The AI Apocalypse: Soft Skills and Hard Thinking

When I was learning to teach Maths, back in the Cretaceous period, one major debate revolved around the use of the then-new technology of hand-held calculators. Would calculators stop students from knowing basic arithmetic? Should we allow them in class? During tests? At home? How would we know what students did by themselves and what they did with calculators? The debate was usually framed between traditionalists and progressives; and there are grains of truth in the opposing positions – ‘calculators will de-skill and make students over-reliant on them, so we need to minimise their use‘ and ‘students will use calculators whenever they want in future at work, so they need to get used to them’.

Over time, a consensus emerged: All students should know basic arithmetic, but not get bogged down in tedious mechanistic operations. Calculator use is very limited in the early years of education, but widespread for older students, for whom sophisticated graphical calculators are not just permitted, but required. Mastering the technology is a significant part of mathematics, enabling a richer curriculum. At the same time, calculator-free assessments are also an important part of assessing student capabilities.

This broad consensus happened because the calculator afforded us the opportunity to change our conception of what was important to teach. I am not talking here about the increasing importance of ‘soft-skills’ like communication and collaboration (important though these are), but of the traditional dimensions of knowledge and skills. When we no longer needed to teach the skills of long division, or trigonometric tables, a re-evaluation of the purpose of mathematics was possible. We reduced focus on mechanical operations to allow more room for problem-solving, creative puzzle solving, and mathematical modelling.  And calculators certainly have not reduced the need for maths skills – you can’t just give someone a calculator and expect them to be a rocket scientist or hedge fund quant.  If anything, calculators have just made maths more accessible to everyone, and to move away from calculators would now feel to me like asking classes in creative writing to chisel their words into wood, rather than write them on paper.  It just misses the point about what those classes should be paying attention to.

The same thing applies to the effect of the internet on schools – though we are still in the midst of the transformation. As with calculators, the internet has forced us to confront fundamental questions about what’s important to learn. As a result, the balance between skills, dispositions and knowledge has shifted, and as a result we are no longer happy with simple measures as adequate measures of attainment. At UWCSEA, our focus on transferable conceptual understandings as a central academic aim is a clear outcome here.

Generative AI (Gen AI) is doing exactly the same thing: forcing us to ask initial questions about practicalities that quickly lead to questions about the purpose of education, and how we allocate the scarce resource of our attention. The difference is that while the transformations engendered by calculators and the internet have evolved over many years, the scope of GenAI, and the fact that it builds on and amplifies the existing internet challenges, means that these challenges are much accelerated.

There is a very profound question about what GenAI will do to education in the long term.  To my mind it’s a bit early to tell, but I want to make one brief observation before addressing the much narrower issue of assessment.  Let’s start with this from my colleague Tim Lovatt:

If you follow the trail of AI into the future, the machines will be able to do more and more cognitive work, better and better. We will be able to use them to replace or augment more and more of this work… There’s no end in sight for this improvement. 

Lovatt goes on to ask what do humans actually end up bringing to the table in 20 years time, when our youngest students enter the job market? And the answer, he claims, is that they bring humanity, context, empathy and emotion. There is a great deal to unpack there, and these soft skills will be the subject of future posts, but I wanted to make that big picture observation here as the most important thing we need to think about, in the long-term.

In the shorter term, however, let’s turn to what has been called a GenAI ‘apocalypse’ for Schools and Colleges.

The Assessment Problem

This video illustrates the worry that schools and Colleges will no longer be able to assess students meaningfully; that students will use GenAI to create what has been called ‘an apocalypse’ for schools. Watch this student use GenAI to do his assignment; it clearly makes the task utterly useless.

As far as authenticity goes, some Higher Education research indicates that 15% of HE students pay someone else to do an assignment, mainly online. And by one reckoning, 20,000 people in Kenya alone earned a living writing essays full-time. So this already is an existing problem that GenAI may greatly exacerbate, and it needs real thought and attention. The danger will be that schools will be forced to go back to supervised exams as the sole means of assessment. But it would be tragic to undo decades of progress, to go back to assessments at a single point in time, that (tautologically) can only focus on things that can can be answered in exams, and that do not allow for iterations, or extended thought, or research. Exams have their place, but they are not the be-all and end-all solution here.

