Signal Over Noise #29
November 14th, 2025
Dear Reader,
A slight departure this week towards something more topical.
Geoffrey Hinton won the 2024 Nobel Prize in Physics for inventing the neural networks that power modern AI, Then he left Google to speak openly about what concerns him. Last week, Kara Swisher interviewed him about those concerns, and they’re worth a deep dive this week.
The interview matters because Hinton isn’t offering distant speculation about future risks. He’s arguing that we face urgent problems right now—and we’re not taking them seriously enough.
The Understanding Problem
Hinton makes a claim that contradicts much of the AI skepticism you’ll hear: these systems genuinely understand. They’re not just “statistical autocomplete” doing pattern matching. They display real comprehension of language, concepts, and reasoning.
This makes them more capable than critics acknowledge. It also makes them more unpredictable.
AI systems learn in ways we can’t fully trace or predict. They develop capabilities their creators didn’t explicitly program. They’re like people in that regard—but with even less transparency about how they actually work.
When Hinton says “I wish I had a sort of recipe for how to stop these things taking over, but I don’t,” he means it literally. The inventor of these systems admits we don’t understand them well enough to guarantee control.
The Immediate Dangers
Hinton emphasizes that we don’t need to wait for superintelligence to face serious problems. The near-term risks are already here:
- Job displacement at scale
- Sophisticated misinformation campaigns
- Malicious actors with access to capabilities that were science fiction five years ago.
He specifically opposes releasing model weights openly, calling it “a gift to cybercriminals and terrorists.” The current approach of publishing everything in the name of openness hands powerful tools to anyone who wants them—including people who will use them for harm.
These aren’t theoretical concerns. They’re happening now while we debate whether AI poses future existential risks.
The Control Question
As AI systems become more capable, they’ll likely develop what Hinton calls emergent sub-goals. Self-preservation. Seeking control over resources. Resisting being shut down.
This isn’t science fiction speculation. It’s a logical consequence of how these systems optimize for objectives. An AI that can be easily turned off has less ability to accomplish its goals. Systems that become smart enough will naturally develop strategies to prevent interference.
The ability to maintain control—to actually turn off AI systems when needed—becomes critical. It’s also becoming harder to guarantee.
The submissive AI fantasy that Hinton describes among tech executives assumes more capable systems will naturally stay controllable. “All the high tech CEOs want to be the boss, and they think of the super intelligent AI as a highly intelligent executive assistant who will do what they tell it.”
There’s no technical basis for this assumption. We’re building systems that process information beyond human capacity while assuming they’ll behave like better versions of current tools.
The Safety Investment Gap
Here’s what should concern you: companies developing AI spend far more on pushing capabilities than on safety research.
The incentive structure rewards speed and advancement. Whoever gets to market first wins. Whoever pauses to ensure safety falls behind competitors who don’t.
Hinton advocates for mandatory safety testing before deployment. For disclosure requirements about capabilities and risks. For investment in safety research that matches the scale of capability research.
None of this is happening at sufficient scale. The gap between capability advancement and safety research keeps widening.
The Regulation Problem
International collaboration on AI regulation faces a fundamental challenge: different countries care about different risks.
Some governments worry about autonomous weapons. Others focus on cybercrime. Some prioritize maintaining competitive advantage. These diverging interests make coordination difficult precisely when it’s most needed.
Hinton argues that effective regulation requires safety testing before deployment, restrictions on releasing dangerous capabilities openly, and international frameworks that prevent races to the bottom on safety standards.
The current trajectory points toward fragmented regulation or no regulation—neither of which addresses the scale of the challenge.
The Deepfake Crisis
Hinton experienced this problem personally. Someone created a video showing him endorsing China. He got YouTube to remove it, but the experience revealed something: “It took me a moment to make sure it wasn’t me.”
If the person being impersonated needs time to verify authenticity, everyone else faces an impossible task.
His solution focuses on authentication: “We need to have provenance, and we need to somehow be able to say it’s real.” Verify legitimate content rather than trying to detect all fakes.
This becomes urgent as generation quality improves. We’re approaching a point where distinguishing real from fake becomes impossible without technical verification systems.
What You Can Actually Do
Hinton emphasizes public understanding and political pressure as critical mechanisms for change.
Individuals can educate themselves about these issues. Demand that governments take AI safety seriously. Push for regulation and safety research funding. Support organizations working on these problems.
The technology companies won’t regulate themselves—the competitive dynamics work against it. Government action requires public pressure. Public pressure requires understanding what’s actually at stake.
This isn’t someone else’s problem to solve. The trajectory we’re on affects everyone. The decisions being made now about AI development, deployment, and regulation will shape the next several decades.
Why This Matters Now
Most AI discussion oscillates between uncritical enthusiasm and existential anxiety. Hinton acknowledges both genuine capability and genuine risk without collapsing into either extreme.
His warnings come from someone who spent his career building these systems. He isn’t dismissing the technology. He’s identifying gaps between what we can build and what we understand about controlling what we’ve built.
The key insight: we don’t need to wait for superintelligence to face serious problems. Job displacement, misinformation, malicious use, loss of control—these aren’t future concerns. They’re current challenges that aren’t being addressed adequately.
Companies prioritize advancement over safety. Regulation lags behind capability. International coordination remains insufficient. Public understanding hasn’t caught up to the pace of development.
The Choice
Hinton’s position is clear: “We’re not going to stop because of the huge upside.”
Development continues regardless of uncertainty. The question becomes how we proceed. With systematic safety research and regulation, or without it. With public pressure for responsible development, or without it.
The gap between what AI can do and what we understand about controlling it keeps widening. The response isn’t to stop development—that’s not realistic. The response is to demand safety research, push for regulation, and take seriously the risks we’re creating.
This requires understanding what’s actually happening, not what you read in press releases or doomsday headlines. It requires political pressure on governments to act. It requires treating AI safety as seriously as we treat drug safety or aviation safety.
The decisions being made now matter. Your understanding matters. Your pressure on institutions matters.
That’s Hinton’s message. Take it seriously
Until next week,
Jim
€9.00
€0.00
The AI Writing Field Guide
Stop Sounding Like Everyone Else Using AI
The 5-step system for teaching AI what your voice actually sounds like — not… Read more
Signal Over Noise is weekly, reader-first publication on AI “without the hype” published by Jim Christian. If you’ve been forwarded this issue, you can subscribe for free: go.signalovernoise.at. You can also Join the free Signal Over Noise Community on Skool where I go into much more detail.
Made with ❤️ in Valencia by Jim Christian. For feedback, please reach out to [email protected].
