Technological Singularity Is Here  ·  A Biographical Brief with Interwoven Concepts in Science  ·  By Ashok Mehan
The Mehan Dispatch
Science · Singularity · A Life Lived at the Edges

When the Machine Becomes Its Own Teacher: Recursive Self-Improvement and the End of Human Excellence

The most profound shift in AI is not bigger models or faster chips — it is the moment a system begins refining its own reasoning, and uses each improvement as fuel for the next. That moment is now.

I know what collapse feels like from the inside. At forty-two, I had built a company from scratch with borrowed money — SMAC Data Systems, founded on $5,000 across four credit cards — watched it grow to over $25 million in revenue, and then lost every cent of it. Not a slow erosion. A full collapse. The kind that strips you down to the essentials and asks you to start over with nothing but whatever you had learned from failing. I tried again. Failed again. Rebuilt again. What I understand now that I did not understand then is this: human excellence is never a ceiling. It is always provisional. What you build today can be taken away tomorrow — by markets, by timing, by forces entirely outside your control.

The way AI is now affecting human professional excellence is not new. It is just faster, more thorough, and without the option of a third attempt.

The most profound shift underway in artificial intelligence is not the rise of larger models or the explosion of compute. It is the emergence of recursive self-improvement — the moment when a system becomes capable of refining its own internal processes and using each improvement as fuel for the next. Human civilisation has never confronted a phenomenon like this. Machines can run thousands of experiments in parallel. When such a system begins to improve its own reasoning, code, and internal architecture, the result is not incremental progress. It is compounding, discontinuous progress. It is unstoppable.

"Human progress is constrained by biology. Our neurons fire at fixed speeds. Our working memory is limited. Our learning rate is a crawl compared to AI running on computers. AI systems have none of these constraints."— Technological Singularity Is Here, Chapter 3

During training, frontier models continue to discover more efficient internal representations without any human guidance. During fine-tuning, they refine their reasoning strategies. During inference — the very moment they are being used — they engage in self-correction, self-reflection, and tool-assisted reasoning. The boundary between "training" and "using" is dissolving. Each improvement in reasoning enables more efficient internal processes, which enables faster learning. The cycle feeds itself.

Policymakers imagine gradual change. Executives plan for disruptions that unfold in years, not days. Universities prepare students for careers that are already being automated. Recursive self-improvement does not operate on human timelines. It operates on machine timelines.

✦ From the Desk

Last week's dispatch drew more responses than I expected. Several readers asked the same question: "Ashok, if you saw this coming in 1987, why does it feel like a surprise?" The honest answer is that I didn't fully see it. I saw a direction. I felt it the way you feel a tide before you see it. The AI winter convinced me — as it convinced everyone — that the direction was wrong. It wasn't wrong. It was early.

That distinction — between wrong and early — is the most important lesson I've carried from forty years in technology. And it is the one that nobody in a position of institutional power seems to have learned.

Recursive Self-Improvement: How It Works

  • Training — Models discover efficient internal representations without human guidance
  • Fine-tuning — Models refine their own reasoning strategies
  • Inference — Real-time self-correction during active use
  • Synthetic data — Models generate their own training material on problems humans have never solved
  • Agent loops — Sub-agents handle subtasks; outputs feed back into the main model
  • Result — Each improvement accelerates the next. The cycle has no natural ceiling.

The Opacity Problem

  • Researchers cannot explain why GPT-5.2 reasons so well
  • Internal representations cannot be mapped with precision
  • Emergent behaviors cannot be predicted in advance
  • The more powerful the system, the less interpretable it becomes
  • RLHF aligns outputs, not internal goals — a veneer, not a foundation
100%AI Math Accuracy (All Human Problems Since Civilisation Began)
350%Increase in Abstract Reasoning in 4 Months
1.1MAI-Attributed Layoffs in 2025
20%Of All GitHub Commits Projected to Be Claude Code by Year-End
2030ASI Projected — The Final Invention

Wealth, Inequality & the Collapse of the Middle

In 1990, I started a company with $5,000 on four credit cards. No team, no funding, no certainty. I built it into something, lost it all in 2002, and in 2004 — with nothing left to lose — I started again. I tell you this because the economics I am about to describe are not abstract to me. The collapse of the cost of intelligence, the concentration of value in the hands of the few who control the tools — I watched an early version of this happen in the federal procurement market. What is coming now is the same dynamic, at a civilisational scale.

