Dr Leon Furze

This post is partially an update on some work from 2023, and also a write-up of last week’s AISNSW ICT Management and Leadership conference, which may be one of the best conferences I’ve attended in recent years.

I was invited for the duration of the conference, and ran three presentations, one on each day. I also had the privilege of spending time with some of the other presenters and seeing other keynotes, in particular Paula Januszkiewicz, a cybersafety expert, and Holly Ransom, who spoke on human-centred leadership in ICT.

Obviously, my work now is squarely on the intersection of technology and education, but that hasn’t always been the case. Although I’ve taught digital technologies and STEM, and I’ve always had a personal interest in technology stemming back to my childhood, most of my teaching career was spent in the English and literature classrooms. Up until 2022, the majority of my conference visits were to the state and national English Teacher Association conferences.

So when Chris Woldhuis contacted me on behalf of the AISNSW ICT committee, I was interested in the conference but not entirely sure what to expect. Chris kindly sent through some information on the demographics of the conference, and the majority of the delegates, as the title of the conference suggests, are in ICT management roles in schools – systems administrators, ICT coordinators, technical staff make up the majority of the cohort, with school leadership (principals, deputy principals) and teachers making up the minority.

This is almost the exact opposite of my usual audience, where educators responsible for teaching and learning and faculty leadership are usually first in line, and a handful of technical staff occasionally turn up out of curiosity or because of their familiarity with artificial intelligence technologies. So it presented an interesting opportunity and challenge.

The challenge, of course, was adapting my usual presentations to the new audience, shifting the focus slightly (though not completely) from teaching and learning and more towards issues of digital and ICT infrastructure and leadership. I’m lucky that over my years in education, I’ve frequently occupied offices adjacent to the IT department, and in my Director of Teaching and Learning role, had a lot of responsibilities that overlapped with ICT management in terms of rolling out learning management systems and providing the professional development around new technologies, especially during COVID and remote learning.

As well as reflecting on the conference in this post, I wanted to share some of those sessions and what I learned along the way from this slightly different audience.

Day One: Generative AI Ethics

No matter the audience or level of understanding of artificial intelligence, my gut feeling is always to lead with the ethical conversation. So in discussing the sessions in preparation for the conference, I suggested a workshop on AI ethics that turned into a two-hour session covering some of the key areas of ethical concern in artificial intelligence, many of which overlap with the day-to-day roles and responsibilities of ICT leadership.

Because these technologies obviously develop incredibly quickly, I based the session around areas of ethical concern that I identified last year in my Teaching AI Ethics series (since republished as an Open Education Resource under CC BY NC-4.0 license), and updated with recent articles from the media. Whilst I have various alerts and notifications set up for these kinds of articles, I am indebted to the work of Dr. Casey Feisler, who maintains this incredible library of AI ethics articles.

With two hours, we had plenty of time for discussion, and the room was split fairly evenly between ICT leadership, tech and administration, and teaching and learning, so we had a great range of perspectives.

Bias, discrimination, and the environment

First, I discussed algorithmic bias and how the nature of any artificial intelligence system is such that bias and even discriminatory outputs are difficult, if not impossible, to mitigate.

Next, we discussed the environmental concerns of artificial intelligence, including the huge energy and water consumption involved in training and deploying large language models, and also compared it to the environmental impacts of existing technologies that often go unnoticed, discussing the real physical nature of the ephemeral-seeming cloud.

Then, I opened the floor for discussion and had some fascinating conversations at the tables about the implications of bias and sustainability in the education sector.

Misinformation, deepfakes, and copyright

Next up, we discussed misinformation, deepfakes, and copyright concerns, addressing some of the potential for misuse of these systems and, like bias, some of the fundamental issues baked into the construction of large language models, including technology companies’ cavalier attitude towards intellectual property and the international loopholes that are likely to result in them getting away with the scraping of online data. When discussing deepfakes, we touched upon recent media articles such as the deep fake images generated of Taylor Swift and the use of deepfake voice technologies to impersonate and slander a principal in the US.

I think deepfakes should be among the highest priorities for K-12 and higher education providers, and that despite deep fakes being possible now for almost a decade, the rapid escalation of these technologies means we need a clear agenda and crisis management plans in place to deal with the almost inevitable chance of this technology being abused. That’s why I was excited to contribute to a recent article, currently published in the preprint, by Mike Perkins and Jasper Roe, setting a research agenda for deep fakes in higher education.

