On-the-Job Training Practices

Picture this: It's 2030. Sarah, an instructional designer, arrives at work. But she's not building courses anymore. Instead, she's orchestrating an AI ecosystem:   • 𝗛𝗲𝗿 𝗥𝗲𝘀𝗲𝗮𝗿𝗰𝗵 𝗔𝗴𝗲𝗻𝘁 analyzes learner patterns overnight, identifying skill gaps traditional assessments miss.   • 𝗛𝗲𝗿 𝗗𝗲𝘀𝗶𝗴𝗻 𝗔𝗴𝗲𝗻𝘁 proposes personalized learning pathways optimized for different preferences and business needs.   • 𝗛𝗲𝗿 𝗖𝗼𝗻𝘁𝗲𝗻𝘁 𝗔𝗴𝗲𝗻𝘁 generates adaptive scenarios and practice opportunities that respond in real-time to performance.   • 𝗛𝗲𝗿 𝗔𝗻𝗮𝗹𝘆𝘁𝗶𝗰𝘀 𝗔𝗴𝗲𝗻𝘁 measures actual behavior change and business impact, not just completion rates. Sarah's role? 𝗦𝗵𝗲 𝗰𝗼𝗻𝗱𝘂𝗰𝘁𝘀 𝘁𝗵𝗶𝘀 𝗔𝗜 𝗼𝗿𝗰𝗵𝗲𝘀𝘁𝗿𝗮, 𝗺𝗮𝗸𝗶𝗻𝗴 𝘀𝘁𝗿𝗮𝘁𝗲𝗴𝗶𝗰 𝗱𝗲𝗰𝗶𝘀𝗶𝗼𝗻𝘀 𝗮𝗻𝗱 𝗺𝗮𝗶𝗻𝘁𝗮𝗶𝗻𝗶𝗻𝗴 𝗵𝘂𝗺𝗮𝗻 𝗰𝗼𝗻𝗻𝗲𝗰𝘁𝗶𝗼𝗻. The old world is gone. No more rigid courses. No more clunky LMS platforms. No more pre-built curriculums. Instead, there are 𝗶𝗻𝘁𝗲𝗹𝗹𝗶𝗴𝗲𝗻𝘁 𝗹𝗲𝗮𝗿𝗻𝗶𝗻𝗴 𝗲𝗰𝗼𝘀𝘆𝘀𝘁𝗲𝗺𝘀 with conversational AI tutors providing just-in-time coaching, dynamic pathways adapting to performance, and seamless knowledge networks connecting people to expertise exactly when needed. Learning is part of the work, not separate from it. People solve real challenges with AI support, peer collaboration, and adaptive guidance. 𝗧𝗵𝗲 𝗜𝗗𝘀 𝘄𝗵𝗼 𝘁𝗵𝗿𝗶𝘃𝗲 𝘄𝗶𝗹𝗹 𝗹𝗲𝗮𝗿𝗻 𝘁𝗼: - Design human-AI collaboration workflows - Build agent systems that amplify human potential - Create learning ecosystems, not learning objects - Measure real-world impact, not engagement metrics 𝗧𝗵𝗲 𝗾𝘂𝗲𝘀𝘁𝗶𝗼𝗻 𝗶𝘀𝗻'𝘁 𝘄𝗵𝗲𝘁𝗵𝗲𝗿 𝘁𝗵𝗶𝘀 𝗳𝘂𝘁𝘂𝗿𝗲 𝗶𝘀 𝗰𝗼𝗺𝗶𝗻𝗴. 𝗜𝘁'𝘀 𝘄𝗵𝗲𝘁𝗵𝗲𝗿 𝘆𝗼𝘂'𝗹𝗹 𝗯𝗲 𝗹𝗲𝗮𝗱𝗶𝗻𝗴 𝗶𝘁 𝗼𝗿 𝘀𝗰𝗿𝗮𝗺𝗯𝗹𝗶𝗻𝗴 𝘁𝗼 𝗰𝗮𝘁𝗰𝗵 𝘂𝗽. 𝘞𝘩𝘢𝘵'𝘴 𝘰𝘯𝘦 𝘈𝘐 𝘵𝘰𝘰𝘭 𝘺𝘰𝘶 𝘤𝘰𝘶𝘭𝘥 𝘴𝘵𝘢𝘳𝘵 𝘦𝘹𝘱𝘦𝘳𝘪𝘮𝘦𝘯𝘵𝘪𝘯𝘨 𝘸𝘪𝘵𝘩 𝘵𝘩𝘪𝘴 𝘸𝘦𝘦𝘬?

