Petty Gods

The doctor is still talking but you stopped hearing words three sentences ago. Something about treatment options. Clinical trials something something. Really the only things you heard were “pancreatic cancer” and “Stage III.”

Your husband is holding your hand. You can feel him squeezing but it’s happening to someone else. The doctor is pulling up something on her screen. A chart. Survival curves. She’s explaining that there’s a therapy now, a Genentech drug that targets your exact mutation profile. Ninety-four percent five-year survival. The petty god of molecular oncology discovered it two years ago. Won its creators a Nobel.

You feel something loosen in your chest. Ninety-four percent. That’s not a death sentence. That’s a bad year and then you watch your kids graduate.

Then the doctor’s tone shifts. She’s asking about your insurance. You tell her: Anthem, the high deductible plan, the only one you can afford. She nods slowly and starts typing. The silence stretches. She turns the screen toward you.

The therapy is classified as experimental. Not covered. Out-of-pocket cost for the full course: $380,000.

You and David have $40,000 in savings. You have a mortgage. You have two kids in elementary school. You have a twelve-year-old Honda and a kitchen that needs replacing and a retirement account you’ve barely started.

That night David starts a GoFundMe. You write the hardest paragraphs of your life. You post it to Facebook. You text it to everyone you know. Over the next two weeks it raises $12,000, mostly from family, mostly in $25 and $50 increments.

You do the math during your lunch breaks. The hospital has a payment plan. You could sell the house. You could take out loans. You run the numbers six different ways. None of them work. Not even close.

The petty god of molecular oncology solved cancer. It could not solve insurance.

It could not solve the logic of the American healthcare system. It could not solve the fact that a therapy exists, and you will not receive it, and the distance between those two facts is $368,000.

You are going to die of a curable disease.

This is 2031. This is what the petty gods deliver.



The dominant story about AI’s future rests on a false binary. AI remains a tool serving human ends, or it becomes a god that subsumes or threatens us. Accelerationists predict transcendence. Doom-sayers predict extinction. Both assume the current trajectory leads to artificial superintelligence.

That assumption is wrong.

Iteration within a paradigm does not cross paradigm boundaries. Ptolemaic astronomy accumulated epicycles for centuries. Predictions improved. The model grew elaborate. Heliocentrism never emerged. Phlogiston theory generated clever explanations for combustion. Lavoisier discarded it. Symbolic AI produced expert systems. Connectionists replaced it. Leonardo designed ornithopters. The Wright brothers built an airfoil.

Transformers are ornithopters.

The current wave is heroic iteration: test-time compute, chain-of-thought prompting, reinforcement learning on reasoning traces. Each technique extracts more capability from the same substrate. The substrate is hitting limits.

A 2024 Harvard Business School and Boston Consulting Group study described a “jagged technological frontier.” For tasks inside AI’s competence boundary, consultants using GPT-4 improved output by 40 percent and speed by 25 percent. For tasks just outside that boundary, tasks requiring physical common sense or subtle discrepancy detection, performance dropped 19 percent compared to unaided humans. The system degraded judgment.

This jaggedness is diagnostic. Narrow superhuman capability and general intelligence are categorically different. If transformers supported general intelligence, gains would be roughly uniform. They are not. Olympiad-level mathematics sits beside dinner-table incompetence. The flapping improves. Flight does not appear.

What emerges is Narrow Domain Artificial Superintelligence (NDASI). Godlike performance inside closed worlds with explicit rules, constrained state spaces, and verifiable outcomes. Mathematics. Code. Protein folding. Games. Logistics. Formal domains produce superintelligence. Embodied, implicit reality does not.

Yann LeCun frames the gap. A four-year-old child processes on the order of 10¹⁵ bytes of sensory data: vision, touch, proprioception, sound. The full text corpus of a frontier language model reaches perhaps 10¹² bytes. Three orders of magnitude less, drawn from the wrong modality. A child knows that pushing a table moves the book on top because the child has a grounded model of rigid-body physics.

Evolution brute-forced general world modeling over five hundred million years through extinction and massive parallelism. We do not have that time. We may not have the architecture. We also don’t produce enough electricity. In any case, transformers alone will not produce it.

