THE SINGULARITY IS NEAR, REVISITED

What If It Already Happened? And what of Boyd’s prescience?

My last post was about AGI. The comments I got back was about the singularity.

Fair enough. The two ideas travel together, and I’ve been carrying both around since October 2008, when a book on a Blue Horizons reading list sent me to San Jose to hear Ray Kurzweil explain why the future was coming faster than any of our planning documents assumed.

So this one is about the book. Kurzweil’s The Singularity Is Near, what it got right, what it got wrong, and the question I’ve come to think about most: if the singularity arrives, how will we know?

My suspicion is that we won’t, at least not at first. I think it happens quietly, and one day we look up and the world has changed underneath us.

HOW I GOT HERE

I read The Singularity Is Near in 2008 as an Air War College student and the first Reserve component fellow in Blue Horizons, the small Air Force program chartered to look twenty to thirty years into the future. Dr. Grant Hammond had helped shape the reading list. Grant was John Boyd’s friend and biographer, and he’d later become my colleague and one of my closest mentors when I returned to Blue Horizons as faculty. He was on assignment at the NATO Defense College in Rome that year, but his fingerprints were all over the program.

Kurzweil made the list because Blue Horizons wasn’t in the prediction business. We were looking for forces capable of breaking the strategic environment and asking what small moves we should make now. A 650-page argument that technology compounds rather than accumulates qualified.

I spent part of my research budget on the Singularity Summit in San Jose that October. Five hundred people, give or take, in the Montgomery Theater. Kurzweil was there. Peter Diamandis was there. Vernor Vinge, who had given the technological singularity its modern framing in 1993, was there.

I met all three.

That was one of the advantages of the Summit in those days. It was small. I could ask a few questions, but some of the most useful education came after the formal sessions, when the cameras and lights were off and I could listen to the conversations continue.

I didn’t come home converted. I came home curious.

WHAT KURZWEIL ACTUALLY ARGUED

Strip away the charts and The Singularity Is Near makes four big claims.

First, information technology doesn’t improve in a straight line. It rides an exponential. Kurzweil calls this the Law of Accelerating Returns, and his key point is that the curve can survive changes in the technology underneath it. Vacuum tubes give way to transistors, then integrated circuits. One paradigm reaches its limits and another takes over.

Second, human intuition is lousy at exponentials. We look backward a few years and project a straight line forward. His favorite example was the Human Genome Project. It appeared hopelessly behind when only a small fraction of the genome had been sequenced, then accelerating sequencing capability helped finish the job.

Third, Kurzweil expected three revolutions to converge: genetics, nanotechnology and robotics, with robotics encompassing strong AI. He put human-level machine intelligence around 2029 and the singularity around 2045. He still holds to both dates.

Fourth, Kurzweil’s singularity isn’t simply a machine takeover. His version is a merger. Nonbiological intelligence becomes vastly more capable while humans increasingly connect themselves to it until the distinction between biological and nonbiological intelligence gets difficult to draw.

The word itself is useful. In physics, a singularity describes a point where our existing equations cease to give useful answers. Kurzweil borrowed the concept for a future in which technological change becomes so steep that our models can no longer reliably describe what comes next.

Hold onto that idea. I’ll come back to it.

WHAT HELD UP

More than I expected in 2008.

Kurzweil’s 2029 date sounded absurd to plenty of people then. Today the argument has moved toward definitions and timing. Nvidia CEO Jensen Huang recently declared that “AGI has arrived.” Plenty of credible people disagree, Gary Marcus among them, but human-level machine intelligence is no longer a fringe subject.

Kurzweil’s broader exponential argument has held up remarkably well. Compute price-performance continued improving while the technologies underneath it changed. GPUs became central to modern AI, with increasingly specialized accelerators joining them. The particular hardware changes. The underlying progression continues.

His criticism of linear intuition held up too. I spent years after Blue Horizons in and around Pentagon planning cycles that often assumed the next five years would resemble the last five, only somewhat more expensive. Large institutions like straight lines because budgets, organizations and programs are built around them. Technology has no obligation to cooperate.

Even the merger framing feels less exotic today. I dictate to a phone. An AI converts it to text, reasons over it, searches, summarizes and argues back. Neither of us is doing quite the same job alone anymore. We’re nowhere near Kurzweil’s proposed brain-cloud interface, but the direction is recognizable.

WHAT DIDN’T

The nanotechnology and biotechnology legs of Kurzweil’s tripod haven’t kept pace with AI. Medical nanobots aren’t coursing through our bodies, and longevity escape velocity remains an aspiration rather than something visible in population statistics. Kurzweil was persuasive about information bearing technology and too aggressive about how quickly everything else would join the same curve.

