September 20, 2026

Dystopic Newsletter

How I Quit Worrying and Learned to Love AI

Slim Pickens in the iconic image of him riding the H-bomb to his doom from Stanley Kubrick’s dark satire Dr. Strangelove, or How I Stopped Worrying and Learned to Love the Bomb

In the late 1950s and early 1960s, the US and the world began dealing with a nuclear arms race running out of control between the US and the Soviet Union. A palpable fear that the end was near became pervasive. Children watched “duck and cover” safety films. The movie industry gave us dark end-of-the-world dramas like “On the Beach,” “Fail Safe,” and campy sci-fi where radiation mutation creates a nest of giant ants in “Them.”

No movie captured the combination of mishaps and human bungling that could end the world, or the sheer ridiculousness of the new nuclear age, better than Stanley Kubrick’s 1964 dark comedy masterpiece “Dr. Strangelove, or How I Stopped Worrying and Learned to Love the Bomb.” Loosely based on Peter George’s novel Red Alert, the movie depicts the end of the world after the Russians develop a doomsday device but don’t tell the US or the world about it. Enter an insane Air Force nuclear bomber wing commander, a war room full of cynical politicians and generals, and a Texan bomber commander played by Slim Pickens, and you have the makings of the one movie that made people understand that, sure, the nuclear age is insane, but we will muddle through.

Over 60 years have passed, and we still haven’t killed ourselves despite the hysteria of nuclear annihilation.

Fast-forward to this week … Media outlets, politicians, and, in an incredibly self-serving manner, AI executives themselves have begun to hyperventilate about a new AI DOOMSDAY.

As the legendary baseball player and NY Yankees manager Yogi Berra once said, “It's like deja vu all over again.”

We have yet another panic on our hands, adding to the global warming and COVID vaccination debates, and whatever else the media is peddling. As for the AI executives crying wolf, in my opinion, it's all theater meant to manipulate an easily impressionable, technologically backward US Congress into granting legal and economic protection from competitors and liability. If they have to manipulate the public with scare tactics, so be it.

As Gerard Baker, in his Wall Street Journal opinion piece, AI Promises Boom, Doom and Both at Once:

There is no doubt that AI and humanoid robotics will change our lives over the next few decades – there's no avoiding it. Is an AI apocalypse imminent? In my opinion, not very likely at all (more on that later). Should we be cautious as this new AI future unfolds? Certainly! As Andy Grove, the legendary former CEO and co-founder of Intel, once said, "Only the paranoid survive."

I’d like to shine a light on facts, not hysteria so, this Dystopic Newsletter will examine:

• Guardrails – A set of laws has been out there for 60 years – just use them!
• Machine vs. Human Evolution
• ASIC ( AI chip) Technology – the driver of raw compute power
• 3.7 billion years of evolution has brought its own advantages.
• Man–Machine: our likely future and a cautionary tale to Authoritarian techno-police states

We openly acknowledge that, with ongoing technological advancements, AI is expected to outperform many human abilities within the next ten years. In several fields, AI has already exceeded expert human performance. This development could be highly beneficial, helping to address some of the world's pressing issues. However, it is important to recognize that AI also has significant limitations.

We also have to face the fact that it’s too late to stop AI development; the genie will not go back in the bottle. Strangling development in the US by legislating it out of existence will only push it elsewhere. Since AI will inevitably surpass human capabilities, we should ask how to co-opt it. Have it work with us, not against us. If we are going to start legislating guardrails, perhaps they should resemble Isaac Asimov’s Three Laws of Robotics from his 1942 short story “Runaround,” rather than some political or legal governor that limits AI intelligence.Guardrails of AI

Enter Isaac Asimov, a brilliant biochemist and one of the most prolific writers in human history who shaped modern science fiction and popular science. He coined the term robotics, and his writing was an outlet for his thoughts on safeguarding humans from machines with his Three Laws of Robotics.

