Abstract:
Every public conversation about AI seems to end in one of two places: the machines kill us, or the machines save us. Notice that both stories tell you to do exactly the same thing — nothing. If catastrophe is coming, your choices don't matter. If utopia is coming, they aren't needed. This is a climate scientist's two-year attempt to replace that story with something more useful: a model, built in plain English, of what would have to be true for AI to change everything, and then an honest look at whether or not those things are true, working from “the boundaries inward”.
Inside, you'll learn:
- Exactly what it would take for AI to substantially accelerate economic growth
- The things that arbitrarily advanced AI will never be able to do (and whether they matter)
- A rigorous perspective on how humans continue to contribute economically in a world where knowledge work is automated
- Why the simulation theory is probably wrong
- A perspective on how to live through periods of disorienting change
No part of this essay is arguing that AI will not be transformative, and no part is arguing special insight into what’s going to happen. What this essay aims to do is to lay out in one place, and in plain terms, what we know about how the world works today, and how that knowledge constrains possible futures. The aim is not reassurance, but preparedness: whether the future remains human-centric is up to us.
Citation:
Powis, C. (2026), 'Policy Discussion Paper: Return from Erewhon - What I learned about the future after two years thinking about artificial intelligence', INET Oxford Working Paper Series, No. 2026-17.