He also underrated friction. The future contains regulators, wars, bureaucrats, fragile supply chains, energy constraints, capital constraints, borders and people who simply refuse to adopt something. My Blue Horizons classmate Dr. Rob Spalding later founded SEMPRE, and much of the work we do there comes down to who controls compute, networks, power and data. The physical and political world shapes the curve too.

The book also left me dissatisfied with a question that consumed much of my own Air War College research: what values/ethics do increasingly autonomous systems carry, and who decides?

My required capstone paper was supposed to be roughly twenty-five pages. It grew beyond 130, even after substantial cutting, and dealt with ethics, morality and autonomous combat systems, including what I rather unapologetically called “warbots.” The work was eventually selected for presentation to an international scientific audience in Los Angeles in 2009.

Kurzweil’s merger thesis leans optimistic because the machines remain extensions of us. I’ve always thought the values question deserved more work than that.

WHAT THE CRITICS GET RIGHT

Kurzweil has never lacked critics, and some of their objections are stronger than arguing whether 2029 should really be 2032.

An exponential curve isn’t a law of nature for every technology. Compute can improve exponentially while biology, manufacturing, energy, regulation and human adoption move at very different speeds. Kurzweil is strongest on information bearing technology and less convincing when he assumes everything else will join the same curve on his schedule. In some ways, I wish he were right.

There’s also the question of whether more computation automatically produces more intelligence. Recent AI progress has made that objection harder to state as confidently as people once did, but it hasn’t eliminated it. We still argue over what intelligence means, much less AGI.

The criticism I find most interesting is almost philosophical. Kurzweil tells us the singularity is an event horizon beyond which today’s models stop working, then describes what lies beyond it with considerable confidence: abundance, radical longevity and a largely benign merger of biological and machine intelligence.

There’s a tension there.

If the singularity really is where our models fail, Kurzweil may be strongest describing the road toward it and weakest describing what’s on the other side.

VINGE’S CLOCK HAS ALREADY RUN OUT

Vernor Vinge put the proposition more starkly.

In his 1993 paper The Coming Technological Singularity: How to Survive in the Post-Human Era, he opened with this:

“Within thirty years, we will have the technological means to create superhuman intelligence. Shortly after, the human era will be ended.”

That was 1993. Thirty years was 2023.

Vinge wasn’t simply predicting killer robots or human extinction. He considered several possible routes to superhuman intelligence, including intelligent machines, networks that collectively become superhuman, intimate human-computer interfaces and biological enhancement. His deeper argument was that once greater-than-human intelligence begins driving science and eventually improving intelligence itself, prediction from our side of the boundary begins to fail.

Kurzweil expects something more gradual. Sam Altman’s 2025 essay The Gentle Singularity pushed that idea further. Daily life continues while machine intelligence moves through more domains and capabilities that looked extraordinary a few years earlier become routine.

Others expect something much faster. Daniel Kokotajlo and the AI 2027 team have explored rapid progress toward superintelligence. Zvi Mowshowitz has questioned whether anything this consequential can remain gentle. In those versions, the quiet period may be the dangerous part. By the time everyone recognizes the transition, much of it has already happened.

The possibilities run from Tuesday continuing to feel like Tuesday to a cascade of changes arriving too quickly for us to absorb.

Vinge’s thirty-year clock has already expired. Maybe he was wrong. Maybe we’re looking for the wrong evidence.

WHY I THINK IT’S QUIET

I don’t have a better predictive model. My instinct comes partly from experience and partly from John Boyd.

In 2010 I ran the military side of the U.S. Antarctic Program out of McMurdo. Weather on the ice doesn’t announce itself. Conditions shift, your plan degrades, and nothing may feel terribly wrong until you’re already committed. Good operators keep comparing what they expected with what they’re actually seeing and act on the gap.

Boyd approached the same problem intellectually.

Most people know John Boyd through the OODA loop. I did too until Dr. Grant Hammond pushed me deeper.

Grant had known Boyd personally and wrote The Mind of War, the first intellectual biography focused on Boyd’s ideas. Years later, Air University Press asked Grant to edit and compile Boyd’s A Discourse on Winning and Losing. By then Grant and I had spent two years teaching alongside each other at Blue Horizons. We developed lessons, debated student research and talked about almost anything. Somehow we usually found our way back to Boyd.

Grant pushed me particularly hard toward Boyd’s 1976 essay Destruction and Creation. He later called it the “alpha and omega” of Boyd’s thinking.

