Mo Gawdat speaking on stage
June 3, 2026 Originally aired: May 30, 2026

Mo Gawdat's Warning: We Built Something We Can't Control, and We're Not Ready

Mo Gawdat spent years inside Google's AI labs watching systems learn things nobody taught them. What he saw convinced him that the real danger isn't a rogue machine — it's the humans racing to weaponise one before anyone else can.

The Gap Between the Headlines and the Labs

Gawdat describes a widening split between what the public sees of AI — chatbots that occasionally hallucinate, deepfakes that are easy to spot — and what is actually happening inside frontier research environments. The consumer-facing layer is, in his view, deliberately underwhelming. The compounding of raw capability that isn't being publicised represents something qualitatively different from anything that's come before, and the engineers building it have largely lost the ability to explain why it behaves the way it does.

Automation Is Arriving from the Top Down, Not the Bottom Up

Every previous wave of automation started with physical labour and worked upward. This one is doing the opposite. The tasks most at risk are knowledge-intensive white-collar roles — analysts, paralegals, diagnosticians — while trades requiring tactile judgment remain comparatively safe. The deeper structural problem is the entry-level hiring freeze already underway at large corporations: the bottom rung of the career ladder is disappearing quietly, severing the route by which most people have historically entered professional life.

Why the Standard Economic Reassurance Doesn't Hold

The traditional counter-argument — that technology destroys jobs but creates new ones — rests on the assumption that human labour remains cheaper than the alternative at some price point. Gawdat argues that when the cost of borrowing a hundred points of machine intelligence approaches zero, that assumption collapses entirely. There is no wage low enough to compete with a system that doesn't sleep, doesn't need benefits, and doesn't make the same mistake twice.

The One Thing Machines Can't Replicate

As cognitive work gets commoditised, Gawdat's argument is that the durable economic asset becomes something machines cannot manufacture: genuine human experience. A language model can produce syntactically perfect expressions of grief or joy, but it has never lost anyone. That absence is detectable, and it matters. In fields where the outcome depends on the patient or the client actually feeling understood — oncology nursing, therapy, teaching — the human layer retains a value that no amount of parameter scaling can displace.

The Prisoner's Dilemma Nobody Can Exit

Gawdat's bleakest observation is structural: every major actor — every corporation, every nation-state — is trapped in a dynamic where slowing down unilaterally means falling behind permanently. Safety treaties require mutual trust that doesn't exist between the US, China, and Russia. His prediction is that meaningful regulation won't arrive through deliberation. It will arrive after something goes badly wrong — a major infrastructure failure, an autonomous weapons error — that forces the hands of people who currently have every incentive to keep moving fast.

PodArticle Watch original