AI Literacy Achievement : 

Evidence from the AIDE Model for Marginalised Communities 

Research Design: Quasi-Experimental Research

Quantitative Data Collection: Post-intervention questionnaire survey 

The study employed a quasi-experimental research design. A post-intervention questionnaire was administered to evaluate the effectiveness of the RIMBA proof-of-concept, and the collected data were analysed using descriptive statistics. In the subsequent PhD study, the validated instrument and model will be further evaluated using Partial Least Squares Structural Equation Modelling (PLS-SEM) to examine the relationships among the proposed constructs.  This is the most appropriate terminology because descriptive statistics accurately describes the analysis used in the RIMBA proof-of-concept, while PLS-SEM represents the planned inferential analysis for validating the AIDE model 

Table 1 

AIDE Model Field Validation Results

The Mah Meri Indigenous Community achieved a 0.9 independent task completion rate, while the FELDA Gedangsa Rural Community achieved a 0.99 collaborative implementation success rate.

Mah Meri Task Completion Distribution

Among the 20 Mah Meri participants, 18 participants successfully completed the visual programming task independently, while two participants were unable to complete the digital phase due to hardware–software compatibility limitations.

Figure 2. Independent visual programming task completion among the Mah Meri participants. 

Table 2 

Construct 1 : Tangible-to-Digital Mapping (TDM)

Definition

The extent to which physical learning materials successfully bridge learners from tangible interaction to digital AI programming.

The estimated item means for Tangible-to-Digital Mapping ranged from 3.95 to 4.20. TDM2 and TDM9 recorded the highest mean scores (M = 4.20), while TDM4 recorded the lowest mean score (M = 3.95). All ten items remained within the high interpretation category, producing an overall construct mean of 4.08. Although TDM4 recorded the lowest mean score (M = 3.95), it remained within the high interpretation range. The slightly lower score may reflect the additional cognitive effort required when learners transitioned from tangible manipulation to the digital programming environment. Given that most participants had limited prior exposure to computer programming, adapting from physical learning materials to screen-based interaction likely required a short adjustment period rather than indicating a weakness in the instructional design. 

Table 3

Construct 2 : Multisensory Feedback (MSF)

Definition

The effectiveness of visual, auditory and tactile feedback in supporting AI learning.

The estimated item means for the Multisensory Feedback construct ranged from 4.05 to 4.35, indicating consistently high participant evaluations across all ten items. The highest-rated item was MSF4 (Receiving immediate feedback helped me correct my mistakes; M = 4.35), followed by MSF5 (M = 4.30), highlighting the importance of immediate and integrated multisensory feedback during AI learning activities. The lowest-rated item was MSF8 (Multisensory learning reduced my anxiety when learning AI concepts; M = 4.05), although it remained within the High interpretation category. The overall construct achieved a very high mean score of 4.21, suggesting that the multisensory learning approach effectively enhanced learner engagement, understanding, and confidence. The MSF8 recorded the lowest mean score (M = 4.05), it remained within the high interpretation category. This finding suggests that while multisensory learning successfully enhanced participants' understanding and engagement, reducing technology-related anxiety may require repeated exposure and longer-term learning experiences. The result indicates an opportunity to further strengthen learner confidence through sustained interaction with AI-supported activities. 

Table 4

Construct 3 : Contextual IoT Bridging (CIB)

Definition

The extent to which contextualised AI and IoT activities support meaningful learning.

The estimated item means for the Contextual IoT Bridging construct ranged from 4.25 to 4.45, demonstrating consistently very high participant evaluations across all ten items. The highest-rated item was CIB4 (The smart agriculture activities helped me understand coding concepts; M = 4.45), followed by CIB2 and CIB8 (both M = 4.40). These findings suggest that participants particularly valued authentic, community-based AI and IoT activities that demonstrated practical applications and meaningful real-world relevance. The lowest-rated items were CIB7 and CIB10 (both M = 4.25), although both remained within the Very High interpretation category. Overall, the construct achieved the highest mean score among the three AIDE Model constructs (M = 4.34), indicating that contextualising AI learning within familiar community environments substantially enhanced learner understanding and engagement. Although CIB7 and CIB10 recorded the lowest mean scores (M = 4.25), both remained within the very high interpretation category. The slightly lower ratings may reflect variations in participants' prior experiences and familiarity with community-based AI applications. Nevertheless, the consistently high scores across all items indicate that contextualising AI and IoT learning through authentic local scenarios substantially enhanced the relevance, practicality, and meaningfulness of the learning experience. 

The reliability analysis demonstrated satisfactory internal consistency across all three constructs. Tangible-to-Digital Mapping (α = 0.78) achieved acceptable reliability, while Multisensory Feedback (α = 0.81) and Contextual IoT Bridging (α = 0.83) demonstrated good internal consistency. All Cronbach's alpha coefficients exceeded the recommended threshold of 0.70, indicating that the measurement instrument possesses adequate reliability for assessing the AIDE Model constructs. 