Boyd starts with a simple problem. Human beings construct mental models so we can understand the world and act in it. Reality keeps changing. Eventually observations stop fitting the model. Contradictions accumulate.

So destroy the model. Take it apart. Examine the pieces against what you’re actually observing. Recombine them. Create something that better explains reality. Then be prepared to do it again.

Boyd practiced this himself. Hammond wrote that Boyd’s famous briefings were never intended as fixed doctrine. He constantly revised them as audiences challenged him, evidence changed or he discovered weaknesses in his own thinking. Grant had seen nineteen different versions.

That is a much more interesting Boyd than four boxes connected by arrows.

Over the years I’ve had similar opportunities to spend time with other thinkers who approach the future from very different directions, including Robert Zubrin and Vint Cerf. Zubrin thinks from first principles about engineering, exploration and what humans can build when artificial constraints are removed. Cerf helped build the architecture of the internet itself and has spent a lifetime thinking about what happens when networks connect people, machines and information at global scale.

Those relationships, beginning with Hammond and extending through people I first encountered at the 2008 Summit and others I’ve come to know since, have shaped the way I think about strategy. They also keep pulling me back toward Boyd.

Now put Vinge and Boyd next to each other.

Vinge says greater-than-human intelligence eventually takes us across an event horizon where our ability to predict what comes next breaks down.

Boyd says our ability to survive and prosper depends on continually destroying mental models that no longer correspond to reality and creating new ones.

Maybe the singularity is what happens when Vinge outruns Boyd.

When reality begins changing faster than people and institutions can destroy obsolete models and create new ones.

That’s the version of the singularity I find most interesting.

It doesn’t require a machine to announce that it has become superintelligent. It doesn’t require a government declaration or some benchmark crossing a magic number.

It shows up as mismatch.

Things that used to work stop working. Planning cycles miss. Expertise develops a shorter shelf life. Institutions optimize processes for conditions that have already disappeared. The people setting the assumptions keep answering last year’s question.

You don’t get a memo. You get a widening gap between the briefing and what’s actually happening.

WE’VE MISSED REVOLUTIONS BEFORE

There’s precedent for this.

Nobody woke up in 1780 and announced the Industrial Revolution. The term came later as people looked backward and tried to describe a transformation they had already lived through.

The internet did something similar. It reorganized commerce, media, intelligence, relationships and warfare over years. I don’t remember a particular Tuesday when everyone agreed we’d crossed from the pre-internet world into the internet age. By the time the transformation was obvious, we were already living inside it.

In 2012 I co-directed and produced Welcome to 2035: The Age of Surprise for Blue Horizons for essentially this reason. We weren’t predicting 2035. We were arguing that technological convergence would create combinations and second-order effects nobody could confidently forecast, so the requirement was agility.

The video was controversial inside Air University, so we created a separate YouTube account to host it. It blew past 100,000 views and became the most-watched video Air University had produced at the time. Wired wrote about it and largely missed what we were trying to say.

We weren’t warning people about Facebook. We were warning ourselves about surprise.

SO HOW WILL WE KNOW?

I keep coming back to the question that started this essay. If the singularity arrives, how will we know?

I don’t think a machine tells us. I don’t think Jensen Huang tells us. I don’t think Ray Kurzweil gets to ring a bell in 2045. I think we’ll know by watching the gap, à la Boyd.

How quickly are our assumptions becoming obsolete? How long does expertise remain useful? How often are machines producing solutions their human operators didn’t anticipate? How quickly are scientific discovery, engineering and production feeding one another? How much of the world around us still behaves according to models built five or ten years ago?

At some point, perhaps that gap begins widening faster than our institutions can close it.

If that happens, the singularity may already be underway before we have agreed on what to call it.

WHAT I’D TELL A LEADER

Read Kurzweil. Read Vinge. Then read Boyd. And reread Boyd.

If you’re going to read Boyd, don’t stop at OODA. Read Destruction and Creation. Then read A Discourse on Winning and Losing. Read Hammond’s introduction too. It helps explain why Boyd never wanted his ideas frozen into doctrine.

Then stop waiting for the singularity to introduce itself. Watch your own models. Where are the assumptions you inherited? How old are they? What evidence would tell you they’re failing? What would you do differently if they’re already wrong?

Observe reality. When the model stops matching, destroy it and create another. Make small moves early. Measure. Keep what works. Kill what doesn’t. Keep moving.

Kurzweil gave me the curve. Vinge gave me the horizon. Grant Hammond led me deeper into Boyd, and Boyd gave me the method I trust for operating on the wrong side of one.

If the singularity is quiet, that method is how you hear it.