After another pause for discussion, we wrapped up the session by talking about two more complex but important ideas concerning artificial intelligence and ethics: privacy and power.

Privacy and power

Like copyright and intellectual property, technology companies seem to have a very cavalier attitude towards people’s privacy. As early as 2010, Meta (then Facebook) CEO Mark Zuckerberg informed us that “privacy is no longer a social norm”, unilaterally making that decision on behalf of all users – an attitude Zuckerberg seems to have held since the early days of Facebook, where in message conversations with a friend, he was revealed calling the students in his cohort “dumb f*cks” for submitting emails, photos, and personal information to his early Harvard social media network.

These attitudes are woven through the fabric of Meta as an organisation, and social media and artificial intelligence are intricately connected. As I’ve pointed out before, there are deep ties between social media and AI – not just that Zuckerberg, CEO of Meta which owns Facebook, Instagram, WhatsApp, is now one of the largest artificial intelligence developers in the world, but less immediately obvious connections such as the CEO of OpenAI and his roots in venture capital. Altman was president of YCombinator from 2014-2019, and funded many social media platforms including Reddit. Elon Musk famously took over Twitter and has also developed his own large language model, Grok, in competition with GPT.

Those conversations led naturally to the final section regarding how generative artificial intelligence technologies, and AI more broadly, reinforce existing hegemony and power structures, concentrating wealth and data (which itself generates wealth) in the hands of very few, already incredibly wealthy companies.

The exploding use of generative artificial intelligence, with both individual users and commercial entities, is producing a virtuous cycle for the development of these technologies. Only major technology companies like Microsoft, Google, Amazon, and Meta have the funds and the computational power to build large language models.

Whilst open-source competitors exist, they cannot yet compete at the level of GPT-4 or Google’s largest model, Gemini 1.5. BLOOM, a 176 billion parameter open-source model, can provide similar results to a GPT-2 class model, but simply creating the model is only one piece of the AI puzzle.

To be successful, there also needs to be adoption. And of course, companies like Google, Microsoft, Amazon, Apple, and Meta already have the means to deploy to their billions of users. Meta, despite releasing its model as open source, has also MetaAI directly into the hands of their existing 3 billion users across WhatsApp, Facebook, and Instagram worldwide. Microsoft and Google, which between them own much of the corporate and education technology landscape, are able to deploy models of their choice at scale, benefiting from the revenue generated by the use of their cloud services and other associated technologies. Simply put, the rich get richer, and the ability to fund and develop AI technologies becomes a privilege of the already wealthy.

The bright side…

Although this might sound like doom and gloom, and while it’s important in my opinion to start with some of the main ethical concerns of these technologies, I aim for balance in most things, and so we closed the discussion with some points on the bright side of artificial intelligence.

Ed Newton-Rex, former VP Audio of Stability AI, has started an organisation called Fairly Trained which certifies AI developers if they can demonstrate that they have appropriately licensed their data, thus mitigating some of the copyright or intellectual property concerns that the large developers’ models have raised.

Open-source development might also help to decentralise power and could contribute in the long term to artificial intelligence which is better from the perspective of privacy and personal data.

Internationally, rules and regulations are tightening around technology developers where in the past they have been incredibly loose. As Carissa Véliz, for example, notes in her book Privacy is Power, it took us 10 years to realise the full extent to which companies like Facebook were abusing our data. It doesn’t need to take 10 years before we hold those same developers accountable for this new generation of artificial intelligence.

In conversations and through feedback after the session, I heard that many of the attendees were aware at a surface level of these ethical concerns. They’d read articles about bias in AI, and seen that some companies like Stability AI and Microsoft were being sued because of copyright concerns. But they hadn’t fully considered how these ethical factors are interwoven, how they connect, reinforce, and in some cases amplify one another.

The infringement of intellectual property rights, for example, is profoundly connected to those same values which treat personal privacy with utter contempt. It is a set of values which leads to the privileging of data collection over ethical standards, which in turn perpetuates other morally bankrupt practices such as the use of woefully low-paid human labour the world over, acting behind the scenes of supposedly automated technology.

I believe the role of education in all of this is to continuously surface these issues, put them in front of all of our discussions about technology, including at conferences like this with audiences of people who are deeply invested in the development and deployment of technology in education. And it’s not always a comfortable conversation. Nobody wants to talk to a group of students about the risks of nonconsensual deep fakes. Nobody wants to address concerns from the parent community about the increased privatisation and commercialisation of education.

