The Uneasy Marriage of AI and Nuclear Power
If advanced AI unnerves people, pairing it with nuclear weapons technology stirs even deeper discomfort. Yet this combination traces back to the dawn of America's atomic program. In fall 1943, physicists Nicholas Metropolis and Richard Feynman at the secretive Los Alamos lab launched a contest: humans versus machines. Human 'computers'—mostly scientists' wives crunching equations on desk calculators—faced off against IBM punch-card systems, the era's pinnacle of computing.
The machines prevailed, outlasting fatigued humans and proving their edge. This marked the start of a symbiotic bond between top-tier computing and nuclear weapons development that persists today. Los Alamos recently teamed with OpenAI, installing ChatGPT on supercomputers handling nuclear test data. It's part of the Genesis Mission, a push to double U.S. scientific output through AI and advanced computing at national labs.
“For the first two days the two teams were neck and neck — the hand-calculators were very good. But it turned out that they tired and couldn’t keep up their fast pace. The punched-card machines didn’t tire, and in the next day or two they forged ahead. Finally everyone had to concede that the new system was an improvement.”
ChatGPT Enters the Classified Realm
Last year, Los Alamos National Lab (LANL) linked with OpenAI to deploy ChatGPT on Venado, among the world's mightiest supercomputers. By August, Venado shifted to a classified network, granting the AI access to nuclear weapons secrets. The Department of Energy, overseeing LANL and 16 other labs, launched the $320 million Genesis Mission to leverage AI for scientific acceleration.
Amid Pentagon-Anthropic tensions and AI's role in conflicts like Iran, military-AI ties draw scrutiny. Less spotlight falls on nuclear lab collaborations under Energy Department watch. Lab visitors find no Skynet panic; researchers stay pragmatic about existential risks while handling the planet's deadliest arms.
“They think we’re building Skynet; that’s not what’s going on here at all.”
From Human Calculators to AI tirelessness
Manhattan Project 'computers' toiled exhaustingly on fragile calculators until IBM machines intervened. Today, AI like ChatGPT offers similar advantages: no need for meals, sleep, or pauses. It tackles problems in novel ways, reshaping research at a premier U.S. institution. Comparisons to nuclear dawn abound, from critics to boosters like OpenAI's Sam Altman, who laments Oppenheimer biopics not inspiring physicists.
The Trump-era executive order likened Genesis to Manhattan's urgency. Yet at ground zero, no such fervor shows. LANL's high-performance computing director Gary Grider showed shelves of magnetic tapes holding nuclear secrets, accessed robotically in a chilled vault.
Evolution of Computing at Los Alamos
- 1940s: Human computers and IBM punch-cards outpace manual calculations.
- 1950s: MANIAC, first purpose-built computer, aids hydrogen bomb and beats human at simplified chess.
- 1976: Cray-1 supercomputer installed, world's fastest then, now eclipsed by smartphones.
- Today: Venado with 3,480 Nvidia superchips delivers 10 exaflops; hosts ChatGPT model weights delivered in locked briefcases.
Simulating Armageddon Without the Bang
Post-1992, U.S. halts live nuclear tests (except North Korea). Over 1,000 prior blasts provide AI training data for virtual simulations on Venado. It stresses weapons under real-world scenarios—disease, drops—without detonation. Grider notes: take a weapon, infect it, explode it 1,000 ways digitally.
This aligns with Oppenheimer's post-Hiroshima stance against more tests, favoring lab methods. Now feasible, these yield dream-level calculations. AI could refine explosives: less reactive, toxin-free, cheaper to make. Webster stresses: weapons deter folly, not enable it.
“The world’s nuclear information is right in there. You’re looking at it.”
Life on the Mesa: Beyond Bombs
Perched on a pine oasis amid desert, 'the Hill' blends remoteness and beauty, Oppenheimer's ideal. Home to 18,000 employees, highest PhDs per capita, it's booming with plutonium pit production in a $1.7 trillion modernization. Expansion plans ignore power-hungry supercomputers' environmental toll.
Weapons dominate budget, but AI aids cancer therapies, neutron science, fusion pursuit—perpetually 20 years away, perhaps not with AI designing heat-proof materials. Post-docs once did grunt work; now AI frees seniors for hypothesis generation, coffee, woods walks.
AI Age Echoes Atomic Origins
Mesa relics—from Trinity site to Little Boy assembly—evoke history amid radiation briefings. Heirs to pioneers mix on Manhattan metaphors: not arms race, but multidisciplinary push. Early nuclear fears mirrored today's doomers; now knowledge spreads openly, unlike scarce uranium.
Unlike past government-led tech, private firms like OpenAI lead; labs adapt commercially born tools. Investments mirror atomic scale: data centers blanketing earth? Critics note industrial waste legacies like Hanford. Los Alamos embodies ingenuity turning theory to reality, fueling progress or peril.
“We don’t talk about it. I don’t think I’ve ever had that conversation.”