𝐇𝐨𝐰 𝐜𝐚𝐧 𝐰𝐞 𝐝𝐞𝐬𝐢𝐠𝐧 𝐀𝐈 𝐜𝐨𝐦𝐩𝐚𝐧𝐢𝐨𝐧𝐬 𝐭𝐡𝐚𝐭 𝐩𝐫𝐢𝐨𝐫𝐢𝐭𝐢𝐬𝐞 𝐥𝐞𝐚𝐫𝐧𝐢𝐧𝐠 𝐨𝐯𝐞𝐫 𝐩𝐞𝐫𝐟𝐨𝐫𝐦𝐚𝐧𝐜𝐞? 𝗧𝗵𝗲 𝗰𝗼𝗿𝗲 𝗮𝗿𝗴𝘂𝗺𝗲𝗻𝘁: Large language models (LLMs) are rapidly transforming knowledge work by improving the quality and efficiency of tasks such as writing, coding, and data analysis. However, their growing use in education has exposed a learning-performance paradox: while they can enhance short-term task performance, they may also undermine genuine learning, including cognitive growth, knowledge transfer, and metacognitive development. I'm thrilled to share that I have been working with the dream team Dragan Gasevic, Shazia Sadiq, Lixiang (Jimmie) Yan, Jason M. Lodge, Jason Tangen, Paul Denny, Kristen Eignor DiCerbo, Simon Buckingham Shum, and Ryan Baker to ground a response that tackles this head-on: 𝗕𝘂𝗶𝗹𝗱𝗶𝗻𝗴 𝗔𝗜 𝗖𝗼𝗺𝗽𝗮𝗻𝗶𝗼𝗻𝘀 𝘁𝗵𝗮𝘁 𝗣𝗿𝗶𝗼𝗿𝗶𝘁𝗶𝘀𝗲 𝗟𝗲𝗮𝗿𝗻𝗶𝗻𝗴 𝗼𝘃𝗲𝗿 𝗣𝗲𝗿𝗳𝗼𝗿𝗺𝗮𝗻𝗰𝗲. 📄 Preprint available here: https://lnkd.in/gbts24D2 This paper addresses the question of how artificial intelligence should be designed and used to support learning rather than merely improve immediate outputs. We introduce the concept of AI learning companions, defined as adaptive, pedagogically informed, LLM-powered agents designed for integration into learning environments, built around three foundations: 🧠 𝗣𝗲𝗱𝗮𝗴𝗼𝗴𝗶𝗰𝗮𝗹 𝗳𝗼𝘂𝗻𝗱𝗮𝘁𝗶𝗼𝗻: How students learn with AI, grounded in deep and interactive learning, guided scaffolding, metacognitive development, and contextual, authentic engagement. 🔄 𝗔𝗱𝗮𝗽𝘁𝗶𝘃𝗲 𝗳𝗼𝘂𝗻𝗱𝗮𝘁𝗶𝗼𝗻: How AI learns about students, through a continuous cycle of capturing learner data, modelling cognitive and affective states, adapting instruction, and evolving over time. 🛡️ 𝗥𝗲𝘀𝗽𝗼𝗻𝘀𝗶𝗯𝗹𝗲 𝗱𝗲𝘀𝗶𝗴𝗻 𝗳𝗼𝘂𝗻𝗱𝗮𝘁𝗶𝗼𝗻: How companions act with integrity, remaining transparent, accountable, inclusive, and secure. The paper features five case studies spanning global K-12 tutoring (𝗞𝗵𝗮𝗻𝗺𝗶𝗴𝗼), AI-assisted co-creation of educational content (𝗥𝗶𝗣𝗣𝗟𝗘), scaffolded programming support (𝗖𝗼𝗱𝗲𝗛𝗲𝗹𝗽), AI-assisted course discussion and formative feedback (𝗝𝗲𝗲𝗽𝘆𝗧𝗔), and institution-wide AI companion design for higher-order learning (𝗥𝗲𝗰𝗮𝘀𝘁 at UTS), each illustrating what a deliberately designed AI learning companion can look like in practice. 𝗧𝗵𝗲 𝗸𝗲𝘆 𝗶𝗺𝗽𝗹𝗶𝗰𝗮𝘁𝗶𝗼𝗻: Designing AI for learning is not a minor prompt-engineering adjustment. It requires building systems that model learners, adapt to their needs over time, and are grounded in pedagogical principles, shifting from optimising task outputs to cultivating learners who are more reflective, more metacognitively aware, and better equipped to learn independently in an AI-rich world. 💬 Have you developed or used AI tools that prioritise learning over performance? We'd love to hear about them in the comments.

