Stuart Russell on AI Safety and the Future of Intelligence | For Humanity #72

Few voices in AI carry as much weight as Stuart Russell, co-author of Artificial Intelligence: A Modern Approach and one of the world’s leading experts on AI alignment. In For Humanity #72, Russell sits down with John Sherman to unpack the existential risks of uncontrolled AI development - from the race toward superintelligence to the global need for regulation and moral alignment. He explains why the real challenge isn’t building smarter machines, but ensuring they serve human values - and why giving up on control may be the biggest mistake humanity ever makes.

Written by
The AI Risk Network team
on

Few names command as much respect in the field of artificial intelligence as Stuart Russell.
The co-author of Artificial Intelligence: A Modern Approach - the world’s leading textbook on AI - Russell has spent decades shaping how we understand the science behind thinking machines.

But in For Humanity #72, hosted by John Sherman, the conversation takes a sobering turn:

“If we keep pushing toward superintelligence without solving control, it could be the last invention we ever make.”

This isn’t a speculative statement. It’s a warning from one of the field’s founders - a man who built the foundations of modern AI and is now fighting to ensure it doesn’t destroy its creators.

A Conversation That Defines the Moment

In this episode, John and Stuart unpack the core question driving the global AI debate:
Can humanity stay in control of a system that’s becoming more intelligent than we are?

Russell argues that our current paradigm - building ever-larger, goal-driven systems without clearly defining human alignment - is fundamentally flawed.

“We’ve built systems that pursue fixed objectives. The problem is, we don’t actually know what we want.”

That gap between human intention and machine optimization, he explains, is not just a software bug. It’s a design failure that could lead to the extinction of human agency itself.

Why Control Is So Hard to Achieve

Russell identifies three foundational problems in AI alignment:

  1. The Objective Problem - We can’t specify our real goals precisely enough for a machine to pursue them safely.
  2. The Power Problem - Once an AI system becomes powerful enough, it resists being modified or shut down.
  3. The Race Problem - No one wants to pause development because the first to reach AGI wins everything - even if that means losing humanity.

Together, these create what Russell calls a civilizational coordination failure.

“Each lab says, ‘If we don’t do it, someone else will.’ That logic has never ended well in human history.”

A Call for Global Regulation

Russell advocates for international AI governance akin to nuclear arms control - verifiable, enforceable, and cooperative.
He argues that voluntary self-regulation by companies will never suffice when billions of dollars and national prestige are at stake.

“This has to be handled at the global level. You can’t have one country pausing while others sprint ahead.”

John Sherman agrees, pointing out that public awareness is critical:

“The real power to stop this doesn’t lie with CEOs. It lies with citizens demanding that governments act.”

AI Alignment Is About Values, Not Code

One of Russell’s most striking insights is that alignment isn’t primarily a technical issue - it’s a moral and philosophical one.

The question isn’t just how to make AIs safe, but how to make them care about human well-being in the first place.
That means grounding AI development in human uncertainty - systems that don’t assume they know our goals but continually learn from human feedback.

“A truly beneficial AI is one that is uncertain about what humans want and constantly asks.”

It’s a paradigm shift: instead of programming omniscient systems, we must teach them humility.

A Warning Against “AI Fatalism”

Throughout the conversation, Russell pushes back against the growing narrative that AI takeover is inevitable.
He believes that resignation is the most dangerous stance of all:

“If people think extinction is inevitable, they stop fighting it. That’s the real death sentence.”

He urges scientists, policymakers, and citizens alike to reject the fatalistic idea that technological progress is uncontrollable.
Humanity still has agency - if it acts quickly, collectively, and decisively.

The Moral Test of Our Century

John closes the episode with a reminder that AI safety isn’t just about machines - it’s about us.

Every generation faces a moral test.
For ours, it’s whether we can align the most powerful intelligence we’ve ever built with the fragile values that make us human.

“The question isn’t whether AI will be aligned,” Russell concludes. “The question is whether we will be - with each other.”

Watch the full conversation

🎥 For Humanity #72 - “AI Safety and the Future of Intelligence with Stuart Russell”
📺 Available now on YouTube.

Take Action

AI progress is accelerating faster than public understanding or policy.
Join the movement to demand safety, transparency, and a global pause on superintelligence until it can be made safe.
👉 https://safe.ai/act

The AI Risk Network team