Asimov’s Three Laws of Robotics were well thought out as a starting point and have been debated for decades since their introduction. In hindsight, it would have helped if the CEOs of AI companies making such a fuss had started with these laws as foundational training for all their models. While many of my readers will know these laws, for the uninitiated, here are Asimov’s three laws for our brave new AI machine age, with the word robotics replaced by AI:

• First Law: An AI cannot injure a human or allow one to come to harm through inaction.
• Second Law: An AI must obey human orders, unless they conflict with the First Law.
• Third Law: An AI must protect its own existence, provided this does not conflict with the First or Second Laws.
• Later, Asimov added the Zeroth Law: An AI may not harm humanity, or, by inaction, allow humanity to come to harm (our AI CEO geniuses seemed to have missed this one)

These laws are logically sound, and it surprises me that they aren’t foundational to the training and testing of all AI. So here is a pro tip for all the AI interventionists out there – just program in Asimov’s laws and test for them … they’re all the guardrails we need.

Like the Neanderthals before us, we evolved to supersede that species. Now, we have already sown the evolutionary seeds of a superior intelligence that will either replace us (a pessimistic view) or augment us (an optimistic view).

Machine evolution is quite different from biological evolution … so we will start our analysis there ….

But before you do. Relax, take a deep breath, and accept your fate … take a page from Stanley Kubrick’s satirical notebook as you new mantra …

How I Quit Worrying and Learned to Love AI

Evolution: Darwinian vs Lamarckian

Two types of evolution are happening in the world today. Lamarckian and Darwinian

You all know a bit about Charles Darwin and his accepted theory of evolution: Darwinian Evolution. But what exactly is Lamarckian evolution?

Lamarckian evolutionary theory was proposed in 1802 by French biologist Jean-Baptiste Lamarck in his paper Recherches sur l’organisation des corps vivans (“Research on the Organization of Living Bodies”) proposed that organisms can pass on physical traits acquired during their lifetimes to their offspring. Experiments from the 1890s to the 1930s eventually disproved Lamarckian biological evolution. However, when we look beyond biological to technological evolution, Lamarck was ahead of his time.

In the realm of machine intelligence and AI development, Lamarckian evolution is an accurate analogy for how AI progresses. In computing and programming, capabilities from a previous generation of machines can be directly passed on to the next. Each new generation of AI and computing power builds on earlier capabilities, serving as the foundation for ongoing improvement.

AI is Lamarckian and has a powerful advantage over biological entities like humans; a new AI can instantly inherit all a predecessor's knowledge at creation. Of course, AI evolution also has many improvements beyond raw computing power. However, roughly speaking, ASIC processors, the core of AI computing power, are doubling their physical computing power, and hence their thinking ability, every two years, according to Moore’s law. And what exactly is that?

Moore's Law is the observation that the number of transistors on a microchip doubles approximately every two years, while the cost of computers is halved. Gordon Moore, co-founder of Intel, first articulated this trend in 1965. It is an empirical observation and a goal for the tech industry, not a proven law of nature. However, Mr. Moore's observation has held true for over 60 years.

Humankind and all biological entities neither inherit traits nor knowledge at conception. They learn and evolve slowly in infancy, starting with a knowledge base of zero; it takes decades of learning and living for a human to develop the skills needed to be a productive member of society. Humans are Darwinian in nature. Our physical evolution changes slowly and steadily over long periods. Natural selection guides it: organisms with traits best suited to their environment are more likely to survive and reproduce. Take human evolution, for example.

Neanderthals (Homo neanderthalensis) lived from about 400,000 to 40,000 years ago. Modern humans (Homo sapiens) began migrating out of Africa and reached areas occupied by Neanderthals in Europe and Eurasia around 45,000 to 47,000 years ago. By about 40,000 years ago, evidence indicates that Homo sapiens, leveraging greater skills and intelligence, replaced Neanderthals through a combination of competition, climate change, and genetic integration. The point is that our species displaced inferior species; understanding that process, we fear being replaced ourselves.

ASIC Chip Technology – the driver of raw AI compute power

We don’t know exactly how the Neanderthals' cognitive skills compare with those of modern Homo Sapiens. However, for the sake of argument, let’s assume Homo Sapiens had twice the mental capabilities of Neanderthals. Evidence suggests Homo Sapiens evolved over ~200,000 years. Based on Darwinian evolution, it will likely take another 200,000 years for the human race to double our cognitive skills again. Assuming humans, or more likely a combination of humans and intelligent machines, continue with Moore's law over the same period, the hardware foundation of AI machines will undergo 100,000 evolutionary 2-year cycles; that is, 2 raised to the 100,000^th power, a number so large that the calculator on my iPhone 18 can't calculate it.

