Picture two Nobel-worthy scientists standing at opposite ends of a bridge. One waves a red flag: “Stop now, or we risk extinction.” The other shakes his head: “This is preposterous; you’re scaring people away.” The bridge is artificial intelligence, and the gulf is how to think about its future.

NPR recently revived the term “AI doomers” in covering Nate Soares and Eliezer Yudkowsky’s book If Anyone Builds It, Everyone Dies. The phrase alone makes for a viral headline. But beneath the doomsaying lies a profound scientific disagreement. And unlike past hype cycles, the voices are not just futurists in fringe forums — they include Turing Award winners, AI lab founders, and global policy advisors.

The debate is not whether AI matters. It is whether it might destroy us, or whether that claim itself is a dangerous distraction.

The Case for Alarm: Hinton, Bengio, Russell, Carlsmith

Geoffrey Hinton, the “godfather of deep learning,” shocked the field when he left Google in 2023. “If you take the existential risk seriously, as I now do, it might be quite sensible to just stop developing these things any further,” he said.

Yoshua Bengio, another Turing laureate, warns that AI systems may acquire dangerous capabilities—such as strategic planning, persuasion, or autonomous scientific innovation—and in his FAQ on catastrophic risks he argues that unchecked AI could pose threats comparable to nuclear weapons, especially when feedback loops accelerate.

Stuart Russell, a perennial voice of caution, describes the risk of “catastrophic success” — when a mis-specified objective leads an AI to optimize relentlessly in ways that destroy what humans value — often illustrated via the King Midas parable.

Joseph Carlsmith, in his influential report Is Power-Seeking AI an Existential Risk?, lays out a six-premise argument that misaligned agentic AI could seek power and cause human disempowerment. He currently estimates a probability above 10% by 2070 for such an existential catastrophe.

Toby Ord, a philosopher at Oxford’s Future of Humanity Institute, places AI alongside pandemics and nuclear war as one of humanity’s most serious risks. In The Precipice, he estimates about a 1 in 10 chance that unaligned AI could cause existential catastrophe within the next century, and about a 1 in 6 chance when all existential risks are combined. As he frames it, these are not the kinds of odds a prudent society would ever accept.

And so the argument runs: alignment is not optional — it is survival.

The Skeptics: LeCun, Mitchell, Brooks, Marcus

Yann LeCun, Meta’s chief AI scientist and a Turing Award winner, has been one of the most vocal skeptics of the extinction narrative. He has called claims that AI could wipe out humanity “preposterous” and “complete nonsense.”

In a WIRED interview, he cautioned that such alarmism risks “scaring people away” from useful technologies. Instead of pausing development, LeCun argues that progress in open research, coupled with technical safeguards, is the most realistic path to ensuring AI remains beneficial.

Melanie Mitchell emphasizes that today’s systems are brittle and unreliable—she worries more about “machine stupidity” in high-stakes contexts than about runaway superintelligence.

Rodney Brooks has long cautioned against hype; his “seven deadly sins” of AI predictions include anthropomorphizing algorithms and overestimating near-term progress.

And Gary Marcus is skeptical of extinction narratives but warns that AI already enables persuasive falsehoods at scale, threatens elections, and undermines trust—risks demanding governance now.

The Middle Ground: Hassabis and Amodei

Some leaders walk between doom and dismissal. Demis Hassabis, CEO of DeepMind, has argued that AI risks deserve treatment on par with other global challenges. In Time, he criticizes “move fast and break things” as a mode ill-suited to powerful technologies and stresses that oversight and global coordination are necessary.

Dario Amodei, in his Senate testimony, referred to AI as posing “extraordinarily grave threats” to national security. He outlined steps his lab pursues—testing, audits, interpretability research—and frames scaling not as a race, but as one that must navigate careful gating and oversight. The aim is not a freeze, but development with brakes and guardrails.

Why This Debate Matters

Apocalypse narratives have a way of drawing attention. But as Blaise Agüera y Arcas has argued in Noema, they can also mislead: “The illusion of existential risk distracts from the real dangers of present systems.”

