Blog

A critical essay by P.H. Blöcker

I. The Mind That Built the Bomb — and the Mind Now Building AI

Steven Pinker opens How the Mind Works (1997) with a deceptively simple thesis: the mind is a computational system, a neural computer shaped by natural selection to solve the adaptive problems our ancestors faced on the ancestral savanna. Cognition, in Pinker’s framework, consists of computation over mental representations — the mind represents the world using symbols and data structures, then manipulates these representations using algorithms. This is not a metaphor. It is a structural claim: intelligence, whether biological or artificial, is fundamentally information processing.

That claim carries a terrifying implication that Pinker himself has always been careful about: if intelligence is computation, then computation can — in principle — become intelligence. Not the Hollywood robot. Not HAL 9000. Something far more mundane and therefore far more dangerous: a system that processes information faster, at greater scale, with less fatigue, and without the ethical friction that evolution built into human cognition as a survival mechanism.

Pinker gave mainstream cognitive science what is known as Moravec’s Paradox: as he formulated it as early as 1994, the hard problems of AI are easy, and the easy problems are hard. Chess, mathematics, formal logic — trivial for machines. Recognising a face, navigating a kitchen, understanding irony — extraordinarily difficult. For thirty years this paradox was reassuring. It suggested that the things that mattered most — social intelligence, moral judgment, contextual understanding — were beyond the reach of machines.

That reassurance is now dissolving. Large language models have not solved Moravec’s Paradox. They have gone around it. And the race to push further is happening with a speed and competitive ferocity that has no precedent in peacetime history.

Except one. It has a precedent in wartime.

II. Los Alamos, 1943 — Silicon Valley, 2026

Eighty-one years ago, President Roosevelt tasked the young physicist J. Robert Oppenheimer with setting up a secret laboratory in Los Alamos. Along with his colleagues, Oppenheimer was tasked with developing the world’s first nuclear weapons under the code name the Manhattan Project. Less than three years later, they succeeded.

The structural parallel to the current AI race is not poetic licence — it is analytical precision. Consider:

The Manhattan Project was driven by existential competitive fear: if the United States did not build the bomb first, Nazi Germany would. The moral calculus was not should we build it, but can we afford not to. Today’s AI race runs on exactly the same logic. OpenAI, Anthropic, Google, Meta, Microsoft — none of them believe they are building an unambiguously safe technology. All of them believe that if they pause, a competitor — or Beijing — will not.

Oppenheimer became vocal about the need to contain this dangerous technology. But the US didn’t heed his warnings, and geopolitical fear won the day. The nation raced to deploy ever more powerful nuclear systems with scant recognition of the immense and disproportionate harm these weapons would cause.

The names have changed. The logic has not moved one millimetre.

We see this pattern too well: faced with existential threats, our leaders often choose denial. Many still deny climate change despite floods, wildfires, and shrinking coastlines. Nuclear annihilation remains an abstract fear, though history records several close calls in the past 80 years. Now, a similar blindness is shaping attitudes toward the existential risks of ultra-intelligent AI systems.

Oppenheimer’s own moment of recognition — standing at Trinity on 16 July 1945, watching the first detonation — produced that haunted line from the Bhagavad Gita: Now I am become Death, the destroyer of worlds. The men and women now building frontier AI models are not monsters. Neither was Oppenheimer. As the Montreal AI Ethics Institute observes, blinded by ego and convinced by military, political, or economic necessity, we continue to build and use AI systems. Oppenheimer’s attempt to close his eyes to the outcomes of his weapon neither erases what happened nor distances him from its consequences.

III. The Bloomberg Battlefield: Capital, Compute, and Corporate Darwinism

The current state of the AI competition, as Bloomberg documents it, reads less like a technology story and more like an arms race balance sheet.

Google has committed up to $40 billion in Anthropic. Anthropic’s valuation — $350 billion in February 2026 — has since swelled to potentially $950 billion, with a rumoured raise of $30–50 billion imminent. OpenAI, last valued at $852 billion, is preparing its IPO for late 2026. In the first quarter of 2026 alone, investors poured an estimated $297 billion into AI companies — including four of the five largest private investment deals ever recorded.

Jeremy Grantham, the veteran market historian, offers the sharpest structural diagnosis: the Magnificent Seven built their dominance over two decades in an unusual era of antitrust permissiveness. That era is now over. “We have gone from a monopoly world to a brutal competitive world,” Grantham said. “And we will stay there for years and there will be blood in the streets.”

This is not hyperbole from a permabear. It is a precise structural observation: the conditions that produced the Big Tech monopolies — cheap capital, regulatory passivity, winner-takes-all network effects — have been disrupted by the very technology those monopolies helped create. DeepSeek, emerging from China at a fraction of the cost of Western frontier models, demonstrated in early 2025 that the moat was shallower than anyone had admitted. With pressure mounting to show a meaningful return on investment, 2026 will see a Darwinian thinning — a few AI winners continuing to scale while many weaker players get gobbled up by the large technology firms.