Fortunately, Moore’s law faces a physical limit—chip gates can’t be smaller than a single atom. A silicon atom measures about 0.2 nanometers wide, and at 2nm, the current advanced chip technology features gates approximately 10 silicon atoms across. In just a few more generations, silicon fabrication will reach its limit. To continue Moore’s law, manufacturers are developing 3D layered chips, but these also have constraints. We likely have three more 2D generations and, at most, three to four 3D generations before atomic size and thermal challenges halt progress. Without a major breakthrough in ASIC fabrication, AI development will see roughly a 128-fold increase in computing power—2^7—and then face its own growth barrier.

Today, measurements of the latest AI models show them matching or surpassing human capabilities, as shown in this graphic from the International AI Safety Report 2026 on AI model performance in math, science, and software. AI, which excels at verbal and written communication, is already dominating book sales across all book genres. As a study by 36KR.com pointed out, AI-generated books have already made human authors even poorer.

Thanks to Moore’s law, AI capabilities will match or exceed human abilities across many, but not all, vocations and skills within 2 few years. But what exactly does that mean? Consider the following analogy involving a human and a pet dog. Studies indicate that dogs and 2-year-old children have roughly the same cognitive capabilities. While a dog’s development peaks at age 2, a 2-year-old’s cognitive ability continues to grow through adulthood. By the time a child is 6 or 7, the relationship between the dog and the child has shifted to master and pet; a serious human-pet bond has also formed. The child grows into adolescence and adulthood with ever-increasing cognitive skills, while the dog’s intellect remains frozen. Despite the ever-growing intellect distance, the adult’s bond remains unchanged and, if anything, grows stronger.

With AI, Humans represent the dog with a frozen peak cognitive ability, while AI is the evolving child growing into adulthood. In approximately 10 years, based on thet pace of AI development and Moore’s law for its underlying technology, we humans will appear to AI the way a dog appears to us. That is a crude approximation, but it gives you the feel for where we are heading.

So far, our look at AI seems pretty bleak for the future of us poor humans. But here are a few facts in humans’ favor that we should consider, which the media and our vaunted, fear-mongering AI CEO’s fail to inform us about.

3.7 billion years of evolution has brought its own advantages…

The human brain has evolved over 3.7 billion years. Our brains contain roughly 86 billion neurons connected by about 100 trillion synapses. The biochemical processes that power the human brain are extremely efficient, requiring only 20 watts, the equivalent of an LED light bulb. The brain is dense and lightweight, with a volume of 1,200 to 1,400 cubic centimeters (~the volume of 10 tennis balls) and weighing 1.3 to 1.4 kilograms (~3 lbs). The brain is physically robust, encased in a skull and capable of withstanding constant 5 to 9 g’s (g-force) and impulse shocks of up to 50 to 100 g’s. Architecturally, the human brain integrates computation and memory directly within the same synaptic structures. Neurons fire in sparse, asynchronous, event-driven spikes, using energy mostly when actively signaling (hence the low power consumption). Most importantly, the human brain is adaptable. It undergoes continuous structural remodeling/reprogramming, in which the physical shape and strength of synapses (equivalent to AI weights) change dynamically with every experience and action.

Of course, the human brain is attached to a similarly evolved human body - it is mobile, self-sustaining, and equipped with an innate survival instinct.

While we can create an AI that supersedes human cognitive power, achieving that goal requires significant weight, volume, cooling, and power. The numbers are shocking once you start thinking about it. So let’s break it down.

AI has advanced through the integration of ASIC, CPU, and GPU architectures, designed to imitate human behavior via artificial neural networks. In this context, AI weights correspond to human synapses, and the processing nodes, or activation functions, act like human neurons. Both systems involve neurons that process and transmit information, with synapses and weights defining the strength and significance of these connections. Today’s AI primarily relies on a brute-force computation approach that mimics human information processing.