And yet, ignoring the alarm entirely is risky. As Holden Karnofsky of Open Philanthropy put it bluntly: “If misaligned AI systems become much more powerful than humans, then AI could defeat all of us combined.”

Somewhere between Hinton’s retreat and LeCun’s confidence lies the hard work of policy, research, and public debate. This is not just a contest of ideas — it is a contest over the trajectory of one of the most powerful technologies humanity has ever built.

Back to NPR: Why Doom Resonates Now

NPR didn’t coin the term “AI doomer,” but it crystallized how mainstream the debate has become. A decade ago, existential risk arguments lived on obscure blogs and at niche philosophy conferences. Today, they headline major news outlets, animate congressional hearings, and surface at dinner tables.

That shift matters. Doom narratives cut through uncertainty because they promise clarity. They distill tangled probabilities into visceral headlines: apocalypse, extinction, survival. Yet, as NPR suggests, the word “doomer” can oversimplify — reducing serious concerns to caricature, and turning scientific disagreements into memes.

This is why the public tension between Geoffrey Hinton and Yann LeCun is so striking. When two Turing Award winners — one urging us to slow down, the other dismissing such fears as “preposterous” — can’t agree on whether AI is humanity’s greatest threat or simply a misunderstood tool, the rest of us are left with both unease and responsibility.

NPR’s framing, then, isn’t just about who is right. It’s about how society interprets the warning bells: as ghost stories to shrug off, or as early system alerts to debug.

For perspective, I’ve linked two complementary explorations of this same debate — one from Diary of a CEO, the other from the BBC’s coverage of the AI 2027 paper — each offering a different lens on why “doom” continues to resonate.

Vocabulary Key

* Alignment: Making sure an AI’s goals match human values and instructions.

* Catastrophic success: When a system perfectly optimizes a flawed goal, with disastrous side effects.

* Power-seeking: An AI’s tendency to gain resources or control to better pursue its programmed objective.

* Superdangerous skills: Abilities like persuasion, strategic planning, and R&D that make misaligned AI especially risky.

* Responsible scaling: A proposed policy of deploying AI in stages, with rigorous evaluation and oversight at each level.

FAQs

Are AI doomers exaggerating? Depends whom you ask. Hinton, Bengio, and Russell say the risks are serious. LeCun, Mitchell, and Brooks say the scenarios are overblown.

What is the core worry? That superintelligent AI, once given imperfect goals, might seek power and sideline humans in the process.

Is superintelligence even possible? No consensus. Optimists believe decades; skeptics think it may never arrive in the form imagined.

Should we pause AI development? Some (Hinton, Yudkowsky) say yes. Others advocate stage-gating, transparency, and oversight rather than a full stop.

What can we do right now? Invest in safety research, regulate misuse, and treat both short-term and long-term risks with seriousness.

Further Reading & Resources

* AI 2027. Daniel Kokotajlo, Scott Alexander, Thomas Larsen, Eli Lifland, Romeo Dean. (The potential timeline of an AI takeover.)

* Becker, Adam. The Useful Idiots of AI Doomsaying. The Atlantic. September 19, 2025.

* Carlsmith, Joseph. “Is Power-Seeking AI an Existential Risk?” arXiv, 2022.

* Ord, Toby. The Precipice: Existential Risk and the Future of Humanity. Oxford University Press, 2020. Podcast version.

* Davis, Ernest. “Ethical Guidelines for a Superintelligence.” Artificial Intelligence, October 23, 2014.

* Agüera y Arcas, Blaise. “The Illusion of AI’s Existential Risk.” Noema, July 18, 2023.

* Bengio, Yoshua. “FAQ on Catastrophic AI Risks.” June 24, 2023.

* Russell, Stuart. Human Compatible: Artificial Intelligence and the Problem of Control. Viking, 2019.

* Center for AI Safety. “AI Extinction Risk Statement.” May 30, 2023.

* Deep Learning with the Wolf. The Singularity: Alan Turing Debates John Hammond (with a little help from Carl Sagan)

#existentialrisk #ethics #superintelligence #responsibleai #aialignment #innovation #deeplearningwiththewolf #dianawolftorres #geoffreyhinton #aidoom #aifuture

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit dianawolftorres.substack.com