AI conference in Hong Kong

by

in

When I was asked if I wanted to attend the Asia Pacific International School Conference (AISC) in Hong Kong I said yes. I had never attended a single in person external Professional Development Session before and now I had the chance to go to an overseas one. Due to flights timing it was also impossible for me to make it back to Ho Chi Minh before the Winter Holiday. So having my final teaching day of the year move up by three days was also a nice bonus. I was also able to travel with my colleague George who runs NoiseSaigon.com. After the first day, we were both in agreement. The greatest takeaway we had from this conference was a feeling of affirmation. Most of what we saw was practice and ideas we already had. Some of it more refined, or alternatively worded, or slightly differently broken down. We should run a “something” in HCMC when we get back. The second day humbled us again, I felt like a caveman gazing upon astronauts. Nevertheless, that did not change my opinion, that at least in the space of in classroom practice and strategies for teachers we have a lot to offer. Especially for non-specialists.

Note: Everything that follows based on my recollection and notes. When I get access to the slideshows, I intend to return and add corrections. It was also written by a human and the AI feedback disregarded.

Leading on Artificial Intelligence

This started when I looked for Professional Development courses for the year 2023/2024. While I had dabbled with ChatGPT when it came out, even running some small session with my class I chose to go out and find some training on it. After finding an course advertised by AI Quintessential that charged the same price for 3-5 Participants I gathered 4 other members of staff to attend this online course. I highly recommend the course, especially for beginners. Since it is their primary job they have time to go, research, compare and prepare prompts for the ever growing list of AI apps. They also worked around our time, our schedule and our requests. While I learned a lot from that event, such as professionally constructed prompts from designed by university professors, the three members of our group who had used LLMs extensively were left with that same feeling of affirmation.

After this we hosted a morning Professional Development Session in our school. It was split into familiar with AI, hosted by Carlos, then two sessions for unfamiliar for Secondary, by Myself and Lo, and Primary by George and Linh. The sessions were highly positively received with almost 100% approval on the feedback form. We were doing the right thing. Immediately we knew we needed to run a follow up session. Indeed that has not happened yet due to other scheduling issues. The head of secondary tasked me with making more sessions, spaced roughly every 5 weeks for secondary. Simply by virture of being the first to ask I had somehow become the AI Lead for Secondary. I began attending the Toddle Leading with AI, series of talks. I also trialed the Toddle integrated AI for planning. Then the conference leaflet came and I was asked if I wanted to attend.

Day 1 of the Conference

When recalling the talks from the speakers I will ensure any commentary and opinion from myself is in italics.

I somehow slept through my alarm and turned the 5 minute walk with an easy crossing into a mad 17 minute dash across multiple lanes of traffic after misunderstanding signs. I still made it on time.

Artificial Intimacy: Navigating the Heart of AI’s Influence

The highlight of Day 1 was Artificial Intimacy by Jason Prohaska and Tracey Chitty. Their goal was to create a sense of urgency and it certainly did. Top of the agenda when we get back would be to invite parents and other teachers to a session on this topic of health and wellbeing in an AI world. A few key quotes are “Move from awareness to action on simulated relationships” and “We can no longer gently engage with digital citizenship”. They stressed, this is the worst version of AI.

They covered the Harvard longitudinal study, the longest study on happiness found that deep meaningful connection is what matters the most. We discusses on the table what are the key differences between Artificial and Human emotion, Artificial and Human love and Artificial and Human Empathy. As this was a discussion all table had different take always. We questioned what does authenticity mean. In some aspect the face that we cannot control who loves us is meaningful. I brought up that it was a key feature of the movie “Her”, that some AIs hated their users. Will humans who interact heavily with bots learn the proper limits of socialisation. People can choose their feedback, it can reinforce negative behaviour.

This was a work shop that examined our thoughts and reactions to inevitable futures of our life with AI. Neural pruning, artificial girlfriends in REPLICA, nudity apps (AI nudes are legally considered non-consensual porn). They recounted a case where a teacher was found to be a victim of this but still faced the professional consequences despite having no control over the events. I have not been able to find this case.

