The Week the Body Became Surplus

OpenAI lost four executives in a single week and the market did not flinch. Fidji Simo, the woman charged with converting artificial general intelligence into a sellable product — her official title was CEO of AGI Deployment, a position that reads like governmental science fiction — took medical leave for POTS, a neuroimmune syndrome she has battled for years. Brad Lightcap, the chief operating officer who held the commercial machinery together for three years, was reassigned to "special projects," that corporate euphemism that means the same thing in every language. Kate Rouch, the chief marketing officer, stepped down to undergo breast cancer treatment. Greg Brockman, co-founder and president, assumed interim command. Meanwhile, a capitalization table reconstructed by Forbes revealed that Microsoft had converted thirteen billion dollars into two hundred and twenty-eight billion — a return of 17.6 times — and that current employees hold, on paper, one hundred and thirty-five billion. Sam Altman, the founder, holds zero. The company is valued at eight hundred and fifty-two billion dollars and its executive body bleeds in silence. In Latin America we know this architecture well: the hacienda prospers while the hacendado lies dying.

Ken Smythe, founder of Next Round Capital, offered six hundred million dollars in OpenAI shares to his network of hundreds of institutional investors. He found not a single buyer. At the same time, the platform Hiive logged one point six billion in demand for Anthropic shares, and Next Round's buyers had two billion ready to deploy. A year earlier, those same OpenAI shares would have sold in days. Capital has no loyalty. It has instinct.

In 1791, Jeremy Bentham designed the Panopticon: a circular prison where a single guard could observe every cell without the prisoners knowing whether they were being watched. The genius of the design was not the surveillance but its permanent possibility — the prisoner disciplined himself simply because he could not rule out that someone was looking. Two hundred and thirty-five years later, Hikvision, the largest surveillance camera manufacturer on Earth, installed in Hangzhou's Secondary School Number 11 a system that scans the classroom every thirty seconds, identifies seven facial expressions — neutral, happy, sad, disappointed, angry, scared, surprised — and classifies six behaviors: reading, writing, listening, standing up, raising a hand, lying on the desk. Each student receives an attention score displayed on a screen inside the classroom. Classes compete by concentration level on a screen visible from the hallway. In Wuhan, Guanggu Primary School Number 9 subjected eight hundred students to AI psychological evaluations: fifteen were flagged for "emotional fluctuations," three for "serious psychological problems." China's Ministry of Education published a white paper in May 2025 declaring the inaugural year of "smart education" and set the goal of universalizing AI in all primary and secondary schools by 2030. Beijing already mandates a minimum of eight hours of annual AI instruction as of fall 2025. One hundred and eighty-four pilot schools operate across more than sixty prefectures. Bentham would have recognized the structure. He would only have been surprised that the prisoners were twelve years old.

But there is a reverse side. If in China the child's body becomes data, in Japan the worker's body becomes absence. The Recruit Works Institute calculated in 2023 that by 2040 the country will face a deficit of eleven million workers. The Ministry of Health projects a shortfall of five hundred and seventy thousand eldercare workers. METI needs 3.26 million specialists in robotics and artificial intelligence who do not exist. Sho Yamanaka, a principal at Salesforce Ventures, summarized it with a phrase worth engraving: "The driving force has shifted from simple efficiency to industrial survival." The ministry responded with three hundred and eighty-seven billion yen for physical AI infrastructure and a target of capturing thirty percent of the global market by 2040. Telexistence already operates restocking robots in more than three hundred FamilyMart stores — one thousand beverages daily, ninety-eight percent accuracy, twenty-four hours without rest. FANUC, which controls twenty percent of the global industrial robot market, signed a partnership with NVIDIA for robots that generate their own Python code by voice command. The missing body is replaced. The surplus body is measured. The operation is the same.

At Tohoku University, Professor Hideaki Yamamoto succeeded in making cultured rat cortical neurons generate sine waves, triangular waves, square waves, and chaotic Lorenz attractor trajectories. The paper, published in March in the Proceedings of the National Academy of Sciences, documents the first case in which living biological neurons perform supervised machine learning tasks — an algorithm called FORCE within a reservoir computing framework. The neurons grew in microfluidic devices that controlled their connectivity to prevent excessive synchronization: too much coherence kills computational complexity, a problem that also applies outside the laboratory. Yamamoto said it without ornament: "Living neuronal networks are not only biologically meaningful systems but may also serve as novel computational resources." Resources. The word matters. A rat neuron ceased to be tissue and became infrastructure.