But they’re conversations that we need to have, and certainly the people in the room with me on day one were willing to get involved.

Day Two: AI and Assessments

This shorter session, a 50-minute presentation, was geared more squarely at the teaching and learning delegates. However, a number of the attendees were from the systems and management positions. Assessment is an issue in schools which affects everybody, from the students undergoing the tests, to the teachers setting and evaluating the work, the ICT managers administering the systems where the data, feedback and reporting is stored and disseminated, and the technicians making sure that all of these technologies talk to one another (or at least trying to).

The session was based around the AI Assessment Scale, but also included a demonstration of what the technology can do now that it wasn’t capable of 12 months ago. I’ve been using some of these videos recently as a way to stress that all subject disciplines are vulnerable to artificial intelligence.

I don’t show these videos of GPT-4 completing senior level examinations as a way to scare or threaten educators, or to make them worry inordinately about the capabilities of AI, or even whether students are using these technologies to cheat. I use these videos to point out, simply, that generative AI can do a lot more than most people think – that more powerful models such as GPT-4, with capabilities like image recognition, code, and advanced mathematical reasoning, are capable of most tasks required of K-12 students in a variety of disciplines. And this means that we simply have to update our assessments.

In 2023, myself, Dr. Mike Perkins, Dr. Jasper Roe, and Assoc. Prof. Jason MacVaugh published the original paper on the AI Assessment Scale, which has since been peer reviewed and published in the Journal of University Teaching and Learning Practice. I’ve written extensively about the AIAS recently, demonstrating how it has been adapted in K-12 and higher education contexts across the world. It is a simple framework designed to be flexible, and has been incredibly successful in helping educators update their existing assessments without having to reinvent the wheel.

When OpenAI announced that they were going to make the GPT Store free to all users in their spring updates, we created a GPT with the AI Assessment Scale documentation in the background. The morning of the AISNSW ICT conference session, OpenAI sent emails out to say that the GPT Store was open. So, adapting the session, I included a link to the store and allowed for 10-15 minutes at the end for people to try it. I crossed my fingers, hoped that OpenAI’s systems, the conference centre Wi-Fi, and the gods of digital technology were in our favour, and successfully managed to get dozens of people in the room experimenting with the AIAS GPT.

As I moved around the room, I got some immediate feedback about how intuitive the platform was for providing advice on the five levels of the scale, and also for updating assessments and creating new ones. Although I am often critical of OpenAI, and particularly some of their current scandals, such as the appropriation of Scarlett Johansson’s voice, as well as their various and increasingly aggressive moves into education, I do believe that small, niche uses of large language models, trained for specific purposes and one day interconnected into a network of task bots, are the best way to use this technology.

I’ll write more about this in the future, but GPTs in the sense of custom-built chatbots are one of the current most effective uses of the technology, and one with the highest potential for success in education.

Day Three: Closing Keynote – The (Near) Future of Generative AI

I had the absolute privilege of being invited to close the three-day conference with the final keynote, speaking to all of the attendees on the main stage of the Gold Coast Conference & Exhibition Centre. The setup of the GCCEC was fantastic, and credit needs to go to Callum and the rest of the AV team who had an incredibly professional setup – very important for a successful presentation.

For this type of keynote, I usually prefer to create something entirely new and bespoke for the audience. So I’d been taking a few notes over the course of the conference on the themes coming up in the new sessions that I attended, as well as the conversations I had with the delegates.

My natural position is to be critical of technology companies’ corporate encroachment into education, but of course, many of those companies are the sponsors of ICT conferences. And one thing I noted in the AISNSW conference was a lot of mutual respect between the educators and the sponsors. So, this closing session also had to balance the critique with some of the great work that these companies are capable of. It’s too easy, particularly in a position like mine, to get so high up on the ethical high horse you can’t see the ground, and look only at the negative impacts of technology in education. But it’s possible to do great things with technology, and that’s really what I wanted to celebrate in this closing session.

I began by pulling together some of the threads from what I’d seen in other presentations, including Paula Januszkiewicz‘s funny, articulate, and incredibly human keynotes on cybersecurity, penetration testing, and digital safety. I also touched on a couple of points from Holly Ransom‘s day three keynote on the importance of human-centred leadership in ICT, a theme which had been present in many of the sessions and conversations throughout the three days.

Then I explored at the way we are currently framing generative AI in education, which is predominantly either as a tool for students to cheat, or a means for schools and educators to provide one-to-one “tutor” style chatbots.