The workshop kept coming back to questions such as “What caring roles would you delegate to a robot?” “Why does it matter that if a relationship is artificial?”, “Does the human need for connection supersede the need for authenticity?” “Who has the right to end my dream?”

One key to cutting through our apprehension was PARO, a therapeutic robot that has been helping “stimulate patients with Dementia, Alzheimer’s, and other cognition disorders.” It stimulates the caregiver response in the patient without the risk of harm, mess or non compliance that may come from a live puppy, and it never gets tired.

We need to move from awareness to action on simulated relationships
We can no longer gently engage with digital citizenship

Jason Prohaska

An uncomfortable thought I was left with following this sections was that a major premise of the conference futility of banning AI. One question I have is how to we guide and navigate the young children through entire new categories of sexual crimes. With sexting, many still did not fully comprehend that taking photos of themselves was illegal. Will we be forced into a world where the default assumption is that someone somewhere has generated an AI nude of us? Will a new version of “unsolicited dick pics” now involve a person sending AI nudes of themselves with plausible deniability?

Keynote – The Future of Education: Leveraging Artificial Intelligence to Enhance Learning in International Schools

Prof. Rose Luckin‘s provided great insight of how to think about AI. I will definitely try to read the list she recommended, Machine Learning and Human Intelligence (2018), Ethical AI in Education (2021), AI for School Teachers (2023) and Who moved my cheese? (1998). One example was mapping out Numeracy and Literacy being core competencies. Then adding AI Literacy along side the other goals of education, Learning Mastery1 and Knowledge Mastery2. These latter three are developed through traditional subject teaching. Education about AI fell into three parts, What AI tools can do, How to prepare AI for Human Intelligence and How it works. The last one was less that every teacher needs to know exactly how to build and AI but more about how it only generates one word at a time.

Human intelligence is fundamentally different from Artificial Intelligence

Rose LuCkin

She introduced the idea of Digital Tracers. Simple acts such as book marking, highlighting and comments. Digital actions that provide evidence of metacognition, reflection and other deeper competencies, or lack thereof. Eventually AI tutors will be able to create metrics for learning a granular level.

Her list of types of intelligence and defining the types AI are bad at was also extremely helpful. The first three are areas that AI thrives.

  1. Interdisciplinary Academic intelligence
  2. Meta-Knowing intelligence
  3. Social Intelligence
  4. Meta-cognitive intelligence – Awareness of thought process
  5. Meta-subjective intelligence – Awareness of emotional impact
  6. Meta-contextual intelligence – Able to navigate new contexts
  7. Perceived Self-Efficacy intelligence – Knowing when it is not providing good information

A CLEVER approach to facilitate senior secondary students’ argumentative writing by using ChatGPT

I also want to commend Dr. Yang Yin Nicole work on developing CLEVER, a structured framework for argumentative writing with AI for senior students. Using AI recursively to refine each step and asking it for feedback, “Do I have enough evidence?” “How is my structure?” “Are my sources varied enough?” Checkpoints that a normal tutor might used but that research has shown as effective chunks for AI feedback.

  1. Claim
  2. Layout
  3. Evidence
  4. Validity
  5. Elaboration
  6. Refinement

The value in this was the evidence she generated regarding how well it worked with a cohort of 5 Chinese student learning to write in English. Definitely one we will ahve to trial an implement in our course.

During the Q&A one of the audience members recounted how in her university course, students have to submit and are graded on the series of prompts they used to generate the given response.

The rest

The other events were sadly a mixture of cancellations or not very helpful in my opinion. For the sake of positivity I will leave at at that.

Day 2 of the Conference

Day 2 was incredible dense compared to Day 1 as you will see by the sheer length. A lot of this was big picture talks rather than premade tools and techniques to use in in our class. As such, these conversation become heavily layered with very little immediately applicable when compared with the desire to run 2 workshops immediately after Day 1.