Can Software Double Learning? Reflections on the Andhra Pradesh PAL Study A major evaluation in Andhra Pradesh’s government schools has made global headlines. A team led by Nobel laureate Michael Kremer finds that Personalised Adaptive Learning (PAL) software doubled measured learning rates for 14,000 students across 1,200 schools. For Class 6, this meant the equivalent of two years of progress in just one year. This is an important result. For decades, Indian classrooms have struggled with overcrowding and diverse learning levels. PAL addresses this by tailoring practice questions to each child’s ability, something a single teacher with 40–60 students cannot easily do. The Andhra trial confirms what earlier experiments in India and Kenya (Muralidharan, Singh, & Ganimian, 2019; Banerjee et al., 2016) had shown: adaptive technology can deliver real improvements in maths and language learning. Yet the story is more complex. Learning Beyond Test Scores The “doubling” claim rests on test outcomes. While foundational skills are vital, education is not reducible to exams. Creativity, empathy, higher-order thinking skills, critical thinking, and cultural understanding remain invisible to the software. Narrowing education to what algorithms can track risks shrinking the purpose of schooling. Unequal Gains The study found boys gained more than girls. This gap reflects entrenched inequities in digital access and social norms, not just software design. Andhra’s classrooms remain stratified and resource-divided. Without deliberate safeguards, technology will mirror and even reinforce these inequalities rather than correct them. The Politics of EdTech The trial is significant because it is publicly funded, unlike many private EdTech apps. But key questions persist: Will PAL support teachers or erode their authority? Who owns the vast learning data generated? Are public schools becoming sites for global EdTech experiments? As research on EdTech warns (Williamson & Hogan, 2020; Selwyn, 2022), technology can bring surveillance, privatisation, and market logics into public education. A Way Forward The Andhra study matters because it shows that personalised learning works. But scale-up must be careful: (i) Keep teachers central and build their professional capacity. (ii) Address gender, community, and rural divides in access and outcomes. (iii) Measure learning more holistically, beyond maths and language scores. (iv) Ensure local ownership of data and curriculum. Adaptive software can accelerate test outcomes, but education’s task is far larger: shaping thoughtful, ethical, and culturally rooted/critical human beings. That remains beyond the reach of any algorithm. Critical EdTech India (CETI) #EducationResearch #EdTech #PublicPolicy #LearningOutcomes #AdaptiveLearning #GlobalEducation #CriticalEdTech #EquityInEducation #DigitalLearning #EdTechForGood #LearningEquity #PolicyAndPractice #IndianEducation #GovtSchools #PAL #EducationReform

Could a new layer of openness help unlock truly adaptive learning? Most learning materials still come in a single flavour: one language, one reading or grade level, one version for all. Open Educational Resources (OER) made a leap forward with free, openly licensed, remixable content. Yet most OER remain ‘fixed’, to be used ‘as is’. 𝐀𝐈 𝐭𝐨𝐨𝐥𝐬 𝐚𝐫𝐞 𝐛𝐞𝐠𝐢𝐧𝐧𝐢𝐧𝐠 𝐭𝐨 𝐬𝐡𝐢𝐟𝐭 𝐭𝐡𝐢𝐬 𝐩𝐚𝐫𝐚𝐝𝐢𝐠𝐦 With AI tools, this is changing. For example, UNICEF’s Accessible Digital Textbooks tool can already convert a single source file into multiple languages and accessible formats for learners with disabilities. Prompts can provide deeper personalisation, and emerging prompt libraries are a good start. But what if we reimagined prompts in the spirit of OER? What if they were openly licensed, shared, remixed and iteratively improved? This leads to a question:  𝐂𝐨𝐮𝐥𝐝 𝐰𝐞 𝐢𝐦𝐚𝐠𝐢𝐧𝐞 𝐬𝐨𝐦𝐞𝐭𝐡𝐢𝐧𝐠 𝐥𝐢𝐤𝐞 𝐎𝐩𝐞𝐧 𝐋𝐞𝐚𝐫𝐧𝐢𝐧𝐠 𝐏𝐫𝐨𝐦𝐩𝐭𝐬 (𝐎𝐋𝐏)? Picture prompts not as one-liners, but as modular, openly licensed learning objects that span subject areas, contexts, themes, and pedagogical models. They could: ● Live in a public, version‑controlled repository under open licences, where