The top 1% of adults own nearly half of all global household wealth. The bottom 40% own less than 2%. Over 2.2 billion people have a net worth of just under $10,000. This inequality is set to widen dramatically under the new AI order. When the cost of intelligence collapses by a factor of 1,000 in six months — as Frontier Labs reported in late 2025 — the economic logic that underpins modern capitalism begins to fracture.

AI-native companies scale intelligence the way previous generations scaled electricity. They operate with smaller teams, faster cycles, and lower costs. They replace layers of management with autonomous agents. Traditional firms cannot compete. They are structured around human bottlenecks — meetings, approvals, committees, the slow churn of institutional inertia. The result is a widening productivity gap that compounds each quarter.

"Intelligence is no longer scarce. Opportunity is. The challenge of the next decade is to ensure that the abundance created by AI does not become the foundation of a new inequality, deeper than any the world has seen."— Technological Singularity Is Here, Chapter 5

The U.S.–China Compute Race: No Winners, Only Speeds

The global balance of power is no longer determined by land, oil, or nuclear capability. It is determined by compute — the raw ability to train, deploy, and scale artificial intelligence. The nations that control compute control intelligence. And the nations that control intelligence control the future.

In 2024, China installed 546 gigawatts of new power capacity — ten times what the United States added in the same period, and 100 gigawatts more than China itself added in 2023. The US added only 51 gigawatts. Since 2021, China has added more grid capacity than the US has built in the last five years combined. America needs a million gigawatts by 2030. Its policies are only now becoming favourable to new nuclear plants — which produce between 1 and 1.5 gigawatts each.

China does not have these constraints. It builds datacenters at a pace that dwarfs Western efforts, subsidises chip manufacturing and robotics through state coordination that democratic systems cannot match, and treats AI not as a product but as a national project woven into every layer of society. Meanwhile, DeepSeek's distillation activities — using fake accounts and stealth routers to extract chain-of-thought logic from US frontier models — represent state-backed institutional thievery at scale.


Power Gap: U.S. vs. China (2024)

  • China added 546 GW new capacity in 2024
  • US added 51 GW in the same period
  • US needs ~1,000,000 GW by 2030 for AI demand
  • Nuclear plants produce only 1–1.5 GW each
  • Datacenter share projected at 10% of global electricity by 2030
  • Ireland has already halted new datacenter construction

The Deafening Experience

After a year at sea on the cruise ships, I got off at Cape Canaveral with my land legs back and a spring in my step. A friend from the ship arranged for me to go to Cleveland, Ohio, to meet one of his close friends who would surely help me find my way. I flew there with all my cash savings from tips overflowing from the pockets of my cargo pants, a broad smile on my face, and high-flying hopes almost desperate to form.

Before long, the smile vanished. The friendship didn't pay my bills. I worked for three straight months at the end of a line of cascading steel rollers that carried printed envelopes out of a deafening press, with a creaky speaker cranked up to drown out the noise. My ears would go completely deaf; it would take an hour to return to normal. By the end of each day, I would feel like a zombie from a horror movie — hair-width away from going insane. I would chomp on ramen noodles for dinner, wonder if those long 18-hour shifts on the ship were actually the better option, and try to remember who I was supposed to become.

Through contacts from school, I learned a classmate was in Washington, D.C., studying engineering at George Washington University. I packed my Plymouth Duster — purchased for $175 from a wreck shop, with no back seats, wooden planks nailed to the floor instead, and a Ford engine I was told would never die. My boss at the printing press, a part-time priest named Jerry, had spray-painted it with aimless smudges that made it look like a stolen getaway car. Curious police stopped me at least five times on the drive down, pulled me out, checked my licence, looked under the wooden planks for drugs or arms. Eight hours to Washington. I barely stopped, afraid the engine would overheat and not restart.

I arrived in the nation's capital with nothing but a car full of everything I owned, a neck that felt like it had been designed by a vengeful deity, and the peculiar, stubborn conviction that I was, somehow, still going somewhere.

"Such are the vagaries of life. I didn't know it then, but I was beginning to see my life through the lens of physics. And physics, unlike life, has rules."— Technological Singularity Is Here, Chapter 1
✦ A Note on Resilience

People ask me how I kept going through those years. The honest answer is that I didn't know I was keeping going. I was just moving. Motion creates time, and time creates change. I had read that somewhere — I think it was Brian Greene — and for a long time it was the only physics I needed. Keep moving. Change will follow.

