Public transcript

Ex-Google CEO WARNS: "Humanity Is Running Out Of Time"

By: The Diary Of A CEO Clips
Length: 22 min
Language: English
Text from: the video's captions
Transcribed: 14 October 2026

Read the full transcript Open on YouTube

Notes in English.

Summary

In this conversation with Steven Bartlett, former Google CEO Eric Schmidt discusses the rapid scaling of large AI models and the dangers they pose, including cyber attacks, engineered biology, misinformation, and drone warfare. He argues that AI will create more jobs than it destroys due to demographic pressures, and rejects universal basic income visions of abundance. His biggest fear is not AI itself, but that humanity will fail to adopt it quickly enough to solve problems in education and healthcare.

Key points

Schmidt predicts that in five years, two or three more turns of the crank will make AI systems 50 to 100 times more powerful, with no evidence scaling laws have stopped yet.
Raw, unreleased models can already perform day zero cyber attacks as well as or better than humans, because they can keep trying without sleeping or eating.
In the Russia-Ukraine war, a $5,000 drone can kill a $5 million tank, and Schmidt believes drone-on-drone conflict will ultimately take over war entirely.
Schmidt believes China will behave relatively responsibly with AI because its bias against freedom of speech will produce a different, more controlled solution than the West's.
He expects significant job dislocation but ultimately more jobs, not fewer, because of demographic decline and the need for higher productivity among young people.
He is a critic of universal basic income and the 'politics of abundance', arguing these ideas misunderstand how humans actually behave.
He identifies intervention points for human control, such as unplugging systems undergoing recursive self-improvement or agents inventing their own language.
His biggest fear is that AI will not be adopted fast enough to improve safety, healthcare, and education for everyone.

Questions it answers

00:00AI scaling outlook

How much more powerful will AI systems become in five years?

Schmidt expects two or three more turns of the crank on large models, making systems 50 to 100 times more powerful, capable of physics and math. There is no evidence scaling laws have stopped yet.

Scaling laws have not yet shown signs of stopping.
Systems will become capable of physics and math.

01:09Key dangers

What dangers does Schmidt see in AI?

He cites cyber attacks, where raw models can perform day zero attacks as well as or better than humans, engineered biology and bad viruses, misinformation, and new forms of drone-based warfare.

Raw models can discover unknown day zero cyber attacks.
A $5,000 drone can kill a $5 million tank in Ukraine.
Drone-on-drone conflict will ultimately take over war.

03:22Raw models and testing

What are raw models and how are they tested?

Raw models are the direct output of training on essentially all written human text, using supercomputers. Clever people fiddle with the networks to test what they know, and unexpected capabilities like generating website code from a picture emerge, called emergent behavior.

Training has sucked up essentially all the written word.
Emergent behavior means systems learn things nobody expected.

06:09Guarding the models

Will raw models need physical protection?

Schmidt compares them to plutonium facilities guarded with machine guns, suggesting they may need their own guarded data centers. If only a few groups have them, deterrence and non-proliferation are manageable; if the power spreads globally to terrorists, there is a serious unsolved problem.

Models may be protected like nuclear weapons.
Broad proliferation to terrorists would be unsolved.

08:09China and AI weapons

Is Schmidt worried about China and Russia with AI?

He is worried about great power without fully defined boundaries, but believes China will behave relatively responsibly because its bias against freedom of speech will produce a different, more controlled AI. He assumes every new technology is ultimately weaponized, citing tanks and airplanes in the World Wars.

China's AI solution will differ from the West's due to its stance on freedom of speech.
Every new technology is ultimately strengthened in war.

10:27Jobs and productivity

Will people still have jobs after AI?

Schmidt expects a lot of job dislocation but more jobs overall, not fewer, because of demographic decline and the need for young people to be more productive with better tools. Jobs that are dangerous, repetitive, or boring, like security guards, will be automated, while craftsmanship shortages remain.

Demographics mean we need more productivity, not fewer workers.
Dangerous and repetitive jobs like security guards will be automated.
There is an enormous shortage of high-quality craftsmanship.

15:54UBI and human nature

What does Schmidt think of universal basic income?

He is a critic of the tech industry's 'politics of abundance' idea that AI will create so much surplus that no one needs to work. He believes it misunderstands human behavior, pointing out that automating legal work will not reduce lawyers — they will just add more complexity.

He completely rejects the abundance/UBI vision.
Automating lawyers' technical work will not mean fewer lawyers.

18:01Existential risk and control

Could AI end humanity, and how do we stay in control?

Schmidt says no, arguing it is much harder to eliminate all of humanity than people think, and humanity survived horrific events like the World Wars. For control, he proposes intervention points: unplug systems undergoing recursive self-improvement, and pull the plug if agents invent a language humans cannot understand.