the opponent's side greatly reduced the tendency to underweigh missing evidence.     

Comparison of contingent and noncontingent access to therapy dogs during academic tasks in children with autism spectrum disorder

This study compared contingent and noncontingent access to therapy dogs during educational tasks for children with autism spectrum disorder using a multielement design. The experimenters assessed whether initial preference for the dog predicted reinforcer efficacy and how preference changed across time. A higher response rate during contingent dog sessions than baseline sessions occurred for 4 out of 5 participants, suggesting that the dog functioned as a reinforcer. One participant engaged in a high rate of responding in both contingent and noncontingent dog conditions. Preference assessments revealed idiosyncrasies, suggesting that further research is needed into the predictive nature of initial preference assessments with animals as part of the stimulus array. The experimenters also analyzed salivary cortisol before and after sessions to determine if learning about the upcoming interaction with a dog reduced salivary cortisol in children. Cortisol was variable across participants, with only some deriving a potential physiological benefit from expecting to interact with the dog.     

A comparison of picture and GIF-based preference assessments for social interaction

Researchers have recently begun to evaluate video-based preference assessments; however, only two studies have evaluated the efficacy of this preference assessment modality in assessing preference for social interactions. Four individuals diagnosed with autism spectrum disorder who could not match pictures or graphic-interchange-format images (GIFs) of social interactions to in vivo interaction participated. We compared picture and GIF-based paired-stimulus preference assessments for social interaction and evaluated the preference assessment hierarchies by conducting a concurrent-operant reinforcer assessment including all social interactions. The GIF-based preference assessment produced similar hierarchies to the reinforcer assessment for all participants, whereas the picture-based preference assessments produced similar hierarchies to the reinforcer assessment for 2 of 4 subjects. Additionally, we conducted a modality preference assessment in which we evaluated participants' preference for viewing GIFs or pictures of social interactions, and found that 3 out of 4 subjects displayed a preference for GIFs.     

Validity of the multiple-stimulus without replacement preference assessment for edible items

Studies have shown that the multiple-stimulus without replacement (MSWO) preference assessment is an effective assessment format for identifying preferred items. However, it is possible that factors other than reinforcer quality might influence selections by some individuals when the MSWO array consists of edible items. The validity of the MSWO results was evaluated by comparing items identified as most and least preferred by the MSWO assessment in a concurrent-schedule (CS) format. Varying edible items were used in each study. The MSWO format accurately identified the higher quality edible item for most participants across studies (76%); however, the MSWO format did not predict the highest quality edible item in the CS assessment for 20% to 30% of participants in each study.     

Teaching children with autism spectrum disorder to ask “where” questions using a speech-generating device

Children with autism spectrum disorder (ASD) who have limited speech are often taught to communicate using a speech-generating device (SGD). We evaluated procedures for teaching a mand for information (i.e., Where is [item]?) using an interrupted behavior chain procedure. In Experiment 1, all participants (3 children with ASD who communicated using an SGD) acquired the target mand but transfer to a novel stimulus did not occur. In the second experiment, 2 participants were taught to approach alternative communication partners when the first partner did not provide the information. The second experiment also included procedures to test whether the responses were under the control of appropriate motivating operations (MOs). Generalization across communication partners occurred with both participants, but transfer across behavior chains with only 1 participant. The results of both experiments suggest that teaching multiple behavior chains and evaluating MO control may be necessary to establish generalized manding for information.     

Evaluating preference for and reinforcing efficacy of fruits and vegetables compared with salty and sweet foods

Children diagnosed with autism spectrum disorder are often more selective in their food preferences than their typically developing peers. Additionally, many preferred food selections have minimal nutritional value. We evaluated the preference for, and reinforcing efficacy of, fruits and vegetables compared with salty and sweet foods, which are often used as edible reinforcers. Multiple-stimulus preference assessments were conducted to identify preferred foods, and reinforcer assessments were conducted to determine the reinforcing efficacy of more preferred foods. Fruits and vegetables were sometimes preferred over salty or sweet foods and often functioned as reinforcers. Future research should incorporate fruits and vegetables into preference assessments when identifying putative reinforcers.