Being on top of AI: when an educator meets a software developer

My highlight was Dr. Alexey Popov who go the the heart of the true AI revolution in education. It is not about the generative AI, but about the analytical AI that will work in the background. He also talked about how these tools are built by industries and then education get the leftovers. The AI revolution is a data analysis and analytics revolution. Schools and Humans are already bad at using data but with AI they can finally get much better. 10 years ago there was one paper published on the science of learning in a respectable science journal. We can never keep up with the cycle of reading these papers, compiling the learning, comparing it to our contexts and deciding on what are applicable to our lesson. He argued for a possible ceiling in AI intelligence, and that the recent growth in AI as come from its massive increase in training data. When first released ChatGPT only had data up to 2021, Now we have internet connect AI. Once it has consumed all available knowledge, the avenues for growth are improved processing power. Showed some limitations of AI and how it handles data. Mabel is alive at 9am. Mabel is alive at 3pm. Is Mabel alive at 1pm? The AI will answer that it cannot know and create an argument that such a fact cannot be known, not just by itself but as a fundamental fact of epistemology.

We should not be afraid of AI. if it is good at what I do then I should change what I do. If it is a better teacher than it should teach

Alexey Popov

He came up with the data innovation framework and all the different ways a school could and should use data. (ADD link as soon as you get access to slides). Then he built so much based on that the following part of his talk came as a constant stream of innovations. I will compile some of them into here but the one that inspired me most into its own paragraph. Anti-bias check, the AI monitors teacher interactions and limits their cognitive biases. Now casting – Far more granular prediction of results, trajectories and reaction to strategies. Trainable chatbots – with questions built from teacher interactions with the students. Real time concerns and interventions – Academic or social struggles, identified but the teacher or by AI are automatically shared with those best placed to support and intervene.

Natural Language Processing for data analytics and to build a driven student dashboard – What if we could input all of the student written reflection to extract information about the cognitive development of the child. Noticing how the student talk about their class experience it could metrics and track the development of meta-cognitive skills. Students who talk about their struggle with content, vs struggle with task. How they talk about how they overcame difficulties. Student who identify specific strategies vs talking about working harder. Feed all this data into a dashboard that can then help teachers see and identify strength and weaknesses. This could also add in teacher comments. It could include the entire history of reports comments from primary to final years. If the comments were also AI assisted the AI, as long as it was done in the same platform the AI could also know and extract the data that created the comment. Crunching data, from the way humans create data about their thoughts and observations.

Using Learning Analytics to Reimagine Assessment in the Age of Artificial Intelligence

Prof. Dragan Gašević also gave a whole talk about learning analytics to reimagine assessment. Same idea as digital tracers. His focus was “closing the feedback loop”. Before going to far into his work, a key point was how to prevent all of this from becoming an oppressive surveillance tool. Alexey Popov echoed the same sentiments. I recall in one of the Toddle AI talks, John Hattie recounted a pilot programme where AI was used to track teacher interactions on a per student basis. Parents immediately demanded the data to see how many minutes the teacher has spent with their child3. The other key point that he wanted to drive home was Automated analysis to detect AI is flawed 40% false rate with a bias against EAL students. The Keynote speaker of the day had also earlier said that these flaws were public domain, and that institutions should worry about potential legal liabilities should they use detection tools in a way that negatively impacts a students future.

Prof Dragan Gašević looked at what other data sources should and can we now leverage to establish and enhance feedback loops. There were three key area, Analytics for Assessment – how we can measure learning, Analytics of Assessment – how we use this to judge progress, strength, weaknesses and next steps, and Validity and Reliability – how we can ensure trust of the metrics.

OnerousFeasible
DiscreteContinuous
UniformAdaptive
InauthenticAuthentic
AntiquatedModern
Comparing Learning Analytics between AI driven and traditional systems

I feel that this has been the promise of many online learning platforms, from duolingo, to education perfect and myMaths. With the ability to totally automate the these lessons what is the purpose of teacher. There are countless ways to teach oneself how to code, real human teachers still remain popular, even if it is via a zoom call.

He goes on to stress Humans need to remain part of the feedback loop. Key quotes from him are “Analytics should not sideline professional expertise” and it “Does not blackbox accountability”. Not only as teachers who use the tools but also us as a profession are responsible to address any inadvertent effect of these tools.