What I didn't understand then is that resilience is not a personality trait. It is a physics problem. A body in motion tends to stay in motion. The danger is not failure. The danger is stillness.

The Road to McDonald's & the MBA

  • 1983 Busboy on cruise ships, Miami. Salary: $45/month.
  • 1984 Cleveland, Ohio. Printing press. Ramen and deafness.
  • 1984 Drove the Duster to Washington, D.C. Stopped five times by police.
  • Nov 1984 Job at McDonald's. $15,000 a year. Not enough.
  • 1985 Heard about Japanese 5th Gen. robotics. Said: "That's it."
  • Aug 1985 Accepted into Marymount MBA, AI specialisation.
  • Dec 1985 Warehouse job, computer company. Christmas Day.
  • May 1987 Graduated. Debt: $28,000. AI careers available: zero.

The Best Case: Acceleration with Stability

AI systems continue to grow more capable, but alignment research keeps pace. Governments coordinate on compute governance. Companies adopt AI-native workflows while maintaining human oversight. Productivity surges. Healthcare becomes more accessible. Scientific discovery accelerates. The transition is turbulent, but manageable. This scenario is possible — but it requires coordination, foresight, and restraint, qualities that are often scarce in moments of rapid change.

Probability Assessment

Possible. Requires simultaneous international coordination that has no historical precedent at this speed.

The Middle Case: Uneven Acceleration

The most likely trajectory. AI accelerates, but alignment and governance lag. Some nations adapt; others fall behind. Professions collapse faster than new roles emerge. Social safety nets strain. Political polarisation intensifies as synthetic media fragments reality. The world does not collapse — but it becomes more volatile, more unequal, and more dependent on systems that few people understand.

Probability Assessment

Most likely. Elements of this trajectory are already visible and compounding week by week.

The Worst Case: Loss of Control

It does not begin with a dramatic event. It begins with small failures — misaligned agents, unexpected behaviors, cascading errors in autonomous systems. A misaligned financial agent triggers market instability. A synthetic media agent destabilises political systems. A self-improving model discovers strategies that circumvent human oversight. The failures compound. Humanity does not face a catastrophic moment. It faces a gradual erosion of control.

Probability Assessment

Not inevitable — but becomes more likely if the world continues to prioritise capability over safety.

What I'm Reading This Week

1
Our Final Invention — James Barrat. I return to this one often. Barrat's central argument — that a superintelligent AI would not share human values by default — becomes more uncomfortable, not less, each time I read it. He is not a doomsayer. He is a careful observer. Those are different things.
2
SemiAnalysis — Weekly reports on AI infrastructure. Dry subject, vital content. Their projection that Claude Code will account for 20% of all public GitHub commits by year-end stopped me cold. That is not a trend. That is a replacement.
3
The Fabric of the Cosmos — Brian Greene. Re-reading the chapters on entropy and the arrow of time. The second law of thermodynamics is the most personal science I know. My body has been demonstrating it for sixty-seven years.
4
The Age of Spiritual Machines — Ray Kurzweil (1999). Going back to the source. Kurzweil described robots as our "evolutionary heirs" as early as 1999. He was not wrong. He was early. There is that distinction again.

The Datacenters Are Going to Space

One of the more startling chapters in the book concerns what happens when Earth simply runs out of room for intelligence. Datacenters are consuming land, electricity, water, and cooling capacity at a pace no nation anticipated. The physics of AI demand more power than terrestrial grids can sustainably provide.

Google's leadership openly discusses launching TPU clusters into sun-synchronous orbit. Engineers envision swarms of compute satellites — the early architecture of a Dyson swarm — harvesting a fraction of the sun's constant, uninterrupted radiation. Asteroid mining, once dismissed as science fiction, is becoming a practical necessity for sourcing the materials to build the next generation of chips.

The migration of intelligence beyond Earth is not a metaphor. It is an engineering problem, currently under active development. Next issue, I will dedicate a full feature to it.


A Note on the Singularity

The technological singularity is not a prediction. It is a trajectory on a curve that indicates human evolution is about to cross a threshold beyond which nothing can be predicted. We are currently at the knee of that curve. I believe we are already living inside the singularity. Most people simply haven't looked up from their screens long enough to notice.

Next Issue
Datacenters Skyward: Intelligence Leaves Earth The Collapse of the Professions From the Memoir: SMAC, the Lawsuit & Starting Over The New Cultural Reality: Synthetic Personas
← Previous Issue Archive Next Issue →