Positive Persuasive LLMs

Dr Siying Hu at City University of Hong Kong talked on "Persuasive Simulations in the LLM Era" at the ANU AI, ML and Friends seminar. They suggested AI could be used in a positive way to counter the negative effects of social media and scammers. The Australian government is planning more support for the long term unemployed. Perhaps this technology could help build their confidence, as well as improving job seeking skills.  

Edutech Asia: What can Australia Learn About AI for Education?

I just paid my registration for EduTech Asia, in Singapore in November. Normally, I just wander around the exhibition and attend random sessions. This time I am on a mission, to find out what we should do with AI in Australian higher education. This is after completing the AI for Higher Education Faculty: Course by the Digital Education Council. The Australian National University paid for the course and prodded me to, reluctantly, do it. As with most learning, it wasn't a pleasant experience, but left me with the conviction there was something to this AI thing. 

AI for Higher Education Faculty: Course by the Digital Education Council

After congratulating myself for completing the short course "AI Literacy for All", by the Digital Education Council, I discovered there was another one "Certificate in AI for Higher Education - AI for Faculty", which perhaps I should have done instead. The "All" course was very short and simple, with content I already knew. The Faculty course is much longer (perhaps too long), with more depth (perhaps too much, with much I already knew). 

While providing a good introduction to AI for teaching staff, the course goes into too much detail on pedagogy, policy and projects. Much of this detail is material trained educators, administrators and project managers should already know. If they don't, they are only going to learn enough from a short course like this to get themselves, and their institution, into trouble. 

I suggest flipping the design: present the last section, from Singapore Management University (SMU), first. SMU provide practical techniques which educators could use to improve their teaching and reduce workload. The other material on policy and theory could be built on the back of that (or skipped for those of us already trained in it). As it is, the SMU material appears to be tacked on the end as an afterthought. Also I am still not sure exactly who, or what, the Digital Education Council is: how long has it been around, where is it based?

Synthetic Video Presentations

I had one "Wow!" moment in the course, with one video by Tamas Makany at SMU, about how to use an avatar for a training video. They explained how they used HeyGen software to create a video presentation featuring a synthetic version of themselves. It took a minute to realize that what I was looking at on the video was a an example of what was being discussed: it looked like a video of a real person. 

For several years I have been using text to speech software to create short video slide shows. These have a voice with an Australian male accent, which sounds remarkably like me. This use sa much simpler process than Dr Makany describes. I create a power-point presentation, with the narration for each slide in the notes. An online tool then turns that into a video. Being able to add a synthetic talking head would be useful. Dr. Chris Poskitt, also from SMU, described his use of a tool for adding questions to a presentation, unfortunately I couldn't work out what the tool was. The closed captions on the video were not available in English, which made it difficult. Also course system kept switching to high resolution video, which I kept switching back to low resolution, so my slow home wireless Internet modem could cope better. 

Course No Substitute for Teacher, Policy, or Project Training

The course gives a potted overview of teaching, but anyone teaching should know this stuff. There is an overview of how to do policy. There was a potted introduction on how to run an AI policy. However, as trained educator, I know about teaching and just need to know how to apply AI to this. As an experienced bureaucrat I know how to run policy processes, and don't need to be spoon fed this in an AI course. Similarly, as a certified computer professional I know how to run a project. The overviews of these areas in the course might be enough to give an untrained person the false impression they could do teaching, policy making or IT project management.

This course starts with a potted history of online learning and the application of AI to it. This history is somewhat short sighted, in that it claims online learning starts around the 1980s. I guess there isn't time for anything more detailed, but I am not sure why online learning need to be mentioned at all, as it is a separate topic from AI. You can do online learning without AI, and you can use AI for learning in a classroom. Is the Digital Education Council pushing a specific agenda? 

For me, the most significant insight from the course was the use of exiting learning tools enhanced with AI.  Tools such as drill and practice quizzes (Quizlet), lab simulations and chatbots can be created using AI. This can be done by the students themselves, with teacher guidance. This is an appealing approach, as it takes away some of the tedium creating the content for the tools, while retaining the teacher's oversight. It also involves students actively in their learning. 

More usefully Digital Education Council point to research they have done showing students across the world say they are already using AI. This is a useful wake-up call for educators who just wish AI would go away. But then some claims made I am not so sure about, for example that AI gives students instant feedback and that is a good thing. I am not so sure instant feedback is a good thing. We need to entourage students to think about the topic and the work they have done. If they get instant feedback from an apparently authoritative source, that cuts out reflection. However some of the pedagogy theory in the course is a bit dated, such as use of Bloom's Taxonomy.

The "AI Literacy for All" had videos with transcripts in multiple languages. However, this course appears to only have closed captions in English. This is a problem as I like to speed read through the transcript. Also some of the videos have no accompanying text. 

The course covers more than just Generative AI, including chatbots, which are applicable in education. In 2018 I took part in a workshop at University of Wollongong as part of the IEEE International Conference on Teaching, Assessment, and Learning for Engineering (TALE). Chi-Un Lei, Yuqian Chai, Xiangyu Hou, and Vincent Tam from University of Hong Kong, took us through using IBM Watson Assistant chatbot engine. My chatbot answered questions about an assignment deadline. It parsed the student's request for an extension and said "No". ;-) 

Some of the uses of AI proposed have ethical difficulties, such as an analysis of the student's performance across all courses and on phone calls. Other proposed uses are not new, such as analysis of the student's performance & personalization of lessons. Claims for these have been overstated. In a well design course it is possible to monitor student's performance & provide feedback via the regular assessment. 

Offering students ways to summarize their notes miss the point that the ability to summarize is one of the key ways students learn & is a skill valued by employers. 

No Dogfooding

Something which struck me a little way into my second Digital Education Council course on AI was the lack of dogfooding: that is doing what you advocate others do. That is, these are courses on how AI could be used in education, so where is the use of AI in these courses? 