community feedback and up‑votes both surface the most effective versions and guide ongoing iteration ● Adapt automatically to learner and teacher profiles (such as language, reading level, accessibility needs, preferred themes and other interests) ● Support peer review, localisation, reuse across platforms, and model-agnostic design ● Integrate with national digital learning systems rather than sitting on the side‑lines We’re already seeing glimpses - like Gemini Gems and custom GPTs that package multi-step logic. But combining open licensing, profile-aware design, cross-platform integration, and iterative improvement could unlock more meaningful, accessible and scalable personalisation across contexts. There would be many challenges, of course: digital divides, bias in outputs, language limitations, and - who builds and maintains it? Would love to hear from others - educators, developers, AI practitioners, accessibility advocates, startups, and anyone exploring the intersection of learning and technology: What might help - or hinder - such a system to accelerate personalised learning opportunities across different contexts?

Self-Learning Agentic Graph Systems and Their Integration of Knowledge, Memory, and Recommendation Mechanisms 🎛️ Self-learning agentic graph systems represent a sophisticated fusion of knowledge representation, adaptive learning, and intelligent decision-making capabilities. A complex architecture can now combines structured knowledge graphs with dynamic learning mechanisms, creating systems that can continuously evolve while maintaining coherent knowledge representation. This structure enables both stable knowledge representation and dynamic adaptation. 1️⃣ . Pattern Recognition Layer: This foundational component enables the system to identify recurring patterns and relationships within the data. Its value lies in: - Statistical Pattern Analysis: Processes large amounts of data to identify significant correlations and trends, enabling the system to recognize meaningful patterns in user behavior and data relationships. - Relationship Discovery: Automatically identifies connections between different pieces of information, helping build a richer knowledge graph. - Anomaly Detection: Identifies unusual patterns or deviations that might indicate new learning opportunities or potential issues. 2️⃣ . Feedback Integration Layer: This component processes various forms of feedback to improve system performance. Its value comes from: - User Feedback Processing: Incorporates explicit and implicit user feedback to refine recommendations and knowledge representation. - Performance Metrics: Tracks and analyzes system performance to guide learning and adaptation. - Adaptation Signals: Generates signals that trigger specific adaptations in response to feedback. 3️⃣ . Knowledge Evolution Layer: This layer manages how the system's knowledge base grows and changes over time. Its value derives from: - Knowledge Graph Updates: Continuously updates the graph structure to reflect new learning and insights. - Weight Adjustments: Modifies relationship strengths based on observed patterns and feedback. - Structure Evolution: Allows the knowledge graph to evolve its structure to better represent complex relationships. 4️⃣ . Temporal Adaptation Layer: This component manages how knowledge and patterns change over time. Its value stems from: - Temporal Relevance: Evaluates the current relevance of stored information and patterns. - Historical Pattern Analysis: Identifies how patterns and relationships evolve over time. - Decay Management: Gradually reduces the influence of outdated information while preserving valuable historical patterns. 5️⃣ . Context Understanding Layer: This layer enables the system to understand and adapt to different contexts. Its value comes from: - Situational Analysis: Interprets the current context to provide relevant responses. - User Context Integration: Incorporates user-specific contexts into learning and recommendations. - Environmental Awareness: Considers broader contextual factors that might influence system behavior.

📉 Learning outcomes in Côte d'Ivoire remain low with only 17% of students reaching proficiency in mathematics, and nearly half of grade 4 students not able to read a simple sentence. Through the World Bank's Youth-RISE project, supported by the Mastercard Foundation, we piloted AI-powered adaptive learning platforms across 25 TVET institutions with approximately 2,000 students. 📊 Impact analysis shows active users gained 0.234 standard deviations in mathematics (about 11 months of learning) and 0.121 standard deviations in French (about 6 months). ⚡ The most striking finding: struggling learners in the bottom 15% progressed 6 to 15 times faster than average performers, showing how adaptive technology can meaningfully reduce educational inequalities when students actively engage with it. https://lnkd.in/dvemmuMd

🚀 New OECD Report: AI to Support Neurodivergent Learners in Vocational Education and Training Artificial Intelligence is not just about automation, it’s about inclusion. The latest OECD report explores how AI and advanced technologies can create more adaptive, accessible, and neuroinclusive VET systems. Here’s what stood out 👇 🔹 Adaptive & Personalized Learning AI can tailor instruction, feedback, and materials to diverse learner profiles, supporting students with ASD, ADHD, dyslexia, and other learning differences in real time. 🔹 XR & Immersive Practice Virtual and Augmented Reality create safe, repeatable environments for learners to rehearse job skills, interviews, and complex workplace tasks, building confidence before entering real-world settings. 🔹 Accessibility by Design Text-to-speech, speech-to-text, live captioning, and generative AI tools are transforming how learners engage with curriculum, empowering independence and participation. 🔹 Bridging Education to Employment AI tools can support CV writing, interview preparation, executive functioning, and workplace adaptation, smoothing the transition from VET to the labor market. But the report also raises critical questions: ⚖️ data privacy ⚖️ algorithmic bias ⚖️ overreliance on AI ⚖️ ethical implementation The message from the report is very clear: AI must be implemented thoughtfully guided by accessibility frameworks, teacher training, inclusive design principles, and strong policy guardrails. If we get this right, AI won’t replace educators. It will amplify human potential, especially for learners whose talents are too often undervalued. Curious: What are you seeing in your institution or organization? #aiineducation #neurodiversity #vocationaleducation #accessibility #inclusiveeducation #edtech #futureoflearning #assistivetechnology

Weekend Research Deep Dive #04 — AI + XR in Adaptive Learning Systems (2024–2025) Continuing the weekend series where I break down one high-value research area for builders, educators, and XR/AI practitioners. This week’s theme: How AI-powered agents, generative models, and adaptive interfaces are reshaping immersive learning (VR/AR) — shifting XR from scripted simulations to learner-aware systems. Latest reads (open access / accessible summaries): 1. “Adaptive Gen-AI Guidance in Virtual Reality” Lau et al., 2024–2025 https://lnkd.in/gYx37iyX Shows how GenAI-driven guidance in VR increases learner engagement using multimodal signals, while highlighting the need to balance personalization with cognitive load. 2. “CLAd-VR: Cognitive Load-based Adaptive Training in VR” Matam et al., 2025 https://lnkd.in/gjpDQZsf Introduces a VR training system that adapts instruction in real time based on cognitive load, keeping learners in an effective challenge zone. 3. “Adaptive VR Learning Using Pedagogical Models” Marougkas et al., 2025 https://lnkd.in/grgcNnD7 Demonstrates how adaptive VR improves learning flow, motivation, and skill progression compared to fixed XR lessons. Why this area matters (and why it’s worth your coffee): AI + XR = personalized, adaptive, skills-focused learning. The research points to three shifts: Adaptive interaction loops: Systems adjust guidance and difficulty continuously based on learner behavior. Cognitive-aware training: XR design is moving beyond immersion toward managing cognitive load. Evidence-driven learning: Behavior traces and in-scenario analytics enable measurable skill outcomes. 3 takeaways for practitioners: 1. Start with one adaptive layer before scaling complexity. 2. Treat cognitive load as a design variable, not an afterthought. 3. Instrument XR early to track learning, not just engagement. Question for the community: If you were designing an AI-infused XR learning experience today, where would you start? (A) GenAI-guided tutoring   (B) Adaptive instruction   (C) Difficulty adaptation   (D) In-XR analytics  #XR #AI #HCI #EdTech #ImmersiveLearning #SpatialComputing #Research