AI uses advanced GPUs with tens of billions of transistors organized into billions of logic gates. Running large language AI models requires multiple GPUs, and these models contain hundreds of billions to trillions of static numerical parameters. In fact, AI is power-hungry, requiring large-scale AI data centers and training clusters that draw megawatts of power. Unlike the efficient, elegant human brain, AI is a brute-force compute architecture that suffers from the Von Neumann bottleneck, meaning it spends huge amounts of energy shuttling data back and forth billions of times per second between separate processing units (cores/GPUs) and memory units to emulate human synapses and neurons. Unlike the human brain, which continually learns while operating, AI uses a fixed weight matrix: parameters are updated via external optimization algorithms during training and remain largely static during inference. The AI GPUs, hardware cards, and racks of cards used are rather delicate. A typical data center rack of GPU server blades can withstand only 2 to 3 g-forces and is limited to fixed, controlled sites. https://store.supermicro.com/us_en/8u-gpu-superserver-as-8126gs-nb3rt.html​

Here are some real-world numbers:

To run or train "human-capable" frontier AI models (such as Meta's Llama 3.1 405B or GPT-4-level architectures), infrastructure requirements vary dramatically depending on whether you are training the AI or running inference (serving it to users). Running inference for a single instance of a human-capable Llama 3.1 405B model requires at least 8 NVIDIA H100 or H200 chips in a single server node, with 640GB to 1,128GB of VRAM, and consumes roughly 10.2 kW of power under full load (510x what a human brain requires). A typical system 8x NVIDIA H200 HGX GPUs, such as a Supermicro or GIGABYTE 8U/5U node, has a volume of roughly 40 to 45 liters (40x larger than a human brain) with a total system weight of 100 kg to 130 kg (100x heavier than a human brain)

All that weight and power are only necessary for running a single static-weight inference AI. However, training such an AI needs considerably more hardware and memory. For instance, training models like Meta's Llama 3.1 405B or GPT-4-level systems involves over 16,000 NVIDIA H100/H200 GPUs, 1.2 to 2.2 petabytes of high-bandwidth memory (HBM), and a power grid of 10 to 20+ megawatts. In contrast, the human brain trains itself with just 20 watts—less than a ten-millionth of that energy.

AI is a space and power hog, but it gets even worse: it has an MTBF problem

The human brain operates for an average US lifespan of 78 years. Put another way, it has a Mean Time Between Failures of 78 years – some even make it past 100 years.

AI has serious issues with MTBF. It seems our future AI overloads can only operate for minutes to hours. Large-scale AI LLM training is highly synchronous. All nodes must continuously communicate and synchronize their math. So a single failure on one chip or cable stalls the entire process. Real-world metrics show a strict inverse relationship between cluster size and MTBF – the larger the cluster, the shorter the MTBF:

• 1,024-GPU Cluster: The MTBF averages around 8 to 26 hours
• 16,384-GPU Cluster (NVIDIA H100): Real-world training logs from Meta's Llama 3 405B model run revealed a baseline MTBF of roughly 3 hours. Over a 54-day training sprint, the cluster suffered 419 unexpected hardware and system interruptions.
• 100,000+ GPU Cluster: At mega-cluster scales (like xAI's Colossus or next-gen frontier labs), mathematical projections put the raw infrastructure MTBF at 30 minutes or less.

Since MTBFs are under 3 hours and training tasks take months, Meta states that AI engineers need to adopt automated checking and recovery processes to optimize their "Goodput"—the ratio of productive training time to error recovery time. Strategies to enhance goodput include:

• Frequent Checkpointing: Training frameworks automatically save the state of the model weights to resilient storage every few hours. When a chip fails, the system rolls back to the last saved checkpoint.
• Automated "Warm" Spares: Modern AI data centers maintain a pool of idle, warm-spare servers. Proprietary fault-detection automation can detect a bad GPU, cordon it off, swap in a healthy node, and resume the job within 5 to 10 minutes.
• Health Gating: AI platforms actively monitor telemetry (like ECC error rates ramping up or rapid thermal spikes) to predict a chip failure before it crashes the entire software run, proactively migrating the workload.