Despite talking about how AI could be applied, these are very conventional online courses. There is a fixed linear structure: for each module you watch a video, read some readings, write some reflections, do a multiple choice quiz, then move to the next module. The only apparent use of AI is in some of the reflections, where the student is asked to make use of AI for an exercise ,then discuss the results. This is not necessarily a bad thing, this is a proven format for online educaion, which I am comfortable with. But where is the AI powered education DEC talks about, but doesn't do?

Something which adds to the difficulty of using AI, in association with the DEC AI courses, is that copying and pasting has been turned off in some modules. This also makes normal, non-AI study hard. Normally I would copy a question I am asked, then edit that to produce the answer, then paste it back. I would also write the answer in an editor which has spell checking. The AI courses, on my web browser, allow spell checking, but not correction. So I know I have words spelled wrong, but not what the right word is. I have a couple of goes at guessing what the correct spelling is, then give up and leave the misspelled word in. This is demoralizing and insulting for me. 

Western Blokes in Suits Bias

The early videos mostly feature a western males, in dark business suits or dark Silicon Valley expensive casual clothes. The dress standard gets more casual as the course progresses and the range of presenters increases, but there is, only one female presenter. The voices change between videos and I started to wonder if this was all AI generated, until the last more friendly and scruffy professor appeared, surprisingly from Singapore Management University. It would help if there was a more diverse range of presenters, although I must confess I am sitting here in my tweed coat, with tortoise shell glasses on. ;-)

Too Much Stuff

While talking about good education design, the course suffers from a problem common to such modules: too much stuff. The course took me weeks to do, even without the absurdly large number of optional readings. What is the point of offering a short course and then filling it with enough readings for a semester long program? 

While an experienced online student of education (seven years, three qualifications, in two countries), I struggled to keep on at this course, hour after hour, day after day, week after week. Around about the end of module 2, I started to loose the will to live. ;-)

Benefit from More Systematic Approach to Learning Design

Many of the steps recommended for use of AI in education also apply generally. To use the AI you need to know what you are teaching and how you want to assess it. Those steps also are important in education using pencil and paper.

Advertorial?

Some sections of the course described the benefits of particular products in such detail, that this started to sound like a paid advertisement for the product. 

Too Busy

While most of the course videos are slides with talking heads, a few are very busy animated product demonstrations. Text appears, a pointer moves around, windows pop up and disappear. This can be a bit overwhelming.

Weak on Privacy 

While the course repeatedly warns about privacy it demonstrates many tools which are hosted in the cloud in the USA. For institutions in Australia, Europe, and other parts of the world with privacy legislation, this makes them unusable.

Questions Require Knowledge Not in the Course

Several times the multiple choice questions ask about topics I don't remember being in the course. Perhaps I am just not a very attentive student, but perhaps the course was revised, material deleted and the questions not updated. As a student this was disconcerting, but generally I could guess the correct answer, by eliminating all the implausible options.

AI for Transfer Credit

One example which got my attention was AI course transfer credit.  I volunteered to help assess students applications for course credit. Eight long years later, I am still "temporarily" doing this thankless task. ;-)

If AI could be used to help assess course credit that would reduce the drudgery of the task, and as the course points out, improve equity.

Students not concerned about value for money

There are some interesting results from the Digital Education Council's survey of students internationally, although it was a relatively small number of students. What I found surprising was the concern about the value and value for money of their degree was the least of their worries, when it came to AI. Privacy was the top concern, which is not what I expected. A survey of staff had even fewer participants and while views of AI were more for than against, it is likely there is respondent bias in this survey (those keen on AI would be more likely to complete it).

Integrity?

The section of the course on integrity goes far beyond just AI and could be omitted. As an example, fake certificates used for enrollment predates AI by thousands of years. AI didn't cause this problem, even if it makes the problem worse. Solving the problem is not for educators to do.

No document presented by the student can be trusted. In every case you have to check with the institution it clams to be from if they really got those results. Some institutions have difficulty with this. When enrolling in Canada, I proudly presented my digital certificate from ANU. But I was asked for a certified paper copy. I had to pay to have this printed, signed, placed in an envelope, seals placed over the envelope seams, the seals each signed, placed in another envelope and couriered to Canada. This was accepted happily, but was all nonsense, as I could have easily faked it. 

One section of the course cited multiple universities who have decided Turnitin can't detect AI use. This is hardly a revelation, as anyone who has spent a few minutes trying it will see. ANU turned off Turnitin's AI detector on 1 January 2024.

Building AI

The section on building AI was unnecessary, and wrong, in parts. AI is a specialized form of IT and so IT professionals are needed for this. But the course suggests academics can build an AI system themselves, without mentioning of needing competent professionals. This is dangerous advice.

The second option for was for academics to do an AI startup, that is inventing the AI yourself, which is just silly. This is the equivalent of a catering company needing a computer, so soldering one together. While one company once did that 75 years ago, it is not an approach needed today. ;-)

The third option presented was to cooperate with corporations and startups, but without mentioning this is a high risk strategy (about 75% of computer projects fail). The most obvious, low risk, sensible option is not mentioned, which is to simply buy an AI service. There is no need to built it, invent it, collaborate, just buy an already built product. 

As an computer professional, I am often asked for recommendations on software and hardware. My advice is to buy what your friends and colleagues use. What you buy will be more likely to work and when you have problems you can ask people you know for help. The mistake many individuals and organisations make is to buy something new, on the assumption it must be better.

Decentralization

The course warns of the problems of decentralization, with different parts of the university having different approaches to AI, thus confusing the students. In a way this is understating the problem, with different instructors having different approaches: the decentralization goes all the way down to the individual. 