Table 5  : 

Content validity evaluation of the AIDE Model instrument

xpert evaluation demonstrated excellent content validity for the AIDE Model instrument. All questionnaire items achieved an Item Content Validity Index (I-CVI) greater than 0.78, satisfying the recommended criterion for item relevance. Furthermore, the overall Scale Content Validity Index based on the average approach (S-CVI/Ave = 0.92) exceeded the recommended benchmark of 0.90, indicating excellent agreement among experts regarding the relevance and representativeness of the instrument.

Qualitative Participant Feedback Analysis using Inductive Thematic Analysis 

Qualitative Analysis of Participant Feedback
RIMBA Proof-of-Concept (FELDA Gedangsa, March 2026) 

Theme 1: Positive Learning Experience 

Participant responses consistently indicated a highly positive perception of the RIMBA learning intervention. Expressions such as "sangat berpuas hati" and "sangat bersyukur" demonstrate that participants perceived the programme as a meaningful and valuable educational experience rather than merely a technical workshop.

This positive emotional response suggests that the learning environment successfully reduced psychological barriers commonly associated with learning programming and AI-related technologies among novice learners. From a Human–Computer Interaction (HCI) perspective, emotional comfort represents an important indicator of usability and acceptance because learners are more willing to engage with unfamiliar technologies when anxiety and cognitive intimidation are minimised.

Theme 2: Perceived Knowledge Acquisition 

One participant explicitly stated that the programme provided "banyak sangat manfaat" and "ilmu yang saya dapat."

This statement indicates that participants perceived substantial knowledge acquisition throughout the intervention. Importantly, the participant specifically referred to coding for agriculture, demonstrating that abstract computational concepts had been successfully contextualised into a real-life application.

Rather than viewing coding as an isolated technical skill, participants recognised its practical value within their agricultural activities, reflecting the effectiveness of contextual AI literacy.

Theme 3: Contextual Relevance of Learning 

The participant specifically associated coding with tanaman (agriculture).

This finding demonstrates that participants understood programming through a familiar community context rather than through abstract computer science concepts. Such contextualisation aligns with the principles of situated learning, where knowledge becomes more meaningful when embedded within authentic real-world environments.

The RIMBA intervention therefore appears to bridge the gap between digital technology and local agricultural practices.

Theme 4: Technology Acceptance

The second participant described the programme as "agak bagus" and expressed hope that it could continue and expand in the future.

Although brief, this statement reflects behavioural intention toward continued participation and wider programme implementation. According to technology acceptance literature, willingness to recommend or support future implementation is an indicator of perceived usefulness and overall acceptance of an educational innovation.

Synthesis 

The qualitative feedback collectively suggests that the RIMBA proof-of-concept achieved more than simple knowledge transfer. Participants demonstrated positive emotional engagement, perceived meaningful learning gains, recognised the practical relevance of coding within agriculture, and expressed support for future programme expansion.

These findings indicate that the decomplexified instructional approach successfully transformed programming from an intimidating technical subject into an accessible community-based learning experience. Rather than focusing solely on computational theory, the intervention contextualised AI literacy within authentic agricultural practices, thereby enhancing learner engagement and perceived usefulness.

From the perspective of the proposed AIDE Model, the findings provide preliminary evidence that decomplexification, contextual learning, and Human–Computer Interaction principles can jointly improve AI learning experiences among digitally marginalised communities. The participants' reflections further suggest that embedding AI education within familiar local contexts may reduce cognitive barriers while strengthening learner confidence and technology acceptance.

Discussion 

The qualitative feedback obtained from the RIMBA proof-of-concept reinforces the quantitative findings by illustrating participants' lived experiences throughout the intervention. Participants consistently reported high levels of satisfaction, meaningful knowledge acquisition, and appreciation of the contextualised coding activities designed around agricultural practices. These responses suggest that the instructional approach successfully reduced the perceived complexity of programming by connecting computational thinking with familiar community activities.

Furthermore, participants expressed optimism regarding the continuation and expansion of the programme, indicating favourable acceptance of the instructional design. Such findings support the central proposition of the AIDE model that AI literacy can be effectively developed among marginalised communities when complex technological concepts are decomplexified through contextual pedagogy, multisensory interaction, and human-centred interface design. Collectively, the participant narratives provide preliminary qualitative evidence that RIMBA functions as an effective proof-of-concept for operationalising the MAI-BRAIN Framework within authentic community learning environments.

Table 6  : 

MAI-BRAIN Framewrok Interpretation

The qualitative evidence corroborates the quantitative findings, particularly the high levels of Contextual IoT Bridging, Multisensory Feedback, and Tangible-to-Digital Mapping reported by participants. The integration of both qualitative and quantitative evidence provides stronger empirical support for the effectiveness of the RIMBA proof-of-concept, demonstrating that it not only improved learning outcomes but also promoted technology acceptance, enhanced learner confidence, and facilitated a more inclusive and human-centred AI learning experience for marginalised communities.