Human Skills for an AI World — a panel at IRAI 2026

Last week I had the great pleasure of joining the panel “Human Skills for an AI World” at the IEEE International Conference on Responsible AI (IRAI 2026), alongside Lily Ballot Jones, Karine O’Donnell, and Behnam Forouhandeh. Special thanks to Shalinka Jayatilleke for organising and chairing the session.

We dug into how sustained AI use affects skill development and retention, where universities fall short in preparing graduates, and what institutions cannot outsource. A few threads stayed with me.

The one I kept returning to: the tasks we offload to AI the most — writing and retrieval — are not obviously the capacities most at risk. Every time I ask for the answer instead of generating it myself, I am bypassing the effortful practice that seems to matter for memory. We do not yet know what this sustained cognitive bypassing does to us, and no study links AI use to long-term decline. But the early signs — from clinicians deskilling once assistance is removed, to shallower learning from a chatbot than from a plain web search — are worth keeping an eye on. My position is deliberately modest: it is a hypothesis to watch, and it almost certainly depends on how and how often we offload, not simply whether we do.

On universities, I argued the gap is not awareness but assessment design. Once a model can write the essay, grading the finished artefact no longer tells us what the student can actually do. We need to assess two things separately — what graduates can do unaided, and how well they can direct, evaluate and correct AI. There is also a line institutions cannot outsource to either a model or an individual: accountability for consequential decisions. Someone still has to be able to say “we stand behind this outcome”, and mean it.

I spoke, too, about where we must be careful — in particular, respecting Indigenous Data Sovereignty, and the dual nature of access. Communities whose languages these tools do not serve are shut out; yet they also face a quieter risk of language loss as people lean on high-resource, Western-centric models whose answers sit closer to Western values than to any other. Asked about the social gaps AI opens — stratification between those who can afford frontier models and those who cannot, and the divide between high- and low-resource languages — I noted it is not all dark: in many developing economies AI is seen as an equaliser rather than a threat.

Asked, finally, what we must protect most as AI grows more capable, my answer was the ability to reach a conclusion unaided — because it is the only thing that lets us verify AI’s answers rather than simply inherit them.