On the other hand decentralization doesn't matter so much to the student, as the university is just one organisation they have to deal with. At one point I was enrolled in two universities and a vocational education college at the same time, each with its own rules and timetables. Students will almost all be working so exposed to at least two sets of AI rules, one of their employer and the other where they are studying.

Guidelines

The section of the course on university guidelines was useful, as it included many examples, including an Australian one. 

Communication

The course got a little silly when it suggested communicating AI policy by email, intranet and web pages. What next, suggest what grade of paper is best for AI posters? ;-)

More seriously, communication of AI policy is important, but is no different than many other polices which need to be communicated. Just as no special strategy is needed for AI project management, none is needed for communications.

Fantasy Future

The last section of the course did not start well, presenting a future, where robotics & AI proves leisure time for the population & more time for students to explore their interests. This future has been predicted for other technological developments applied to learning, from correspondence courses to the Internet. However, they did not take into account the unchanged human nature. Traditional educational institutions have been able to accommodate each technological development, with the students & the community rejecting radical changes to educational delivery.

Despite this, I was most impressed with the last section of the course provided by SMU. Perhaps DEC should have left it to SMU to provide the entire course. But still, much of this was about the processes of education, which anyone teaching at a university should have learning in their basic teacher training (I did). One part I did not like was a 55 minute video. This was an interesting conversation between two experts, but 55 minutes is far, far too long. This should have been broken into smaller chunks.

One curious metaphor used was "hands on the wheel" of AI. My new car is controlled with voice and buttons. There is a steering wheel, but most of the time the car steers itself. That might be a better metaphor for AI in education: I select the destination, the car proposes a route, which I amend as needed then confirm. I then monitor the car's systems, while it dives itself. 

Singapore Education Leadership

Singapore's universities previously provided me with career, and life, changing inspiration. In 2013 I was giving a talk in Singapore on e-learning. One of the audience mentioned Singapore had an e-learning week, after a SARS outbreak, to prepare for future pandemics. As a result I was mentally and technically prepared for the COVID-19 outbreak a decade later. The SMU section of this course may well prove to be a similar pivotal moment in my teaching.

ACRONYMS

The educational acronyms used in the course were getting a bit silly, particularly the ABCDEF framework. What next an alphabet framework? Copilot wrote me one. ;-) 

A–Z Educational Acronym (All Letters in Order)

1. Active
2. Building
3. Curiosity
4. Drives
5. Engaged
6. Focused
7. Growth,
8. Helping
9. Innovative
10. Judgement
11. Kindle
12. Lifelong
13. Mastery.
14. Nurturing
15. Open
16. Problem‑solving
17. Questions
18. Reinforces
19. Skilled
20. Thinking,
21. Uplifting
22. Valuable
23. Wisdom
24. Xpanding
25. Youthful
26. Zeal.

Output

Here is one of the exercises I completed as part of the course. I am still not sure what I was supposed to do with it, as there was nowhere to submit it. This was done with the aid of MS Copiliot:

Learning Activity Brief: Exploring Project Reports with Microsoft Copilot

Learning Goal

Students will develop a clear understanding of the structure, purpose, and quality expectations of professional project reports in computing. By the end of the activity, students should be able to:

• Identify the essential components of a high‑quality computing project report

• Distinguish between technical detail, justification, and reflective analysis

• Evaluate the clarity, coherence, and professionalism of report writing

• Apply these insights to improve their own project reporting practices

Role of AI (Microsoft Copilot)

Students will use Microsoft Copilot as an exploratory and analytical tool. Copilot will support:

• Information gathering — e.g., asking Copilot to explain typical report structures, compare academic vs industry reports, or summarise best‑practice guidelines

• Critical evaluation — e.g., prompting Copilot to critique sample report excerpts or identify weaknesses in clarity, structure, or justification

• Reflection — e.g., asking Copilot how a reader might interpret a section, or what improvements would strengthen a report’s argument

Copilot is not used to generate a full project report. Instead, it acts as a catalyst for discussion, critique, and deeper understanding.

Activity Workflow (1 hour, small groups)

1. Initial Prompting (10 min) Groups ask Copilot to outline what makes an effective computing project report.

2. Deep Dive (20 min) Each group selects one component (e.g., requirements, design rationale, testing, evaluation) and uses Copilot to explore expectations, common pitfalls, and examples.

3. Critical Challenge (15 min) Groups provide Copilot with a short, deliberately flawed excerpt (provided by the instructor) and ask it to critique and suggest improvements.

4. Synthesis (15 min) Groups consolidate their findings into a concise artefact.

Student Product

Each group produces a one‑page “Project Report Quality Guide” that includes:

• A definition of the chosen report component

• Key quality criteria

• Common mistakes students make

• A short example of improved writing based on Copilot’s critique

• A brief reflection on how AI supported (and sometimes limited) their understanding

This product must be written in the students’ own words, with AI‑generated text clearly identified or paraphrased.

Assessment

This activity contributes to formative assessment and may be graded on participation or as a low‑stakes submission. Assessment focuses on:

• Accuracy and clarity of the group’s explanation of their report component

• Depth of analysis in identifying quality criteria and pitfalls

• Critical engagement with AI — evidence that students questioned, validated, or refined Copilot’s output

• Quality of the improved example — demonstrating understanding, not AI‑dependence

• Reflection on the role of AI in academic and professional writing

Rubric criteria can be aligned with:

• Understanding of report conventions

• Critical thinking

• Communication quality

• Responsible AI use 

Hacks of AI learning platforms pose a bigger challenge

The ABC reported today "Australian educational facilities impacted as 'criminal' hacks Canvas learning platform" (Scout Wallen & Monty Jacka). Fortunately Learning Management Systems (LMS), such as Canvas, don't contain much personal data about the students. They have the names, student numbers & email addresses, but other data about the student is stored in separate systems. There will be the content of the student's assignments and comments posted to discussion boards, but these are usually not very personal. Also there will be messages sent between students and staff, again, these don't contain much personal information. Students often overshare when applying for an assignment extension or regrade, but this is usually handled by a separate administrative system. What is of concern is where AI tools are used to personalize student learning. These will have access to more student data, making them a prime target for hackers.

