Page 147 - AIH-2-4

Artificial Intelligence in Health AI-driven personalized learning in residency

of interactions, and completion rates of educational
modules. This study also assessed subjective feedback to
gauge resident satisfaction and perceived benefits of the
AI-driven learning approach. Data collection extended
to performance outcomes on simulation-based quizzes,
aligning engagement metrics with academic progress.

2.2. AI-driven learning platform
The study utilized the edYOU platform, an adaptive,
AI-based educational system designed to personalize
learning experiences in graduate medical education. The
platform’s adaptive content delivery is grounded in key
principles of adaptive hypermedia, which personalize
learning pathways based on user modeling. The platform
15
also integrates two core components: The personalized
ingestion engine (PIE) and the intelligent curation engine Figure 1. Adaptive AI beings on the edYOU platform
(ICE). The PIE technology, which follows adaptive learning Abbreviation: AI: Artificial intelligence.
principles, continuously curates diverse instructional
16
materials from validated academic sources. It uses – including total time spent on the platform, percentage
natural language processing to tailor content delivery of topics completed, frequency of interactions with AI
based on each learner’s demonstrated knowledge, response beings, and depth of follow-up queries – were passively
patterns, and engagement history. This dynamic tailoring and systematically logged by the platform’s analytics
allows the system to adjust the difficulty and sequencing of infrastructure. These objective measures of engagement
content in real time, supporting individualized progression were paired with performance on simulation-based
through the curriculum. The ICE, in turn, ensures the formative assessments and standardized in-service
integrity, relevance, and safety of the educational material examination scores to evaluate learning outcomes. In
by implementing automated content validation protocols, addition, subjective feedback was gathered through
including toxicity filtering, bias mitigation, and source post-intervention surveys, which included quantitative
verification. The above-mentioned safeguards were ratings and open-ended questions to assess perceived
purposely designed to maintain academic rigor while usefulness, ease of use, and the platform’s ability to support
minimizing the risk of misinformation or inappropriate individualized learning. This multi-modal evaluation
content. Medical residents interacted with AI beings strategy allowed data triangulation across usage,
(Figure 1) capable of conducting naturalistic dialogue, performance, and user perception domains.
providing immediate, context-sensitive feedback, and 2.4. Statistical analysis
tracking learner progress through analytics-driven
personalization. This combination of adaptive delivery Descriptive statistics were employed to summarize
and content governance enabled a responsive, structured resident engagement and platform utilization metrics,
learning environment aligned with competency-based including total hours spent on the platform, number of
medical education principles. completed modules, and frequency of interactions with
AI beings. Pearson correlation coefficients were calculated
2.3. Implementation and data collection to assess the association between platform usage and
Medical residents were introduced to the AI-driven flipped academic outcomes, with quiz performance as the primary
classroom model through an initial orientation session dependent variable. Statistical significance was determined
that provided an overview of the platform’s capabilities, a priori at a two-tailed alpha level of p<0.05. All analyses
including navigation, interaction protocols with the were conducted using the IBM Statistical Packages for the
AI beings, and expectations for independent learning. Social Sciences Statistics for Windows, version 28.0 (IBM
Following orientation, residents were encouraged to Corp., Armonk, NY, USA).
utilize the platform beyond scheduled didactic sessions 3. Results
to supplement their self-directed study. The platform’s
design allowed for asynchronous interaction, enabling Data are expressed as mean and standard deviation unless
learners to access content, receive feedback, and revisit otherwise specified. Descriptive statistics are summarized
complex concepts at their own pace. Engagement metrics in Table 1.

Volume 2 Issue 4 (2025) 141 doi: 10.36922/AIH025130023