He found that there was good reliability for the automated scoring of rhetorical structure and that these could be graphed. Not only could his tools classify structures used by students, it could measure how well they would integrates the various elements together into a cohesive argument, and it could be visualized4. He demonstrated the FLoRAProject.org – Self-Regulated Learning with Personalized Scaffolds. My notes on this section is sparse as he visualized the students’ progress in multiple graphs. The end results was better performance for the lower attaining student. During the Q&A I asked if he say this increasing or decreasing the performance gap between the top and bottom performing student. He said that they saw the “Expertise Reversal Effect“. That the intervention actually hindered the performance of the best. Later on in the round table with Rose Lucking he talked about an increase of the gap, but this was between students who learned how to use AI effectively and those who did not.

The teams he heads has been involved in have produced far too much research to cover in detail and so he presented only a few. Three I had in my notes are Epistemic Games – Chats with real and artificial agents. Multimodal sensory tech – Bring simulations into physical spaces. Feedback Comics – Generated comic strips that provided visual, humorous and personalized summary of performances, feedback and illustration for areas to improve

Keynote – The Challenges and Opportunities of Generative AI for Learning and Assessment

Asoc. Prof. Jason Lodge was quick to point out that the fastest thing on earth appears to be the speed at which people became “experts” in ChatGPT. He showed samples of outsiders from the space talking about “learning styles”, ideas about education that we debunked over 14 years out of date.

Good assessment design is good assessment design

Asoc. Prof. Jason Lodge

A lot of his talk was based on the integrity, but with a shift away from an obsession with student integrity but of educational. Of the experiences in class, the assignments and tasks we set. Cheating is only a small, albeit important part. It only takes a few cheaters to undermine trust in a system. There will be an element of Police → Catch → Punish but it is not the focus. He was very clear that electronic detection was not valid. These flaws were public domain, and that institutions should worry about potential legal liabilities should they use detection tools in a way that negatively impacts a students future.

He posits that assessments have always been broken. The fundamental issue with assessment is that we assess artifacts of knowledge, exams, essays and interviews, because we cannot assess the students learning directly. Dylan William also used the same language with the added point that parents, siblings, and tutors were always able to interfere with many these assessments. He called AI democritising cheating. Well designed assessment will be able to deal with a lot of this anyway.

It was once again gratifying to see how closely of this breakdown of the possible responses matched my breakdown5. Advance, avoid, adapt and authenticate. This also had a time axis and “this is the worse version of AI we will ever have.” If we design an AI around a perceived weakness of AI, we should be ready for the fact that a lot of money is being spent to close such gaps. The AI resistant task of today will become a single prompt in the future.

He talked about how we need look beyond AI literacy or improving critical thinking. We have to understand not only how students learn, but how they learn with and from AI. These practices will evolve as the AI systems we have evolve. Teaching practices and educational policies need to be adaptive to accommodate the evolving interactions. The self-regulations learning that AI will bring to dominance needs the support of co-regulated learning. That there has not been enough time to integrate the research into practice. We needs to make observations, build and read theory, analyse this against the mechanisms of learning. Reconcile these aspect and then take action. They proposed 5 points

Assessment should emphasis…

  • appropriate, authentic engagement with AI.
  • a systemic approach to program assessment aligned with discipline/qualifications
  • the process of learning
  • opportunities for students to work appropriately with each other and AI
  • security at meaningful points across a program to inform decision about progression an completion

Assessment and Learning Experience should equip students to ethically and actively engage in a society where AI is ubiquitous

Forming trustworthy judgement about student leaning in a time of Ai requires multiple, inclusive and contextualized approaches to assessment

asoc. Prof. Jason Lodge

The last this I will share from him is this conceptual of how AI affects us. People rarely take the effort to remember phone numbers, or plot a route on a map but the AI can give me a checklist on how to build a good portfolio and guide me through every step. It can answer every question about cancer and therapy techniques but would you want a human guide through the process or a screen. AI should be used to build better relationships with out students. The things we should offload are the things that get in the way of building those relationships.

Living in a world of AI

Ross Parker’s talk came at an odd time in the sequence. His section will be short because it made such a great intro but came as one of the later session. He is the author of a book called Screens that eat children.

Teachers always mediate the output of AI tools they use
Students are invited to consider the perils of inexpert use

Ross Parker

The premise of this talk was that we are at the beginning of a new epoch, and that we need to think about how to navigate this new world. There will be several millionfold improvement over the next 20 years. Our future might require us to have a relationship with AI, the same way we now need smartphones. Kissinger wrote in The Age of AI about how we will have far less patience with regular humans

People will come to prefer digital assistants over humans for humans will be less intuitive of their preferences, more disagreeble

Henry Kissing – The Age of AI: And Our Human Future

Ross Parker reminded us of an incidence where the snapchat AI told a researcher posing as a 13 year old girl how to make her first time with and 18 years older “boyfriend” special and drafted a fake letter to her parents about a fake school trip.

His school created two guidelines around the use of AI. 1. Teachers always mediate the output of AI tools they use. 2. Students are invited to consider the perils of inexpert use. Then he went on the explore what it means to be an expert and to vet the output. 20 year veteran teachers vs fresh NQTs who are loss for ideas. Will students develop expertise as we offload so much onto the AI through the educational journey. These skills were built up incrementally, do we now face a loss of experience. What if quality improves to the point where we do not need to vet or it only emerges rarely so that we miss it. How do we decide on the minimal amount of AI use without getting out of touch. To avoiding tech can increase our connection with being human, but we sacrifice our economic viability

We must be careful with the Humanity of Children

Ross Parker

The round tables

There were two round table discussions. Due to the nature of a round table there was lots of unfocused discussion but I want to draw on a few great moments.

The Future of AI in Education – Rose Luckin and Dragan Gašević, Moderated by Nancy Law

DG – Think about where we do an do not want AI. Prefers AI gates at border controls, want humans in school as it is a place of growth and nurtehr

RL – Must move beyond ethics into answering the moral question of “What must we Maintain?” Moral imperative

Question – What is the time lime for the full epistemic, ontological, contents, skills and concept, and curriculum change?

RL – Yesterday was when it should be done. We knew it was coming. AI learns but doesn’t understand so we must focus on understanding. Flaws were predicted 20 years ago

DG – I don’t want adaptive learning systems, we want adaptive learners. How to we find and create that. Policy and frameworks are slow to follow

Audience – Are we early adopters of AI or just he first respondents.

A – One school dropped 1 IGCSE for a futureproofing programme. Students love it.

A – We are still driven by what is assessed.

A – What is the technical Debt of rushed AI implementation?

NL – It is a network problem. Ai will sit alongside the current methods, Same way new transport fists along side other transit. (My thoughts – Or books and plays exist along side movies and the internet)

A – We need a compass even if we don’t have a map or where to go

DG – Avoid creating a “Sophisticated shell of nothing”

NL – We cannot see ourselves at the receiving end

A – This is the student’s second mess. Their conception of the world is build on uncertainty is normal

RL – We still haven’t talked about parents and their place in this conversation

A – This is a tool and our relationship with tool shape our relationship with the task

DG – While my students had improved learning the outsourced the locus of control when they couldn’t do a task. They blamed the AI, or blamed the tools. We need to make sure they keep a sense of accountability.

NL – teachers are designers. We should not think of assessments as something done to the students but embedded in the loop between the design and outcome of learning.

DG – We cannot just think about generative AI, but also data science, design of learning and pedagogical knowledge

RL – In parallel to the development and improvement of Ai will be and increase in out knowledge of learning

A – What can we do before exam boards change? especially since we already know we probably wont be happy with their solutions

A – Parents have seen AI in their jobs. We should co-construct with parents. My thoughts have as series of parents as guest speakers who talk about industry impacts

DG – Increase of Productivity does not remove workload.

RL – There will be more different work and jobs, such as DATA trustees

Unleashing the Power of AI: Transforming Education at Scale – Helen Meng and Jason Lodge, Modetrated by Kong Siu Cheng

JL – My organization are working to create tools that can work for 9000 schools. Heavy emphasis on privacy, equity and set a foundation for all levels. They avoid being specific on Teaching learning and assessment. They want to tools to power the teacher rather than replace them. 50% haven’t use AI due to lack of awarness. they want to make it easy by collecting and presenting case studies. Technology is progressing faster that the research. Unclear what aspect of GenAI will interface with classrooms. Always harder to do things at Scale

HM – My university offered first AI major in 2019 but then also co created a course for secondary schools for use in Hong Kong School with 5 major themes, Awareness, knowledge, interactions, ethics and empowerment.

JL – We need to build system that protect data rights. System with one way data transfer. His org has a private version of Bing

KSC – My university keeps for 3 months for academic integrity

HM – Tell my students to never upload the thesis, you are just giving it away. Do not do meeting minutes. Also make sure students can always to the task without the tool. Similar to back of the envelope calculations. If you cannot do that do not use the AI

JL – In certain places it doe not makes sense to use AI, such as physiotherapy where it can help but not carry out the practice

HM – For language learning, detect types of mispronunciation. AI can create a reference voice for the speaker in their own voice.

JL – It needs to take away drudgery. Tasks that do not add to relationship. AI should facilitate between student and teache

HM – We are building an ecosystem in Honk Kong. Open curriculums and hope schools can share experiences

KSC – When you upskilled teachers, 1 day workshops will last months. Bring in parents and run workshops with them

HM – Currently bad at symbolic reasoning. it cannot answer how many UN secretary generals were not from Europe. Students need to learn to paly with it. Fact check it. 40% of a doctors time is on documenting their interactions, can we shift the attention back to the patient, or in schools’ case, back to the students.

JL – Language is an attempt to transfer information between individuals. AI has hacked the space between that. We infer that there is something like a mind when we see AI communication because we are used to seeing those things produced by minds. AI has never had access to your though. Only the fraction we can translate into artifacts

The convention

The convention floor was an interesting space to tour. There were another set of talks taking place but mostly in Cantonese. It was a strange experience seeing all the money being invested into EdTech innovations and be left wondering where is the innovations. Countless companies vying to be the one to finally crack the robotics kit market, or online math learning. Drawn into conversation with another assignment distribution platform. That is not to say there is no room for innovation. Someone out there has truly figured it out. Toddle has done some great stuff but it’s true utility was setting up forms that matched the IB’s planning forms in a friendly interface. Managebac already did a lot of that. I personally make sure my students work in office suites or real environments because the skills are lacking. They may one day need to know how to manage word’s tab stops and picture anchors. I am not just teaching Physics/Science but also office skills.

Even worse was is many companies seem to forget that STEM include Science (and Maths), not just Technology and Engineering. We are currently going through an extensive rebuild, redesign and reimagining of our science laboratories and the only interesting addition I could find was some hydroponic systems. There were six vendors for this. Either that or augmented reality dinosaurs on your desk. Seeing a VR planet does not help me learn fractal distillation. Where are the interesting student focused lab equipment? All in one desks with built in power packs and multimeters. Who is going to be the first to build school scale scanning electron microscope? Used can cost $2,500. New can cost $100,000. I did finally find a school that had built internet of things science labs but I will need to follow that up by emails as it was not made as a readily available resource.

  1. Knowing how to study and gain knowledge ↩︎
  2. The sum of factual knowledge that we carry with us and allows us to verify what we read ↩︎
  3. Do we have a greater duty to protect the teacher or to the serve the students or the parents who pay a lot of money to private schools? All three categories are flawed humans that are doing what they can. ↩︎
  4. Add graphs when available ↩︎
  5. My attempt to make this digestible for in school training.
    Authenticate – Citations​, Interviews​, Exams​
    Adapt – Design scenarios that the AI will struggle to answer. This is what future jobs will need​
    Avoid – Use non written tasks such as physical performances and handmade artifacts​
    Advance
    AI is the assignment. We want to see them learn how to use. ​ ↩︎

Leave a Reply