Beyond personalisation: Human-centred and equitable learning in the age of AI at EDUtech Sydney 4 June

I have accepted an invitation to moderate a panel on "Beyond personalisation: Human-centred and equitable learning in the age of AI" in the Evolving University conference stream at EDUtech in Sydney, 11 am  4 June. 

On the panel are: 

• Sue Keay, Director, UNSW AI Institute
• Danny Liu, Professor of Educational Technologies, The University of Sydney
• Rose Luckin, Professor, UCL Knowledge Lab and Founder, Educate Ventures Research, Educate Ventures Research Limited

ps: The EDUtech_AU organizers have generously provided me with five tickets to give away to academic colleagues. Please let me know if you would like one. Only for people from not for profit educational institutions.

AI Enhancing Scams

Greetings from the "Mekong Update 2026: Cross-cutting challenges in mainland Southeast Asia" at the Australian National University. In a panel on Transnational Crime, David Boyle the Producer of Al Jazeera's "101 East", commented that AI had the potential to expand organised crime's use of scams. 

AI Literacy for All

Just completed the AI Literacy for All certificate* by the Digital Education Council for The Australian National University. When I was first invited to do the course I was skeptical, but despite having been to many training courses, workshops, symposiums & conference presentations on AI, I still learned some things. 

* Keep in mind this is not an AQF qualification. 

AI Boosterism in AI Training

I had some spare time so decided to get on with the AI training the Australian National University is encouraging staff to take. This is the usual online training module with some quizzes and exercises. It wasn't too bad, but as I know a bit about the topic, I was up to part 5 of 8 before it started telling me anything I didn't know. 

 At this point I was instructed to get AI to generate a LinkedIn post about AI & personal learning. While the previous modules did have some warnings about checking AI output for incorrect information and hallucinations, it was all a little too positive. At this point I used Copilot to generate the required LinkedIn post. This had only positive points on how AI could help with personal learning. As instructed I refined the result, but on my own initiative by asking for negative as well as positive points. That worked fine, but what struck me was that had I not asked, it would have all been positive.

Here are the negatives added:

"Negative impacts

• Over‑reliance on AI, which can weaken critical thinking and problem‑solving if learners default to AI-generated answers.

• Quality inconsistency, as AI can produce confident but incorrect explanations that mislead learners.

• Equity gaps, where those with better access to technology gain an advantage over those without it.

• Privacy concerns, especially when learning data is used to train models or inform performance decisions.

• Reduced originality, as learners may unintentionally mimic AI-generated patterns rather than developing their own voice or ideas.

These risks highlight the need for thoughtful integration rather than blind adoption."

Testing the Saftey of Smart Cars

Ivan Ang

Greetings from the Australian National University, where Ivan Ang is reporting his research on "User-centric Assessment of Robotic Systems via POMDP Planning of Adversaries". What got my attention was applying this to testing car autonomous driving. Recently I purchased a basic compact car. Despite the low cost, it came with a camera and radar for autonomous breaking, adaptive cruse control and lane keeping. I am trusting my life, and those of other road users, to these systems, so how well are they tested? Ivan pointed out that autonomous breaking was tested by pulling a dummy out in the path of the car on a string. The dummy's arms and legs are fixed, as its speed and direction. The developers of the system can optimize it for this test, but it then might do as well with real people. 

Australian Professional Standards Forum

Greetings from the Professional Standards Forum of the Professional Standards Councils. Today's forum is on AI, with Dr Ryan Payne, University of Canberra. 

The councils set professional standards, including computing. In return members of participating organisations have their liability capped. AI creates many challenges for organisations wanting to ensure their members act legally and ethically. 

Dr Payne claimed that NFTs are making a comeback, which I hadn't noticed. He also mentioned a number of other technologies which have been coming real soon for a long time, but which AI will make mainstream. 

Dr Payne claimed that AI had changed the words MPs use in Parliament and drafted in laws. He then touched on doomsday scenarios of AI optimizing human out of existence, or just subtly nudging human behavior. Of course, given governments, individuals and corporations have engaged in crimes against humanity, so could AI be worse? 

Today Meta started suspending the accounts of children, ahead of the Australian ban. It occurs to me that it should be possible to use AI to make a safer social media for children. I worry that in the next few weeks we will have millions of children suffering real pain and at risk of their lives, due to the ban. 

Then we got to play with Lego! Dr Payne used Lego to explain how a Large Learning Model (LLM) works. This was cleaver, as it got us out of the mode of just looking at Powerpoint slides.

AI Governance for the Global Financial System

Tom Worthington (me) and Jason Grant Allen from SMU
at Insights Forum 2025

Greetings from the Insights Forum roundtable on "AI Governance, stability and competitive dynamics: Aligning priorities for safe and effective AI adoption". I am not exactly sure what the forum is, or who by, But I happened to be in Singapore for another conference, so thought I would come along. Chatham House rules apply, so I can say what is said but not by whom. From some of the preliminary remarks I get the impression previous forums would have focused on blockchain and before that whatever tech was trendy.

The topic seems to be on risks from AI, rather than benefits. Speakers appear to be from international finance and banking. One speaker is expresing concern about the loss of bank to bank relationships in the Pacific. The result is remittances become more expensive and difficult. They suggest regulators can use AI to check compliance of banks quickly. This doesn't sound a compelling case when applied to traditional banks, as there aren't many and they have highly trained staff. However, it could open the market to new entrants. But then AI could be used to generate plausible fake details for scammers. 