So there is a danger, but there is also an opportunity, because like calculators and the internet, GenAI affords us the chance to reconsider what we are trying to do in education. So the challenge and the opportunity is to ensure we use GenAI to find new and better ways to meet our goals, rather than defensively limit the scope of our ambitions.

The Challenge and the Opportunity

In this section I want to accept that the GenAI genie is out of the bottle and see what it might mean. What might be possible now that was not possible before?

We know that students learn when very specific long term-memories and cognitive structures are laid down. And, as Daniel Willingham memorably put it, memory is the residue of thought. So education is fundamentally about getting students to think about things. And just not any thoughts, but targeted thoughts that take ideas, wrestle with them, search them for meaning, interrogate them, resolve or accept their contradictions, and establish links to other thoughts. Researcher Coe et al call this ‘hard thinking’, and it is, well, hard. Really, really hard. The art and the science of the classroom is all about getting groups of children to do this hard thinking when there are a million easier things to do. Put like that, it’s not a new problem, but it does allow us to reframe the ‘AI problem’ as an opportunity: How can we use GenAI to help students engage in hard thinking, rather than as a substitute for thinking? And then it all comes down to creative and engaging pedagogy, as it so often does.

Ethan and Lilach Mollick summarise their experiments with this approach in what I think will be a seminal paper; Instructors as Innovators: A future-focused approach to new AI learning opportunities, with prompts. They use GenAI to develop personalised learning scenarios. Now, pre-GenAI I was deeply sceptical about this concept, but there is something genuinely personal and different here; the critical point being that GenAI may allow a move from student-as-recipient to student-as-mentor or student-as-coach. I tried a few of the exercises here, and they seemed to me to have some promise. It’s obvious that they would not substitute for warm personal relationships, but that was never the goal; and as long as GenAI sits in a supportive atmosphere where those relationships exist, it has the potential for transformation.

We are trialling GenAI use for teachers in a number of spheres.

• Using tools like Scribo, which coach students on their writing skills, helping them identify their habits and work with them to enhance their written work, without giving them the answers or autocorrecting. This, in turn, frees teachers up to provide feedback on the conceptual, nuanced or contextual elements of the written work.
• Using chatbots as individualised lesson coaches to provide just in time learning experiences for students doing independent work, and AI summaries for teachers of the interactions that have happened. 
• Getting AI to write sample work that the students can then give critique or feedback as part of an assessment.  My colleague Jay Douglass notes that here the impersonal nature of AI is an advantage: “The kids will be way more brutal in their feedback to AI than they will be to each other”
• Using AI as a thought partner to develop project work for assessment (e.g. in our new UWCSEA science course, students use AI research tools to understand the environmental impact of the supply chain for manufacturing electric vehicles (EV)and then use this to inform a local, contextual, human question “should UWCSEA only allow EVs on campus?” – this is still in development!)
• The writing feedback tool in Magic School AI has been used from G3-G5 at East to add an extra layer to the writing process. Students have been motivated to focus on specific goals when writing, knowing that the AI will provide feedback focused on those goals. Teachers emphasise the importance of students fully engaging with the planning and drafting phases before any interaction with AI, which is then used as part of the feedback cycle alongside teacher input. The level of personalised feedback that AI can provide every student within a single lesson is something that would be impossible for one teacher to match.  The students are able to interact with the AI, which has been very carefully designed by the teacher to only speak to the learning goals of the lesson, to further refine their work. Students are developing skills in how to craft effective prompts for the AI, learning that the quality of their input directly influences the quality of feedback and suggestions, which in turn helps them bring out their best ideas and enhance their writing. The level of motivation in these lessons has been remarkable.

So as always, we are using the challenges as a springboard, and should embrace it as a means to enrich our educational practices.  It’s an exciting time to be in education.

With thanks to Tim Lovatt, Angela Newby and Ellie Alchin for the insights.