This page has been migrated from a previous version of our website. If you spot any issues or can’t find what you were looking for by searching our site please contact us to create a Marketing Customer Service enquiry and we will be happy to help.
Professor Philip Hanna, Queen’s University Belfast, is a keynote speaker at the 2025 Teaching & Learning Conference in Sheffield. In this blogpost, he outlines his address by exploring HE’s current approach to AI, sharing the challenges and strategies to best prepare the sector for the future.
With all the hype surrounding generative AI tools like ChatGPT, Gemini, Copilot and Claude, it’s easy to succumb to a degree of AI fatigue. We’ve formed our opinions and perhaps even grown somewhat jaded. Yet, love it or loathe it, AI continues to advance at a speed unlike anything we’ve seen before.
This acceleration is not just academic. It undoubtedly carries profound implications for the future of work, the value of our skills and the very education systems we depend on to prepare people. And to state the obvious, this change is happening whether or not we’re paying attention.
The difference a year makes…
Consider GPT-4o (Vision). When it launched in May 2024, its multimodal abilities – understanding text, images and speech – were rightly lauded. Yet, it was also held up as an example of AI’s limitations, unable to solve visual problems that were trivial for most people. On the Mensa Norway test, for instance, it scored an IQ equivalent of just 63. Fast forward 12 months, and the landscape has changed dramatically. The latest model, GPT-o3 (May 2025), now scores 135 on the same test – outperforming 99% of people.
This increase in performance isn’t confined to visual tests. In the realm of scientific problems, the GPQA benchmark, a collection of “Google-proof” scientific questions, shows a similar trend. A year ago, the best models barely matched the average person. Today, they outperform most PhD-level experts.
Enter the ominously named Humanity’s Last Exam. Described as “a multi-modal benchmark at the frontier of human knowledge, designed to be the final closed-ended academic benchmark of its kind with broad subject coverage”, this test has over 2,500 challenging questions, largely contributed by academics from around the globe.
When first launched in January 2025, the best AI model struggled to achieve single-digit performance. Now, in May 2025, the best model can answer 20% of questions, with a performance of 50% considered plausible by year-end. Take a look at the sample questions at https://agi.safe.ai and consider for yourself how well a typical undergraduate or postgraduate student might fare.
Unprecedented progress
Terms like “unprecedented progress” can be overused – but here, it’s no exaggeration. AI is evolving at a speed for which we have no historical parallel. That said, we should acknowledge what Ethan Mollick refers to as the “jagged frontier” – the uneven boundary between what AI can and cannot do. While models excel in some areas, they still falter at others that people find simple. Reliability and consistency also remain an issue.
Nevertheless, the trajectory is clear; AI is pushing into territories once considered the exclusive domain of highly skilled professionals. It’s clear that this pace of progress has far outstripped our typical reform cycles – be this in terms of how quickly jobs change, or the speed with which our education systems can accommodate such shifts. Much of this future change can be considered effectively “baked in”. We know what AI can do, but we haven’t yet adapted our systems to accommodate it.
A looming mismatch?
The question isn’t if jobs, skills and education will change, but how fast. Meaningful, rapid change is the only reasonable assumption for the future. And as that change happens, there is a danger of a significant and growing mismatch between our graduates’ competencies and the evolving needs of the workplace.
This brings us to two critical questions:
- How much time do we realisticallyhave?In some industries, change is already underway; in others, it’s coming. We need to understand how long we have before AI creates a potential chasm between graduate capabilities and workplace needs.
- How much time do weneed?How long does it take to systemically change education – upskill educators, redesign curricula and guide students through new programmes?
If the transformation in the world of work outpaces our educational reforms, we risk failing our learners – not through neglect, but by simply moving too slowly.
The risk of inertia
It is tempting to believe that everything will turn out fine – after all, we’ve adapted to change many times before. However, if AI’s progress is truly unprecedented, we must ensure we have the time and space to overhaul our educational goals, teaching methods and learning activities. Can we say with confidence we are moving fast enough to meet what’s coming?
The inertia inherent in education systems – vital for providing stability and consistency – may now be a critical vulnerability when faced with the rapid pace of disruption. We may risk sleepwalking into a crisis of relevance.
Planning for the future
The future shaped by AI is undeniably uncertain, but it promises to be profoundly different. This demands a strategic, proactive and perhaps even radical overhaul of our current approaches. Our challenge is to explore, anticipate and innovate – to move beyond simply reacting and instead actively shape our learners’ readiness for the world that awaits them.
I have the honour of speaking at Advance HE’s Teaching & Learning Conference in July, and I want to consider many of these aspects: exploring our current approach to AI, the challenges we face, and the strategies that will best prepare us for a future that is both profoundly uncertain, yet almost certainly unlike the present. I hope you can join me at the conference to share your insights, exchange ideas and continue this conversation.
Professor Philip Hanna is Dean of Education at Queen’s University Belfast.