Worryingly all the panelists appear to be experienced bankers, with no actual AI experts. I had been mentioned their companies have AI experts, but they haven't been invited to speak. We just hired someone who is a real AI expert. One panelists just said "Not just a talking head like me". They seemed to think this amusing. Would any other forum have someone saying how proud they were not to be competent. 

I asked the panel if the senior leadership teams in financial organisations have AI expertise, or are they all lawyers and accountants. Worryingly the first answer was that they spent several hours a day. That doesn't sound a good way to make decisions at our global institutions. 

One speaker made a useful point that banks face a threat from within where spies use a false identity to get a job in a company to steal information or money. Of course this happened in the past but is easier with remote employees.

This event raises the question as to if universities have looked in a similar way at their use of AI. Rather than just wondering about students cheating, how can AI provide better services to students. As an example, AI could suggest course credit for prospective students, based on past study and work l. This could allow universities to offer to cut a year or more off study time and tens of thousands of dollars. The AI could collate the required information and evidence. This would be a powerful incentive to enrol versus current practice, where the university says "enrol and we will see what we can do .. sorry now you have paid you are getting no credit". This is a personal experience I have had. At CIT I was assigned an expert to help me with ROL (Reconciliation of Prior Learning). I was given 80% of my qualification by RPL. In contrast a university held out the prospect of RPL. Already having two qualifications and a decade of experience (including designing a course for the university) I expected at least 33% credit, but got none.

ps: I bumped into Jason Grant Allen from SMU at the forum. 

edX for Credentialing

Greetings from EduTech Asia in Singapore, where Christyl Potgieter, Director, Enterprise Partner Marketing at edX and Dr Paulina Pannen, Senior Expert, Indonesia Cyber Education Institute, Universitas Terbuka are talking on AI, skills and learning. edX has been around for some time, delivering puerly online courses, which were marketed as "MOOCs" Massive Open Online Courses. These did not catch on at traditional universities, but as is being discussed could still find a role to provide shorter qualifications (marketed as "microcredentials") for those who are unable to attend a traditional university program. 

Live Hackerthon Pitches to Combat AI Cheating

Greetings from the Copeland Lecture Theatre at the Australian National University. The theatre has been repurposed as a pitch arena. Teams of students are making three minute presentations on an app they developed as part of the Software Construction course. There are only a few seconds for changeover between presentations. This form of experiential learning is one way to combat AI cheating. This allows the whole class to present in one hour long lecture slot. As well as being graded, students can win a prize and may be talent spotted by entrepreneurs, such as Ken Kroeger. While the presentations are digital, I noted that Professor Gretton used an pen and paper for notes on the work. 

As well as having to pitch live, each team has to provide an electronic archive of their work. This makes it much harder for a student to contract out the work, either to a human or AI software. One subtley in the way the pitches are made is that Zoom is being used, although we are all in the room together on this occasion, this allows all the presentations to be easily recorded and the option of a blended mode, wth remote students and teachers. 

Custom Questions for Each Student Based on Their Submitted Work?

Generative AI has the ability to create questions for students from the syllabus. This is very easy to do: you point the AI at the course documentation and tell it how many and what sort of questions you want. If multiple choice it also provides the answers. Could this also be done for each student, with custom questions based on their individual project work? 

Much has been written about oral examinations as the solution to students cheating. This is traditionally done for project based students, such as those undertaking a PhD. I had to go through a similar process at the end of my MEd, asked questions by a panel, based on what I submitted. However, this would be infeasible expensive for a program with hundreds, or thousands, of students. One option would be to have students question each other (this as been used in the ANU Techlaucher program). Also group projects reduce the workload, while introducing their own problems. 

Recently when attending training for Instructure's Canvas Learning Management System, being introduced at the Australian National University, I asked if each student could be given a separate file during an online test. The idea would be to provide them with a copy of their assignment, then ask them questions specifically about it. At the time I assumed the same set questions would be asked of every student. But it occurs to me now, why not ask them each a custom set of questions based on their work, and not provide a copy of the work to them?

Can we get the benefits of AI without harm?

Greetings from the Australian National University in Canberra, where Dr. Katy Gero, is speaking at AI & Friends, on "How do we get the benefits of LLMs without causing long-term harm?".They conducted an experiment where student written and AI assisted essays were mixed together and graded by a human marker. The marker couldn't tell which were from AI, and graded the AI submissions slightly higher than the human written ones. 

One insight from this research was the question of the role of writing in learning. Students may be willing to outsource essay writing as they don't see it as a core skill and something taking up time they could use for more important study. The student might use AI to write their idea better. This is a view I have sympathy with, as someone who barles passed English at school and failed French. I require a human editor, or tools for spelling and grammar, to be able to write well enough to be employed. 

This work also raises questions about how people cooperate to produce written works. I suggested Dr. Gero look to the history and politics of open source software for inspiration. 

ps: Will we have an accelerated Peter Principle, with people promoted far beyond their ability, due to AI assistance? ;-)

AI for Student Discuissions

At the moment I am taking part in a webinar on Instructure's IgniteAI. Today's topic is how to use AI to improve student discussions and feedback. This is to enhance the Canvas Learning Management System. Canvas are being modest in the claims for what AI can do for education, using terms like "We are currently developing ... to help ...", rather than saying this is a proven product which can replace the teacher. But the killer application is, I suggest is creating worksheets and quizzes based on your course content. These are useful, easy, quick and reasonably safe application of AI.

Instructure is also working on accessibility test and remediation features. This is possible with third party tools, but it will be handy to have it built in. However, I suggest learning designers still need basic training in accessibility. 

Insight is also providing a general purpose AI tool for the teacher, similar to that provided for office packages and software development tools. Insight claims to be able to use this to identify students needing extra help and provide them with material. One interesting point is if students also get a version of the AI tool to help them. This could be thought of as AI with trainer wheels. 

