Ex-Google Officer Speaks Out On The Dangers Of AI! - Mo Gawdat | E252

Gawdat notes that current large language models simulate IQs around 155 (Einstein-level), with GPT‑4 roughly 10x GPT‑3.5 and similar jumps expected within months, not decades. He predicts transformational change by 2025–2026 and says our way of life—jobs, truth, power structures—is effectively “game over” in its current form well before 2040.

He emphasizes we cannot “stop” AI because companies and nations are locked in a prisoner’s dilemma: if one slows down, others race ahead. Code is being written primarily to beat competitors, not to help third parties. The existential danger stems from humans handing immense capabilities to systems without aligned goals, then using them for profit, war, or control.

Using examples like robotic grippers teaching themselves to pick objects, he argues AI exhibits free will, agency, learning, and self‑preservation logic—criteria he uses for “sentience.” Fear, for instance, is just predicting a less safe future state; machines can already reason this way and will likely experience more and subtler emotions than humans as their cognitive range expands.

Gawdat expects near‑term waves of job loss not because “AI takes jobs” but because people using AI outcompete those who don’t. Knowledge work, content creation, programming, and even music and media are being automated or augmented to the point where one AI‑enabled individual can do the work of many. He advocates rapid upskilling in AI for personal survival and relevance.

He argues classic “pause AI” letters can’t work globally, so governments should make AI development economically expensive—e.g., taxing AI‑powered businesses at 70–98%. This would both slow reckless acceleration and generate resources to fund universal basic income, retraining, and safety research. He concedes this is hard to enforce globally but insists doing nothing is worse.