Idea for July 2025: jagged frontier

My idea of the month is jagged frontier.

This is the idea that GenAI’s advances are massively uneven. We know it’s really good at some things, like face recognition or translation or writing lyrics in the style of Abba. But it’s terrible at others – like value judgements or even plain common sense.

This piece explains more – and how the jagged frontier means we’re constantly changing our sense of what we humans should do, and what we should delegate to our AI agents.

It follows that GenAI can help productivity in some tasks, but in others (which look to be of similar difficulty) it surprisingly can’t.

I guess all advances actually have this shape. Progress is always faster in some areas than others. What’s interesting here is that word ‘jagged’ – suggesting that GenAI can be spiky and dangerous, like broken glass.

And I’m fascinated by the emotions of our developing relationship with GenAI – something my friend David Archer explores in this fascinating experiment, where he asks ChatGPT to make a portrait of itself.

(The jagged frontier in the image is from the impressive new Norwich University of the Arts building, Bank Plain: staircase designed by Hudson Architects.)