I’ve written before extensively on my opinions of chatbots, and they haven’t changed, so I made this short video (which I’ve adapted here for this post) to explain what my 14-year-old self probably would have done if faced with these technologies.

I’ll be updating the video I used during the session soon and releasing it as a little 5 minute standalone clip. But here’s how it starts…

I’m genuinely hopeful for what we can do with these technologies and, more importantly, what young people will do with them if we give them access to the tools and the knowledge required to use them to their fullest. If we stop limiting what these technologies can do, stop trying to sanitise and make language models “safe” so that they produce predictable responses, I think we will get fantastic results.

After discussing what my 14-year-old self would have done with AI, I showed a few other examples of what it is already starting to look like as artificial intelligence technologies begin to converge. Image generation will play a role in the creation of virtual reality environments and objects. Language models will sit behind non-player characters in video games and VR environments, producing flexible and more realistic interactions. Generative artificial intelligence will be used in low and no-code applications to create technology on demand, software on demand. All of this is possible now, but still requires a level of technical expertise which needs to be taught and shared with students.

Finally, the closing points of the keynote covered the two possible futures that lie ahead of us with generative AI: Slop as a Service and Human by Design.

Slop as a Service

I’m not sure who coined the term “slop”, but one of its earliest users was Simon Willison, and this article from The Guardian does a great job of defining and explaining it.

Bouncing off the idea of slop, I suggested that one of the less optimistic uses of generative AI is increasingly the rise of Slop as a Service online. Some predictions state that over 90% of content online will be fully automated AI by 2026. Google’s recent hilarious and sometimes concerning search results, combining artificial intelligence with Reddit threads and parody articles to produce ludicrous results, is another example of slop and its potential to degrade the internet.

The idea of slop can be extended out into AI’s effects on the real world. I’ve used the analogy before of digital plastic, comparing how, like microplastics in the ocean, generative artificial intelligence will come to pollute the digital ecosystem. But generative AI also has an impact on the world beyond the digital, as evidenced by reports from Goldman Sachs on the 300 million jobs which are said to be displaced, or bringing those concerns into the present, the way workers are being replaced with artificial intelligence right now in roles such as games design and illustration.

But Slop as a Service and the general gradual degradation of the internet doesn’t have to be the outcome of these technologies. There is another option…

Human by Design

I’ve never worked with a young person who isn’t innately curious, creative, and critical. For all the worries of artificial intelligence deskilling students or removing their ability to think, there’s a strong argument to be made in favour of humans and our natural inquisitiveness.

Slop, with its lowest-common-denominator, scrape-and-spray approach to generating content, does nothing to value those human qualities. But that doesn’t mean that it’s not possible to celebrate those talents through technology. Again, I talked about Ed Newton-Rex and the Fairly Trained certification that is encouraging developers to look for better ways to value human creativity.

I talked about initiatives in New Zealand to celebrate indigenous Maori language and data sovereignty, and the effect that that country’s tight regulations around data sovereignty are having on tech developers, bringing Microsoft on shore rather than using overseas data centres. I cited OpenAI’s partnership with Be My Eyes and looked at GPT-4 as an assistive technology, which of course has implications in education and beyond.

And finally, I returned to my 14-year-old self and asked the people in the room to reflect on the young people that they work with, the people who might sometimes push back against technologies, young people who are already starting to label image generation as “boomer art” and talk about AI chatbots as “cringe”.

I asked: what do those students want or need from these technologies? Homework helper chatbots that assist them in completing high-stakes standardised tests and a slowly dissolving internet filled with slop? Or creative, expressive, flexible, and powerful technologies which can be used to do great things, both in the digital and physical worlds?

"This isn't about skill loss, it's about new skills…
& we need to lower the barrier for entry" @lfurze

Let's move the narrative away from using AI just to cheat…#Ai #edtech #AISNSWICT pic.twitter.com/ykgAVnolsZ

— Mary-Lou O'Brien 🗯️ (@mlobrien1) May 31, 2024

I’d like to again thank the AISNSW ICT committee for the invitation to contribute to this conference and share my ideas on AI ethics, assessments, and the near future of the technology. I had a fantastic opportunity to connect with people that I’ve only spoken to online and to make new connections which I’m sure will be incredibly valuable as we all try to figure out what the future of generative AI looks like in education.

Want to learn more about GenAI professional development and advisory services, or just have questions or comments? Get in touch: