Cogniti - webinar - Student engagement with an Ai agent

 Notes taken at this month's Cogniti Connect. (the second session)

 Notes and video of the first session found HERE -  'when a tool meets a community'

This month's topic is 'what makes students engage with an AI agent or not?'  aka 'built it and they will come!' The session is facilitated by Olga Kozar.

A survey on the webinar indicated the majority of attendees have poor Agent usage rates.

Three case studies are presented:

- the 10% to 60% engagement shift Cogniti in chemistry: student engagement across different implementations. Presented by Dr. Pierre Naeyaert from the University of Sydney. Students access the Cogniti agent Canvas LMS. Pierre does not have physical access to the students! Agent available since August 2024 with 3,5000 students enrolled. By the start of the year, only 180 students have accessed. From start of this year, Pierre worked towards increasing engagement. A report helper agent was developed for 71 students have to access this before they can submit the report. 11 students engaged. Another agent developed in semester 1 2026 for 140 students. Agent focused on practicing their viva. 50 students used this agent. An embedded video to provide support for how to use the agent may have been useful in bringing students in.

First year survey on AI indicate students use AI 2/5 and 3/5 responded that they did not. Chem1 Cogniti agent was rolled out to 1900 students. 1280 students used. Again a video was used to show students how to use the agent. 

Q&A followed. Ideas on providing information to students to encourage using agents shared (posters, naming the agent with a friendly name, 'how to use' video)

- Design dilemma: How to 'use it to leverage it' with Robert Annabel from University of Sydney on 'to and through the door: the question generator in ECON6021'. This writes practice questions and is available to 143 Masters students from the first week of the course. Access through Canvas and information also on the course outline. Students reminded about the agent during lectures. 

24 days (7 weeks) there were 41 conversations, 88 messages. Postulated that students had high workload and concerns on the reliability of the agent. A conversation starter could be useful to get them going. Worked through several system prompt redesigns including - agent to ask for missing topic. Allow one question at a time. Solution to be provided after an attempt. 

In week 12 tutorial taken to do a live demo of the tool. Information on the pedagogical purpose of the agent. Class discussion was useful. Students provided feedback. This session helped increase conversations to 109 and messages to 833. 

Discussion on how much reminder and time need to put into encouraging students to use the agent. Comments in the chat revolved around how students 'want the answer' and not willing to engage with 'Socratic' style conversations. 

Dr. Carrie Falling from University of Otago shared that first impressions are important. Short time window to obtain their attention and to see the advantages for using the agent. Agent performance has to align with the learning outcome, be easy to use and concise feedback (not long screes of text). Well-designed agents are therefore important to ensure that the user experience is encouraging further use.

- Cogniti sand boxes: 'Sandboxes for critical AI literacy and graduate attributes' with Dr. Carmen Vallis from the University of Sydney. The course Business consulting practicum simulates a project, with industry input. AI changes what 'professional' and 'authentic' mean. The AI agent is an 'AI consultant'.

Cogniti sandbox used to have students create their own agents (with speech and avatars). A safe space to experiment, with secure data and Canvas access. Students could select their own AI models For teachers, a sandbox is easy to set up and they can see agents and conversations in one place. Teachers can share editing.

A 2 hour workshop helped students write a professional profile, choose the info to share (or withhold), built an agent and add to their profile. The then tested their agent (and their values) against real scenarios. Then debriefed in groups on personal, professional and ethical implications.

Small cohort of 24 students. There were 104 conversations across 1 workshop and 2 sessions. Students needed more time and some did not create one. It was messy but the students were engaged. 6 students returned, unprompted to use the AI more. Language shifted between informal to formal. Reflected that next time, they would create a future consultant persona instead of students' professional profile. The sandbox needs to be simplified to have less strict prompt control, use new in-build features and provide more time for students to work on the AI agent.

Q & A followed. 

Olga introduced the next session in October - Bendigo Kangan Institute on Cogniti AI tutors and Manakau Institue of Technology and Unitec on their bespoke platform for end to end AI agent governance and oversight. Also call for presentations on this year's Cogniti Symposium. 

UNESCO report - An algorithm in the room

UNESCO report 2026 on AI in education - The algorithm in the room : seeing and confronting the implications of AI for the future of education. 

Similar to the four scenarios from Singapore, summarised recently in this blog, this report collates 26 think pieces organised into 6 sections.  

After an introduction on the concept of 'algorithms everywhere' along with a discussion on educational purposes and the difficulties of aligning these algorithms with these. Each section presents several think pieces. 

Part one asks 'who will hold agency, autonomy and oversight'?

Part two on 'How will young people develop relationally, emotionally and cognitively?

Part three goes into 'what will be meaningful and relevant for pedagogy and assessment?

Part four moves in 'how will we think, question and express ourselves?

Part five poses the challenges on 'how will we govern education for the common good?

Part six closes with 'how might AI affect imagination about the future?

Each section has several articles, many speculative case studies. Topics discussed provide food for thought. They also provide good conversation starters and work well as thought provokers. 

The conclusion discusses the 3 horizons to empower education in algorithmic times - to see, to steer and to rewrite the algorithm.

Many of the key ethical and educational challenges are discussed through the report. Overall, a report to dip in and out of as interests in each topic arises.

ATAIN - Using LMS 'Nuku' to read and engage with policy - Dr. Lorena Gibson

An AI policy was designed for the Cultural Anthropology programme at Te Herenga Waka-Victoria University of Wellington. The presentation, hosted by ATAIN is provided by Dr. Lorena Gibson - lecturer in cultural anthropology.

The presentation is on 'Discussing AI use with cultural anthropology students'. Shared the guidelines for using AI - AI assessment scale (Perkins, Furze, Roe & MacVaugh, 2024) and how this has been used with her colleagues and students.

Her students work through the guidelines through a LMS module. There are 6 modules and students have to complete a quiz. Modules on [ need to complete before assessment!, AI, how and why to use, how to use AI in assessments, acknowledging Ai use, what not to do with AI, lecturer's own practice with AI.

Shared content of some modules which provide not only links and information but also examples which are of relevance to their context. Modeling by Lorena, and transparency from her examples, set up good practice and show authentic use of AI. 

Shared the section on using AI with course tutors. A pragmatic approach and for them to reinforce the advice provided through the AI modules. Good to see this as well as most of the time, teachers do not access information provided to students.

Interesting Q & A followed.

Important to also ensure students understand the difference between various assistive technologies and AI. Students are embarrassed or ashamed of using AI and clear, visible and educative information is always good practice. AI anxiety is very real and reported in a recent 2026 report on NZ students well-being.

Discussion on the challenges encountered in academic settings. In all, educators need to be sensitive to learners' difficulties. However, students need to engage in 'productive struggle', otherwise, they do not actually attain academic writing skills. 

Ai agents - use in education - Sacha Garrity from Unitec

Sacha Garrity, academic advisor in digital learning from Unitec, presents on how AI agents are used to support learning. The event is convened by Whitirea/Weltec. The work has received national recognition through the Tertiary ICT Association. 

Lee Smith from Whitirea/Weltec welcomed everyone. A karakia was offered before introductions to Sacha and Sharnell Aumua. Their presentation is on the Ako AI agents project: Nursing.

Agents have been used for about a year. The Ako AI agents project involved supporting kaiako to customise mini ChatCPTs, hosted on Cogniti to be used by ākonga anytime when required, help to be responsive to ākonga needs and provide additional support to kaiako. Agents are embedded in Moodle so students do not have to move between two platforms.

An overview of Cogniti provided. Agent development informed by design-based research and Te Tiriti o Waitangi health principles. Agent developed with kaiako through co-design.

Quality assuring the agent is important. Testing undertaken through a broader group of kaiako. Follow a test protocol - does it do what is intended, promote learning, avoid unwanted responses. Then test with ākonga and gain feedback.

In short - interative design and test, extended test by kaiako, followed by student test and final Q & A. 

Shared the nursing Ako AI. 3 nursing kaiako for each agent - (de-escalation agent, deteriorating patient and drug calculations. 4 developers and technical support. Stressed importance of drawing from and integrating Te Tiriti o Waitangi health principles.

Evaluation involved questionnaire, focus groups. Two iterations of testing and data collection for each agent. 

Went through the drug calculation agent capabilities - providing fomulae for conversions/rations and calculations. Provide tutorial support. A demo ensued - basically a quiz, providing support when the student is unsure or unable to do the calculation. 

ākonga feedback indicated that their learning confidence and competence was enhanced; user experience and design interactivity was appreciated; future AI tools and expanded learning support indicated - better integrations between LMS and Cogniti, videos to do revision, gamification / progress bar useful.

Demonstrated the deteriorating patient agent. A scenario agent. Two scenarios - hospital and community. Agent generates verbal and non-verbal behaviours, reactions based on student's response. New patient profiles generated at each use along with clinical/vitals. Feedback involved debrief, reflection and overall feedback - a challenge to build.

The agent has a 'toolbox' which is a link to 'help' in using the agent. Demo of agent in role as patient, perceptor and going through the debrief loop. 

Overall, positive responses from students. Students trusted using the agents as they were developed by their tutors. Interest in having a range of agents to support learning across the nursing curriculum. Support for more clinical scenarios prior to going out to clinical.

Q & A 

Oral responses - challenge with Te Reo Māori as pronunciation not the best. 

24/7 support useful. Agents able to provide more in-depth feedback when they have time or need to undertake practice. A safe place to practice or clarify concepts etc. can be done. 

Intentional learning is a key to developing how the AI agent is structured. Cogniti is user friendly and does not need IT degree to set up. 

Volume of use brought up. Currently Cogniti coping but not sure of maximum number if a agent is open to entire institution e.g. for learning, academic or library support.

Toolbox discussed and how it can be used intentionally. Access to documents supporting the learning can also be accessed through the toolbox.

Security aspects discussed. Important to only provide resources which are relevant and Cogniti itself should not have access to internal organisational data. 

Time from start to implementation averaged 3 - 4 weeks - mostly for SMEs to test, run staff test groups and gather feedback etc. 

Four future scenarios for training and adult education in Singapore - horizon 2035

This report - Four future scenarios for training and adult education in Singapore - Horizon 2035, is written by Ang, B.J., Huang, M. & Chen, Z., the Institute of Adult Learning, Singapore University of Social Sciences. 

The horizon in the report is less than a decade away. For some, a long way off, but with AI, perhaps a shorter window may apply. The report is also deeply integrated into the Singapore training and adult education ecosystem, Therefore, the report needs to be read with the context in mind.

After a preamble section, explaining the approach, philosophies and methodology for the report, the future scenarios are presented. These are:

- the trusted system - very much based on the current system whereby national standards and skilled educators provide and enable completion of qualifications. Currency is the main challenge as technology moves quickly and standards and educators struggle to keep up.

- AI skills network - based in AI being used to provide personalised, continuous and 24/7 available training. Human support and accountability become difficult to built in when required.

- the reputation economy - draws on expertise of trainers but depends on reputation and networks.

- skills market - AI is used to build capability but visibility and trust on credentials is uneven.

The scenarios are of interest as they show how difficult it is to see into the future, given the pace of change and the many challenges related to shifting an embedded credentialling system. Each country will need to undertake similar exercises, to test the efficacy and flexibility of their current training and qualification system and to think through implications, challenges and solutions from within their own contexts. 

A new direction for students in an AI world - report from Brookings Institute

This report - A new direction for students in an AI world: Prosper, prepare and protect was prepared by M. Burns, R. Winthrop, N. Luther, E. Venetis and R. Karim and published at the beginning of 2026 by the Centre for Universal Education at Brookings Institute.

The report focuses on the formal school context. It is just over 200 pages including appendices.

It is framed on the premise that at this moment in time, AI can be useful in enriching learning experiences.

The 3 pillars anchoring the report are:

- Prosper - to use AI carefully to enhance student learning

- Prepare -ensure not only educators and students, but the communities they serve are prepared through understanding the implications of AI on education.

- Protect - safeguards are important to assure privacy, safety, and support emotional and social development.

Discussion is undertaken on how AI can enrich education AND how AI can also diminish learning if overused without adequate guardrails.

Annex A provides a good summary of various AI literacy frameworks.

ASQA - principles for the responsible use of AI in VET

The Australian Skills Standards Authority (ASQA) have released their principles for the responsible use of AI in Vocational Education and Training (VET). 

The guidance is summarised on the site - AI use in VET.  The site contains links to the principles, case studies, how the principles were developed, and an AI transparency statement. The principles are found in the section on the responsible use of AI in VET. They are high level and examples/case studies are provided to inform how the principles can be applied to teaching and learning.

Critical thinking - some models to inform teaching and learning

 The application of critical thinking towards evaluating information has been and is now an even more crucial skill. AI generates plausible responses but these must always be checked or triangulated. Ensuring everyone understands the underlying principles of critical thinking, and how the process to apply it towards becoming better equipped to deal with the avalanche of material many people wade through daily, is now more important than ever.

Three recent articles provide guidance. Each offers a model/framework to teach and preserve critical thinking skills.