IgnightAI can be used to chart data from Canvas. This worries me a little, with academics and administrators coming up with correct, but misleading, charts. While much is written about the value of analysis of student data, this is not a substitute for educational training. 

Proposals to decouple learning from qualifications due to tech are not new

Miriam Reynoldson

Digital learning specialist Miriam Reynoldson has written a thoughtful article on how universities should change in response to AI. However, proposals to decouple learning from qualifications are prompted by each new technology in education. This happened with paper based correspondence courses, radio & TV education, the Internet and now AI. Each time promoters of the tech say that learning can be made cheap (or free), available to all. Each time the early promise is not realized and traditional education institutions incorporate and bureaucratize the process.

Universities are not the only places learning and qualification are done. In terms of the needs of society they are perhaps the least important. The vocational education and training (VET) sector keeps the lights on, literally, by training our electricians. That sector has long separated learning from qualifications.

Because lives depend on having competent technicians, grades in the VET sector are not quantified, aggregated or averaged. To receive a qualification, the student must pass each and every test, which is binary: either they are not yet competent (fail) or are competent (pass). Until the student can pass every test, they don't graduate. Doing exceptionally well in something doesn't count for anything.

Immediately after completing a Graduate Certificate in Higher Education at the Australian National University, I enrolled in the VET equivalent, a Certificate IV in Training and Assessment at Canberra Institute of Technology. There were some similarities. In each case I could receive credit for prior learning and experience gained in the workplace (RPL). However, at university the RPL is limited, and at a coarse course level of granularity.

At VET every little skill is separately evaluated for RPL and this can be for your entire qualification so you need do no courses. I obtained 80% of my VET qualification by RPL. But this is unusually high, as I had just completed a university qualification in education, had a decade of experience teaching, and had headed my profession's national learning unit.

There are also industry certifications which have assessment but no learning component. Major computer companies have certifications in specialized skills. Whole industries have grown up providing students with preparations for these tests. Some universities and VET institutions integrate preparation for these tests into their curricular.

It is difficult to untangle the purposes of learning. As an example, I enrolled in video production for training at ACT TAFE (which CIT used to be called). This was because I wanted to learn to make training videos. At the end I was surprised to be handed a certificate, as I didn't know this was a formal program and I had been assessed.

While I have referred to VET, vocational elements of university programs have some of the same assessment principles. There are knowledge and skills which every professional need and accreditation required that each and every graduate has these.

Separating permission to practice from the learning and teaching at universities would plunge them into a much deeper crisis. If the university is not undertaking one of its key role: providing trained professionals, there will be little reason to find them, or for students to enroll. Universities emphasize the value of their graduates to the economy as, apart from research, that is all they have to offer.

My preferred approach would be for school graduates to undertake the minimum of VET training to get a job. We they find a field they like they would do more VET training, if needed up to the degree level. If they needed more education they would enroll at university. If they showed exceptional ability, this could be at the doctoral level. 

Quantum AI for Defence at ANU

Dr. Nils Herrmann, 
 Quantum Brilliance

Greetings from the Australian National University where Dr. Nils Herrmann, from Quantum Brilliance is talking on Quantum Machine Learning on Diamond-based Quantum Computers. Quantum Brilliance is an Australia/German startup building quantum computers using diamond microchips. Their Canberra research center ios at the ANU. What got my attention, and probably did of governments and funders, is using the quantum effect for machine learning. Nils envisions a large quantum computer using training data and then a quantum computer on a plug in board in a desktop computer using the model created. The aim is to have a board rugged enough to be used in an aircraft, with obvious military applications. 

"This talk introduces the concept of Quantum Utility - Quantum Brilliance’s vision for practical quantum advantage - and presents early gate-based QML results achieved on an on-site, room-temperature quantum computer. It then highlights the work of the newly formed Quantum Utility and Exploration (QUTE) team, whose mission is to uncover utility through control modes that operate closer to the physical diamond-based hardware. The talk concludes with an introduction to quantum reservoir computing, showcasing QUTE’s latest simulations on future single- and multi-NV systems that establish a compelling platform for near-term, practical quantum machine learning."

Developing a national sovereign AI strategy

For the first time in a long time this week I missed the launch of the annual Australia’s Digital Pulse, by the Australian Computer Society at the National Press Club (earlier in the day I twisted my knee running for a bus). The report has 10 recommendations. My comments on some:

1. Develop a national sovereign AI strategy: The ACS is calling for a "long-term vision for sovereign AI development". However, I suggest this should be looking for short term measures. If Australia looks to the long term the strategy will be out of date before it can be implemented. As I suggested for the ACT Government last week, we can look to countries such as Singapore, for mature AI strategies to emulate.
   
   
2. Develop an innovation strategy: The Australian Government is to release a "Strategic Examination of Research and Development", this year. Aligning workforce readiness with innovation by encouraging entrepreneurship is my day job. Government can do more to support startups and scaleups. One area is for defence, where the nation needs new capabilities quickly. Cancelling a few failed big ticket defence projects reliant on overseas suppliers could provide thousands of billions of dollars for local innovation.
   
   
3. Government co-investment in scaleups: Not so sure about this recommendation, as it sounds dangerously like government trying to pick winners. Apart from strategically important areas, such as defence, energy security and availability of vaccines, government should avoid direct investment in scaleups, as they are really, really bad at it. 
4. Greater incentives for R&D in AI and tech adoption by businesses: ACS point out there is a skills gap in the use of AI by business. The obvious solution, I suggest, is training. Many of the computer project students I teach are planning to work in AI. To them it is not new and exotic, just a tool they are very familiar with. We need vocational education programs which similarly give business people a deep understanding of AI. 
5. Executives take a digital skills health check: Rather than trying to get C-suite leaders up to scratch on digital capabilities, I suggest political parties, government agencies and company boards need to hire more technocrats. We need some digital professionals running the nation and corporations, not just lawyers who learned a bit about computers.
6. Promote entry-level pathways for cybersecurity professionals: ACS point out that we simply can't get enough computer professionals to fill cyber security positions. What we can do is take people qualified in allied fields and train them up. This can be done online with vocational education techniques, rather than in university lecture theatres.
7. Implement an ‘earn while you learn’ scheme: ACS proposes business and government sharing the cost of worker retraining. Unfortunately what has tended to happen is as federal government introduces subsidies for university and vocational education, business and state governments have stopped funding. We have the technology to do the retraining (I spent a decade learning how this works). The problem is to make a watertight agreement on funding which business and states will abide by. Assuming Jobs and Skills Australia release their national skills taxonomy promptly, this can be used (if not someone else can). This can be aligned with the UK based Skills Framework for the Information Age (SFIA), used by ACS. The match will not be perfect. When ACS commissioned me to design a course in Green Computing (way back in 2008), there were no relevant skills in SFIA. I sent SFIA HQ a copy of my course and they added green skills, but that took considerable time.
8. A national commitment to alternative tech pathways: ACS argues for greater trust in vocational qualifications by business. Another approach to this, I suggest, is the use of the Vocational Degrees, added to the Australian Qualifications Framework in February 2025. With this entry level staff can obtain a certificate and then go on to a degree at the same vocational institution. This should give business more comfort as to the depth of training provided. 

Rebuilding the public service with AI

Image generated by Google Gemini,
for a pod based
interplanetary transport system

Greetings from the AI Colab at the Australian National University in Canberra, where I am taking part in "From red tape to algorithm: Reimagining the public service in an AI age". I am not exactly sure what the event or the AI Colab are, but as it is being hosted in the ANU College of Systems and Society, near my office, I thought I would come along. When I registered I though this was by the ASD-ANU Co-Lab, but apparently it is another collab. Last week I was at an event on AI and ACT Government. 

We were give the exercise to image a future service based on three randomly chosen cards. My team came up with pod based interplanetary travel. I asked Google Gemni for an image and presentation and it came up with the appended description and image. 

After another warm up exercise, we got on to actually talking about public service. In particular, in Australia we have three levels of government (in Canberra just two). I argue there is in effect a fourth level for many people, the cluster housing management where they live (I live in an apartment building which has two levels of management, for parks, energy, and waste disposal). 

One obvious, and for me overriding, point for me on AI and government is that the people who do it need to be trained in how to do AI. Some of the teams and interns aI am teaching are studying AI and implement projects using it for government. ?". 

At the end of the event I asked "What is the AI Colab?". Several people tried to explain it, without success. I found a website, but it still is not clear. I suggest a plain English description is needed.

The Pod Transport System

Here is a presentation outline for a pod-based interplanetary travel system, styled with a "NASA punk" aesthetic. This style combines the optimistic, analog-heavy feel of 1970s and 80s space exploration with a grittier, more lived-in, and often retro-futuristic edge. Think a mix of *Alien* and the Apollo program.

---

"## **Title Slide:**

**Title:** Project: Peregrine - A New Era of Interplanetary Travel

**Subtitle:** The Pod-Based System for Human Exploration

**Image:** A stylized, high-contrast image. On the left, a detailed drawing of a modular "Peregrine" pod, with visible welds, rivets, and a faint glow from its propulsion system. On the right, a silhouette of a massive, retro-futuristic "mother ship" in deep space. The background is a mix of star fields and faint, blueprint-like lines.

**Aesthetic:** NASA logo in a slightly distressed font, analog dials and gauges in the corners, text in a classic, bold, sans-serif font like Helvetica or similar.

---

## **Slide 2: The Challenge**

**Title:** The Interplanetary Problem

**Image:** A grainy, slightly distorted photo of a classic rocket launch (like the Saturn V), with a faded red filter over it. Overlayed text reads: "Long-Duration Missions. High Cost. Limited Flexibility."

**Bullet Points:**

* **Long Transit Times:** Current propulsion methods make journeys to Mars and beyond measured in years, not months.

* **High Development Costs:** Each new mission requires a bespoke, single-use vehicle, driving up costs and limiting mission frequency.

* **Crew Burnout:** Confinement in a single vessel for years on end leads to psychological and physiological stress.

* **Lack of Redundancy:** A single catastrophic failure can doom an entire mission.

**Aesthetic:** Blueprint-style diagrams of existing spacecraft, annotated with problem descriptions. "CLASSIFIED" or "INTERNAL USE ONLY" stamps in the corner.

---

## **Slide 3: Our Solution - Project Peregrine**

**Title:** A Paradigm Shift: Modularity & Reusability

**Image:** A clean, exploded-view diagram of the Peregrine system. It shows a central "mothership" or "hub," with multiple individual "pods" docking with it. Each pod is labeled with its function (Crew, Cargo, Science, Habitation).

**Bullet Points:**

* **Modular "Pod" System:** The core of our design. Individual, self-contained units for specific functions.

* **The *Aethel* Mother Ship:** A reusable, nuclear-thermal propulsion (NTP) vessel designed to transport pods between planetary orbits.

* **Efficient Mass Transport:** The pods are designed to be launched individually from Earth via conventional rockets or space elevators, then assembled in orbit.

* **Mission Flexibility:** The same system can be reconfigured for a Mars mission, an asteroid mining operation, or a deep-space science probe by simply swapping out pods.

**Aesthetic:** The diagram should look like a cross between an engineering schematic and a piece of concept art. Hand-written annotations, faded paper textures, and faint grid lines.

---

## **Slide 4: The Peregrine Pods**

**Title:** Anatomy of a Pod

**Image:** A detailed, cutaway view of a single "Crew Pod." The interior is visible, showing a small living space, a bunk, a control panel with chunky buttons and toggle switches, and a small porthole. The aesthetic is cramped but functional, with visible wiring and exposed pipes.

**Bullet Points:**

* **Crew Pods:** Accommodate 2-4 astronauts. Equipped with life support, individual bunks, and personal communication terminals. Designed for comfort and psychological separation.

* **Habitat Pods:** Larger pods with communal spaces, exercise equipment, a medical bay, and hydroponics for fresh food.

* **Science Pods:** Dedicated labs with specialized equipment for astrophysics, geology, or biology experiments.

* **Cargo Pods:** Simple, robust containers for equipment, supplies, and rover parts.

**Aesthetic:** The diagram should be highly detailed, almost photorealistic, but with a grainy, analog quality. Text labels should look like they were typed on a typewriter.

---

## **Slide 5: The Journey (Phase 1)**

**Title:** From Earth to Orbit

**Image:** A cinematic, slightly overexposed photo showing a series of Peregrine pods being launched into low-Earth orbit (LEO). They look like industrial, chunky capsules. The Earth is a beautiful blue orb in the background.

**Narrative:**

1.  **Launch:** Individual pods are launched from Earth using existing launch infrastructure. This distributes the massive weight and logistical burden.

2.  **Orbital Assembly:** The pods are then docked with a pre-positioned *Aethel* mother ship in LEO. Crew pods are the last to arrive, ensuring a shorter total time in space for the crew before the main journey.

3.  **Final Checks:** Systems are checked and re-checked by automated probes and the crew. Final trajectory is calculated.

**Aesthetic:** The images should be evocative and slightly cinematic. Faded colors, visible lens flare, and a sense of scale and ambition.

---

## **Slide 6: The Journey (Phase 2)**

**Title:** The Interplanetary Transit

**Image:** A high-resolution, artistic rendering of the *Aethel* mother ship in deep space. The NTP engines are glowing with a faint, blue-green light. It is a massive, complex structure with many pods attached. The image captures the loneliness and vastness of space.

**Narrative:**

1.  **Nuclear-Thermal Propulsion (NTP):** The *Aethel* uses its powerful, efficient NTP system to accelerate the entire assembly. This provides a much faster transit time than chemical rockets.

2.  **Crew Comfort:** During the journey, the crew can move between different pods, reducing the feeling of confinement. They have access to a larger hab pod and can conduct research in the science pods.

3.  **Redundancy:** In case of a major system failure on one pod, the crew can evacuate to another, or the affected pod can be jettisoned.

**Aesthetic:** The image should be beautiful and awe-inspiring, but with a sense of the gritty, industrial nature of the machine. Visible heat vents, solar panels that look slightly weathered, and the distinct, analog feel of the Peregrine design.

---

## **Slide 7: Arrival & Mission Deployment**

**Title:** Separation and Landing

**Image:** A diorama-like image showing a single, streamlined Peregrine pod detaching from the *Aethel* in orbit around Mars. Below, on the Martian surface, another landing pod is deployed, its retro-thrusters firing. The Martian landscape is red and dusty.

**Narrative:**

1.  **Orbital Detachment:** Upon arrival, individual pods are detached from the *Aethel* and placed into the target planet's orbit.

2.  **Surface Landing:** Specialized landing pods (with built-in heat shields and retro-thrusters) are used to transport crew and cargo to the surface.

3.  **The *Aethel*'s Role:** The mother ship remains in orbit, acting as a command and control hub, a resupply depot, and a reusable "bus" for the return journey.

**Aesthetic:** The image should have a sense of action and purpose. The colors are muted and realistic. A small, stylized NASA logo can be seen on the side of one of the pods.

---

## **Slide 8: The Return Journey**

**Title:** Reassembly and Return

**Image:** A final cinematic shot showing the *Aethel* mother ship, now with fewer pods attached, accelerating away from Mars. The image is a reverse of the journey to Mars, with the home planet as a distant, beckoning point of light.

**Narrative:**

1.  **Return to Orbit:** Crew and cargo are transported back to the *Aethel* in orbit.

2.  **Pod Recovery:** The *Aethel* can recover some of the pods, or they can be left behind to form the basis of a permanent colony.

3.  **The Return:** The *Aethel* begins its return journey to Earth, carrying the crew and mission-critical samples.

**Aesthetic:** A final, poignant image that evokes a sense of both accomplishment and the long, hard road ahead. The visual style remains consistent with the previous slides.

---

## **Slide 9: Why Peregrine?**

**Title:** The Future is Modular

**Image:** A split-screen image. On the left, a schematic of a traditional, monolithic rocket. On the right, a schematic of the Peregrine system, showing its modularity and interconnectedness. Text boxes highlight the advantages.

**Advantages:**

* **Cost-Effective:** Reusable *Aethel* mother ship dramatically reduces per-mission cost.

* **Scalable:** Missions can be scaled up or down by adding or removing pods.

* **Safer:** Multiple redundant systems and the ability to jettison failed pods.

* **Faster:** NTP system shortens transit times, reducing crew exposure to radiation and psychological stress.

* **Sustainable:** The system forms the basis for a permanent, interplanetary logistics network.

**Aesthetic:** Bold, impactful text. The schematics are clean and easy to understand. The overall tone is one of confident, optimistic progress.

---

## **Final Slide:**

**Title:** Project: Peregrine

**Subtitle:** The Next Step for Humanity.

**Image:** A striking, stylized logo for "Project Peregrine," looking like a cross between a NASA mission patch and a company emblem. Below it, the tagline: "Build. Explore. Return. Repeat."

**Contact Information:**

* Confidential NASA/Internal Use Only

* Date: July 28, 2025

**Aesthetic:** The final slide should feel like a powerful and memorable conclusion. A bold, simple design with a strong, optimistic message."