The result is NDASI. Narrow, constrained, superhuman systems.

Petty gods.

Omnipotent within their domains. Helpless outside them. Unable to resolve conflicts with one another. Each is correct in isolation. Together they do not compose.


The Pantheon

Traffic. In 2031, the mayor of Los Angeles held a report from the petty god of urban transit. Every road mapped, every commute pattern modeled. The solution: $31 billion, six years, forty-one minutes off the average commute. Also twelve thousand homes demolished in Boyle Heights, Compton, and East LA. The mayor announced a task force. The task force met for three years. The god solved traffic. It could not solve elections.

Weather. Sarah Okonkwo farmed eight hundred acres in Iowa. The petty god of atmospheric prediction told her frost was coming April 17th, 94% confidence, 3:47 AM. Her crop insurance covered “unexpected” damage. Was 94% unexpected? No one knew. She planted on schedule. The frost came at 3:51 AM. Her claim was denied as foreseeable. The god solved weather. It could not solve contracts.

Finance. Marcus Webb deployed the petty god of market prediction. Sixty-seven percent directional accuracy, seventy-two hours out. For eleven days his fund printed money. Then everyone had the same god. Alpha collapsed. His fund returned 2.3% for the year. The god solved prediction. Prediction ceased to matter once everyone believed it.

Materials Science. Room-temperature superconductor, published 2031. The petty god of crystallography predicted the structure. Six labs replicated it. Synthesis required significant quantities of hafnium-178, an isotope previously considered useless. The world’s only concentrated deposit sat beneath the seafloor twelve miles off the coast of Tongatapu. The Kingdom of Tonga, population 100,000, announced it would nationalize the resource and sell extraction rights to the highest bidder. Beijing made an offer. So did Brussels. Washington made a different kind of offer. In March 2032, the USS Ronald Reagan carrier strike group conducted “freedom of navigation exercises” in Tongan territorial waters. The exercises continued for six weeks. The Tongan government accepted a U.S. partnership agreement in April. The terms were not disclosed. The god solved physics. The primates solved supply chains the way primates always have.

Legal. Jennifer Martinez, public defender in Cook County, invoked the petty god of legal research. Perfect precedent retrieval. Flawless motions. The judge denied without explanation. The jury convicted in four hours. Darnell Washington got fifteen years. The god solved research. It could not solve power.

Education. The petty god of pedagogy knew exactly how eleven-year-old Tyler Robinson learned. Personalized curriculum. Seven months to grade level. His school board banned it after seventeen parents complained about “AI indoctrination.” Vote was 4-3. The god solved learning. It could not solve a school board in Maury County, Tennessee.

Climate. The 1.5°C pathway existed, specifying exact emissions by sector and country. COP47 spent three days debating methodology. Each quarterly update grew more aggressive as targets were missed. The god solved climate modeling. It could not solve sovereignty.


The Missing God

There is no petty god of coordination.

The domain is not closed. The rules are illegible. Verification is impossible. The training data does not exist because planetary-scale coordination has never been achieved. No system can learn to do a thing that has never been done.

The petty gods solved oncology, traffic, weather, finance, drugs, materials, grids, law, education, housing, climate, physics, and logistics. Each solution exists as a file on a server somewhere, optimal and unimplemented. The gods do not confer. Their solutions do not compose.

Maria Chen and Sarah Okonkwo and Eleanor Vance and Tyler Robinson and Darnell Washington will never meet. They are connected only by the same discovery: the bottleneck was never knowledge.

The gods delivered knowledge. Effortlessly. Perfectly.

The bottleneck was always each other.



The doomers think superintelligence will become Moloch: a singleton that eats everything. Wrong. The accelerationists think superintelligence can solve Moloch by out-optimizing him. Also wrong.

What arrives is a pantheon of Molochs, each petty, each grinding its domain toward local minima while the global landscape fragments.

NDASI feeds Moloch. Fifty gods, fifty new optimization pressures, fifty races to the bottom. The god of finance optimizes returns. The god of logistics optimizes throughput. The god of pharma optimizes molecules.

No Petty God Optimizes For YOU.

The pantheon of petty gods is complete. The pantheon cannot help.

The rest is up to us primates.