1) 

Vendrell, M. & Johnston, S-K. (2026). Scaffolding critical thinking with generative AI: Design principles for integrating large language models in higher education, Computers and Education: Artificial Intelligence, Volume 10, https://doi.org/10.1016/j.caeai.2026.100572

Firstly, an update on critical thinking from Vendrell and Johnston (2026). The paper presents an analysis of research in cognitive psychology, educational theory and AI ethics. These are then synthesised to develop a pedagogical framework for integrating LLMS into teaching and learning. The principles of critical engagement are defined as: conceptual interpretation, intellectual curiosity, inferential reasoning, evaluative judgement, metacognitive regulation, and epistemic integrity. The paper recommends design principles that preserve cognitive friction, scaffold LLMs as partners, embed evaluation as standard practice, activate metacognitive regulation, encourage humility and curiosity, foster epistemic integrity, align assessment with thinking and balance AI mediated and AI free phases. 

2) 

Brcic. M. & Frljic, S. (2026). The effortless trap: Productive struggle, AI, and the illusion of learning. https://arxiv.org/pdf/2606.26181

A preprint which extends on the article in 1). it advocates for ways to avoid cognitve debt but ensuring that there are elements of 'productive struggle' in learning. The framework involves working through the following when working through new concept:

- Prime - help learners find a reason/motivation 

- Probe - provide a 'hard' problem so that learners have to tryout new ideas, learn from making mistakes and problem solve their way through.

- Point - using Socratic guides or cognitive apprenticeship to help make the problem solving strategies visible

Attach - connect the new to already learnt or new concept

Strengthen - reinforce through deliberate practice so that the concept/process is anchored.

3) 

Liubertaitė AD (2026) EFFORT-AI: an effort-preserving module for cognitively aligned human-AI learning. Frontier of Education. 11:1849821. doi: 10.3389/feduc.2026.1849821

This article also advocates for the need to ensure cognitive effort is undertaken to assure learning takes place. Introduces the EFFORT-AI model as a way to iintegrate AI into learning without allowing AI to do all the work.

- Elicit - Help learners make known to themselves their prior knowledge on the topic and their motivations to learn.

- Formulate - undertake a first attempt at a problem, either through retrieval or explanation

- Feedback - use feedback cycle to find out progress (or not)

Organise - make sense of the concept/process

- Reflect - review, revise and work out how learning undertaken , gaps in understanding and where to move to next

- Transfer - apply and review (spaced) to keep the concept/process active.

Articles 2) and 3) provide examples, mostly in STEM topics but frameworks cane be adapted to many other disciplines. The frameworks in themselves are therefore useful as starting points for adapting signature disciplines and pedagogy towards pedagogical integration of AI.

Being you:: A new science of conciousness - book overview

e-book borrowed from the local library and read on Libby, across a wet winter weekend. This book 'Being you:: A new science of consciousness' by Anil Seth, was one of the books (and authors) featured in  Pollan's book 'A world appears'.  The book was published in 2021 and sets out a biological/evolutionary perspective on consciousness. 

The book has positive reviews - exampled by this one from the Guardian. In itself, the book undertakes an update on neuroscience understanding of how the brain works and how its organic nature, may not be easily replicated. 

The book has 13 chapters with introductory and concluding chapters as well. The 4 sections, lays out the argument, each building on the next to provide details of how the workings of the brain, provides each person with a highly individualised perspective on the world at large. Being organic, the brain draws on the various senses, memories and experiences of each person. Hence, it is inherently a 'work in progress'. Each chapter has several pages of notes at the end.

The processes described in the earlier chapters explains some of the challenges for transforming people's mind-set. It takes effort and persistence to over-ride what each person has learnt, through a life of being in the 'university of life' and to shift perspectives. This is not impossible, which is where education plays an important role, but moving one's way of being, doing, thinking and feeling, requires strong social and individual motivation. It is an intense  undertaking to shift tracks which is why items like political affiliation (see Haidt's book - the righteous mind) takes some doing.

As to whether AI will become 'conscious' is dependent on how consciousness is defined. As per this book, the argument is that 'qualia' a sense of what something is and the degree of someness, is already difficult to quantify as each sentient being, brings their senses and experiences to gauge the experience. My estimation is that currently, Gen AI does not exhibit qualia. Gen AI cannot 'feel' or be conscious of its selfhood. It remains to be seen what may happen into the future. For the moment, our humanness affords us many qualities that are unique to us. 

The book has many examples explaining complex concepts and processes. It is an accessible read providing a good update on current understandings of how the brain works. Important now when claims are made each day about AI and yet, there are so many things AI is not able to do. AI looks proficient, but it is still a pattern matching algorithm. The patterns look plausible to us because our brain favour tidy patterns and black and white answers. However, many aspects of life are not quantifiable, predictable or straight-forward. As such, the human brain, with its inherent malleability and ability to 'think outside the square', draw on experiences /wisdom, and actually change tack when required, will still be difficult to replace.