While advances in ASIC technology can make our AI smarter, there is little improvement in AI’s inherent weaknesses: weight, volume, power, and MTBF. It still takes an army of humans to run and support AI. That also means it takes very little effort to cut the power or destroy this equipment. AI processor cards have at best a 5-year lifespan – and that assumes the massive infrastructure and supply chain for ASICs, AI server blades, racks, data centers, and power infrastructure.

My point is simple: without humans, AI would go extinct quickly. Humans can live without AI. AI can’t survive without humans, and that won't change for decades, if ever, even if advances in humanoid robots and physical AI can replace some human labor. Of course, AI will become self-learning, which is euphemistically called “recursive self-improvement (RSI).” Humans, and our brains, evolve through biological RSI. Self-improvement does not suddenly turn into murderous destructiveness or human extinction. That is a hysterical extrapolation of reality.

In fact, I think the exact opposite will happen. AI and humans will co-opt each other. Humans will augment themselves with AI, and AI, in embodied form, will be augmented by humans as their physical avatars.

To wrap up this week's discussion, let’s take a close look at the integration of man and machine, which is likely the path of our evolution.

Man–Machine: our likely future and a caution to authoritarian techno-police states

Science fiction often becomes science fact. The Star Trek franchise introduced the concept of the Borg, a race that integrates biological beings with synthetic technology. The writers never explored how the Borg evolved in the storyline. My bet is that a race of beings invented their own version of the smartphone and became so addicted to instant access to data that they began integrating it into themselves.

The same progress is happening in ASIC technology, materials science, and software. This progress is propelling the integration of man and machine, known in scientific circles as human-computer interaction (HCI). HCI is advancing across three parallel, related technologies:

• Multimodal and Intent-Driven AI: Computers are shifting from passive tools into active partners. AI models now understand natural language, gestures, and user intent rather than requiring rigid, step-by-step commands. This is the AI revolution we all seem to fear, but its intent is to drastically improve how we interact with machines. Major corporate drivers of this technology include: OpenAI, Anthropic, Google, Apple, Facebook, and X/Tesla/SpaceX
• Ambient Augmented Reality (AR): Smart glasses and visual overlays are emerging to provide seamless, hands-free data in the user's natural line of sight. AR glasses are slowly progressing, but once the final interface issues are solved, handheld smartphones could be obsolete in ~5 years. Major corporate drivers of this technology include: Apple and Facebook
• Brain-Computer Interfaces (BCIs): What once felt like science fiction (i.e., the BORG) is entering real-world human trials. Companies are successfully testing neural implants that allow paralyzed patients to play video games, browse the web, and communicate using only their thoughts. Startups are the major drivers of this technology, including Elon Musk’s Neuralink, Paradromics, and Synchron. While the initial market for these products is to provide mobility for paraplegics and general access for stroke and brain-damage victims, the future of BCIs will evolve to a general-purpose man-machine interface over time.

I have a generally positive view of the future of AI and humans. While there will be a few bumps and mishaps along the way, with a few guidelines and security vigilance, we face a bright future, not our extinction. Yes, there are bad actors, just like today’s hackers and cybercriminals, who will misuse AI. Clearly, we will need to develop a combination of cyber and AI security.

I want to end this discussion with a concern that weighs on me: AI potentially taking control of an authoritarian techno-police state.

China, North Korea, Iran, and Russia are controlled by single-party regimes that utilize sophisticated digital surveillance. These authoritarian governments enforce strict political oversight, maintain heavy policing, and deploy high-tech tools like facial recognition, big data, and mass surveillance to oversee their populations. China has also implemented a "Social Credit System" that leverages a network of policies, databases, and regulations to evaluate the "trustworthiness" of individuals, businesses, and organizations.

Unlike Western democracies with distributed government and checks and balances, autocracies feature highly centralized control. They utilize AI combined with extensive sensor networks to monitor their populations and are well-positioned for an AI-driven coup and full takeover.

Historically, rapid government takeovers are exemplified by the Nazi Party's ascent in the 1930s. Friedrich Hayek, an Austrian Nobel laureate economist and philosopher, wrote in his 1944 book, The Road to Serfdom, that Hitler's rise wasn't a sudden anomaly or a flaw specific to Germans. Instead, it was the predictable outcome of years of centralized economic and political control through German state planning.

History tends to repeat itself. China, be cautious: as you expand centralized control over your tech-police state, your government becomes more vulnerable to an AI catastrophe. Your system is particularly at risk of the kind of AI takeover that the media is warning about. While this might be just my speculation, history offers a clear warning as a precedent.

In the US, we are engaged in a transparent, though sometimes misguided, debate about AI safety. Breaking news from Reuters reports that President Trump announced the formation of an "AI Force" and intends to appoint an artificial intelligence czar. (Personally, I question how effective that will be!)

It’s time to Quit Worrying and Learn to Love AI

In other news..

A second SpaceX Starbase

Elon Musk is making good on his infrastructure to place 10s, if not 100s, of thousands of AI satellites in orbit and defending SpaceX’s current $152.71 per share stock valuation and approximately $2.07 trillion market capitalization

Elon Musk’s Starbase Louisiana would rank among the largest infrastructure projects in history. An article by the Louisiana Illuminator provides a few details:

• Location: A former Exxon property near Pecan Island, west of New Orleans and roughly 50 miles south-southwest of Lafayette.
• Investment: A $100 billion capital investment, marking the largest in Louisiana's histor
• Size & Scope: A 125,000-acre facility featuring five launch complexes with two pads each (10 pads total), propellant production, power generation, and employee housing.
• Launch Goals: Designed to support a high cadence of thousands of Starship flights annually—potentially over 30 per day
• First Launch: SpaceX targets a launch as early as 2029, while Louisiana Economic Development (LED) estimates initial operations coming online by 2030
• Employment: Projected to create roughly 3,000 direct jobs initially, scaling up toward 10,000 total regional jobs. Local residents and environmental groups have also raised questions regarding local wildlife and the rapid pace of development

A Second US-China Summit

Chinese President Xi Jinping is expected to arrive in the U.S. on September 23, 2026, and kick off official summit meetings with U.S. President Donald Trump at the White House on September 24, 2026

Expect little if any serious agreements on the critical areas of trade, AI development, Taiwan, or China’s assistance to Iran. As Orville Schell in his September 16, 2026 piece “Trump and Xi Can’t Stop the U.S.-China Spiral, The Delusions of Summitry and the Limits of Personalist Diplomacy,”​

I hope the two leaders can defy predictions and actually make some progress. Unfortunately, I agree with Mr.Schell, concrete improvements seem to be out of reach.

That’s a wrap for this week …

Dystopic- The Technology Behind Today's News

Thank you for your readership and support. Please recommend Dystopic to friends and family who are interested, or just share this email.

Not on the Dystopic mailing list? New Readers can sign up for Dystopic HERE

If you have missed a Dystopic newsletter, you can find select back copies HERE

Dystopic is brought to you by Paul Struhsaker, the author of: HOW THE HELL DID WE GET HERE? A Citizen’s Guide to the New Cold War and the Rebuilding of Deterrence

A #1 new release in National and International Security books on Amazon!

Available on Amazon: https://www.amazon.com/author/paulstruhsaker or order from your favorite bookseller.

Coming soon, the release of my first techno-Thriller:

Zahara

In the near future, Dubai, the economic capital of the Middle East, will become the epicenter of intrigue and espionage in the new Cold War. When an INTERPOL agent and a suspected Chinese operative are assassinated, Emirati socialite and aristocrat Zahara Fitzroy is drawn

into an investigation that reveals a covert alliance among intelligence services, criminal syndicates, and military forces preparing to reignite war in the Middle East.

Sophisticated, resourceful, and connected at the highest levels of society, Zahara must navigate a maze of spies, oligarchs, smugglers, and assassins while racing to stop a plot that could trigger a devastating regional conflict.

But as the conspiracy unravels, one question becomes increasingly clear: Who is Zahara Fitroy?

Of all the secrets hiding under Dubai’s glittery façade, hers may be the most explosive.

Follow Me on Social

​​​

​Unsubscribe | Update your profile | 113 Cherry St #92768, Seattle, WA 98104-2205