面向儿童执行功能评估的低碳自适应导入设计:一种测量效度保持型人工智能交互框架

面向儿童执行功能评估的低碳自适应导入设计:一种测量效度保持型人工智能交互框架

蓝炜¹ 任怡霖²

(1.庆应义塾大学媒体设计研究科，日本，999081;

2.*通讯作者，暨南大学珠海校区，广东珠海，519070)

摘要:计算机化执行功能评估能够提升儿童认知测评的标准化、时间控制与评分自动化水平，但儿童在正式测验前的任务进入困难可能与目标执行功能混杂;与此同时，若自适应支持延伸至正式计分阶段，又可能改变任务结构并影响测量效度。面向儿童心理评估中的绿色数字化、教育公平与低资源部署需求，本文提出一种“测量保持型自适应导入”设计框架。该框架将系统支持严格限定在任务导入与练习阶段，并在正式计分阶段锁定刺激材料、呈现时序、反应映射、反馈政策、评分规则和汇总指标。研究以维度变化卡片分类任务和倒背数字广度任务为应用场景，基于重复错误、长时间犹豫和输入映射困难等练习阶段遥测信号，设计简短提示、规则特异性动态示范和支架式恢复练习等轻量化、可撤回支持机制。基于清洗后的52名儿童会话数据及8名观察者配对评分，本文比较了标准练习与自适应练习条件下的任务进入、支持触发、练习恢复、正式阶段表现和现场可部署性。结果显示，自适应导入能够在练习阶段形成可审计的支持与恢复过程证据，并改善任务进入流畅性、错误后恢复、协调员负担和流程标准化;正式计分阶段未呈现一致的自适应得分优势，尤其在被直接支持的DCCS任务中，两种条件下正式准确率基本一致。研究表明，有界自适应的目标不是提高正式测验得分，而是在不改变正式测量核心结构的前提下，以低算力、可审计、边界清晰的人机交互机制降低儿童评估中的非目标交互负担与无效流程成本。由于本文尚未直接测量设备能耗、云端调用成本或碳排放变化，相关发现应被理解为具有潜在低碳部暑价值的初步证据，而非已经证明碳减排效果。

关键词:儿童执行功能;计算机化评估;低碳人工智能;自适应导入;测量效度

Low-Carbon Adaptive Onboarding for Child Executive Function Assessment

Lan Wei ¹Ren Yilin²

(1. Embodied Media Laboratory, Graduate School of Media Design, Keio University, Hiyoshi Campus, Yokohama,

Kanagawa, 223-8526, Japan: 2Jinan University Zhuhai Campus, 519070, Zhuhai, Guangdong Province, China)

Abstract: Computerized executive-function assessment improves standardization, timing control, and automated scoring in child cognitive assessment; however, task-entry difficulties before scored trials may be confounded with the target executive-function construct, while adaptive support that extendsinto scored trials may threaten measurement validity. This study proposes a measurement-preservingadaptive onboarding framework for low-resource and potentially low-carbon child psychologicalassessment. The framework confines support to task entry and practice, while locking formal stimuli,timing, response mappings, feedback policy, scoring rules, and summary metrics during scoredmeasurement. Using the Dimensional Change Card Sort and backward digit span as applicationscenarios, the system provides lightweight, withdrawable, and auditable support through briefprompts, rule-specific animated demonstrations, and scaffolded recovery practice triggered byrepeated errors, prolonged hesitation, and input-mapping difficulty. Based on 52 cleaned child sessionsand paired ratings from eight observers, the study compares standard and adaptive practice in terms oftask entry, support triggers, recovery, formal-block performance, and field deployability. Results showthat adaptive onboarding generated auditable practice-stage evidence and improved task-entrysmoothness, post-error recovery, coordinator burden, and process standardization. Formal-blockmetrics did not show a uniform adaptive score advantage; in the directly supported DCCS task, formalaccuracy was broadly comparable across conditions. These findings suggest that bounded adaptivityshould be understood as a validity-preserving interface strategy rather than an intervention for raisingtest scores. Because energy use, cloud-call cost, session-level carbon emissions, and invalid-testreduction were not directly measured, the low-carbon claim is limited to potential deployment valuerather than demonstrated carbon reduction.

Keywords: child executive function; computerized assessment; low-carbon Al; adaptive onboarding; measurementvalidity

[原文下載][Download PDF]