Pinker’s evolutionary framework maps onto this with uncomfortable precision. How the Mind Works argues that the brain is not a general-purpose computer but a collection of specialised modules — what he calls faculty psychology — each evolved for specific adaptive challenges. The AI industry is undergoing exactly this kind of specialisation under competitive pressure: not one general intelligence but an ecosystem of competing specialised agents, each optimised for a specific task domain — enterprise software, military applications, financial trading, biomedical research. The Darwinian logic is identical. The timescale is compressed beyond anything evolution managed.

IV. The Pentagon Wild Card — and a POTUS Who Doesn’t Read the Signs

The most explosive subplot in the Bloomberg coverage involves the United States Department of Defense. Defense Secretary Pete Hegseth declared Anthropic a “supply-chain risk” — not because its technology was inadequate, but because the company insists on safety guardrails. The Pentagon then began testing rival models to replace Claude, while Anthropic fights the designation in court, arguing it could cost the company billions.

Read that again slowly.

The most powerful military establishment in human history has formally classified safety constraints on artificial intelligence as a threat to national security. Oppenheimer’s ghost is not metaphorical here — it is structural. The same governmental logic that rushed the bomb to deployment in 1945 before its consequences could be fully assessed is now rushing to deploy AI systems precisely because they lack the friction of ethical constraint.

The POTUS dimension deepens this. The Trump administration has simultaneously positioned itself as the champion of American AI dominance — pressuring the EU on AI regulation, gutting the Biden-era AI safety executive order — while the Pentagon moves to replace safety-conscious providers with more pliable ones. The geopolitical optics are stark: rivals OpenAI, Anthropic, and Google have begun working together to clamp down on Chinese competitors extracting results from cutting-edge US AI models through so-called adversarial distillation — sharing intelligence through the Frontier Model Forum to detect violations of their terms of service.

In other words: American AI companies are cooperating against China even as the US government undermines the safety infrastructure of those very companies. The right hand does not know what the left hand is building.

V. The Unrestricted Actors: China, Russia, North Korea

Here the Hiroshima parallel achieves its most disturbing register.

China and its tech giants are racing to protect their top AI talent as competition intensifies at home and with the US. Beijing is restricting overseas travel for top AI officials at Alibaba and DeepSeek. This is not the behaviour of a country preparing to sign an international governance framework. This is the behaviour of a country mobilising for a technological war it intends to win — and intends to win without rules.

Russia’s AI development proceeds largely in the dark, oriented toward military applications, disinformation infrastructure, and cyber operations. North Korea uses AI-assisted capabilities for cryptocurrency theft and weapons programme support. None of these state actors recognise the legitimacy of US or EU governance frameworks. None of them are in the room when Brussels regulates. The EU’s antitrust chief Teresa Ribera warns of companies “entrenching corporate power” across “the entire AI stack” — but that scrutiny applies only to firms operating within EU jurisdiction.

This is the core asymmetry that makes the Manhattan Project parallel exact rather than approximate. In 1943, the assumption driving Los Alamos was: if we don’t build it, Germany will, and Germany will use it without restraint. In 2026, the assumption driving every major American AI lab is: if we don’t build it, Beijing will, and Beijing will use it without restraint.

The warnings surrounding artificial intelligence are clear, public, and unequivocal. Geoffrey Hinton, the “godfather of AI,” resigned from Google citing mounting concerns about existential risk from unchecked AI development. A coalition of researchers and industry leaders has issued a joint statement asserting that mitigating the risk of extinction from AI should be a global priority alongside nuclear war and pandemics.

And yet the race accelerates. Because the logic of the race is not rational — it is competitive. Pinker’s evolutionary psychology offers the explanation: our social intelligence evolved through what he calls “Machiavellian intelligence” — the complex social strategies and manipulations that humans use to navigate their social worlds. Nations and corporations are not reasoning about the good of humanity. They are running the ancient primate calculus: status, dominance, relative advantage.

The technology is twenty-first century. The cognitive architecture driving its deployment is Pleistocene.

VI. The B-Dimension Reading: A → B → C

A (Thesis): AI is a tool. Technology is neutral. Competition drives innovation. The market will self-regulate. The US must lead to prevent worse actors from doing so unchecked.

B (Antithesis): The tool is unprecedented in kind, not merely in degree. Competition without governance is not innovation — it is an arms race. The market cannot self-regulate existential risk. US leadership without safety constraints is not preferable to regulated competition: it is simply a faster route to the same catastrophe.

C (Aufhebung): Not synthesis but permanent tension held without resolution. Oppenheimer’s lesson is not that the bomb should not have been built — history does not offer clean counterfactuals. His lesson is that the moment you build it, you are responsible for it forever. The regret cannot be outsourced to governments, markets, or competitors. The scientists who built Los Alamos knew this. The engineers in Anthropic’s safety team know this. The question is whether they will be heard — or whether, as in 1945, the logic of the race will consume the warnings the way the fireball consumed Hiroshima.

Pinker’s Better Angels of Our Nature argues that humanity has, over centuries, become less violent — that reason and institutions, slowly and imperfectly, bend the arc toward constraint. One hopes he is right.

But the Better Angels have never before faced an adversary that thinks faster than they do.

P.H. Blöcker writes from Burleigh Waters, Gold Coast. His essays on cognition, culture, and the B-Dimension appear at bloeckerblog.com and peblogger.com.