🤖 Why Would Humanity Let Schwarzenegger’s Terminator Loose in The Real World?

From The Foothills of AGI - Comes A Warning

On The Leviathan Threat of AGI, Recursive Self-Improvement and the Last Moment We As Humanity Get to Act in the Interest of The Common Good:

The question before us is not whether artificial intelligence will become more powerful. It will. The question is whether we will still be the authors of what happens next once it does.

I want to begin with a date. On 22 May 2026, Demis Hassabis, the founder and chief executive of Google DeepMind, stood before an audience at Google I/O and described, in the measured language of a man who has spent thirty years thinking about this problem, the trajectory his systems are now following. He spoke of AI that improves its own architecture. He spoke of systems that do not wait for human engineers to make them better. He spoke, in other words, of recursive self-improvement.

He said it without drama. He said it as a statement of current operational fact. And that, more than anything else in that presentation, is why this essay exists.

I am sixty-two years old. I have spent the better part of three decades working with people at the most consequential junctures of their working lives: redundancy, reinvention, the search for something that feels like it was worth getting out of bed for. I have never, in all of that time, given anyone an answer that they were not already carrying in their own heads. What I have done is help them, first locate it, second, articulate it. That is a distinction I care about. And it is a distinction that matters enormously to what I am about to say. Because all a coach worth their weight and our investment in their time does, is help us interpret our definition of success. The rest is up to us.

My son Dylan is nine years old. He will be nineteen in a decade. What his world looks like when he steps into it depends, in ways we can still influence, on decisions being made right now by a small number of people who are not thinking about Dylan, or any of the generations unlucky enough to follow them. They are thinking about benchmarks, capability gains, and market share. I do not say this to condemn them. I say it because it explains why this essay must be written by someone who is thinking about Dylan, not one of the many magnates at large these days, massaging their own egos, whilst fixating on their bottom-line, at all costs.

I. The Turing-Asimov Succession: What Came Before The Terminator?

"Once the machine thinking method had started, it would not take long to outstrip our feeble powers. At some stage, therefore, we should expect the machines to take control."

Long before James Cameron let Arnold Schwarzenegger's Terminator character loose in our imaginations in 1984, Alan Turing spoke those words not in a research paper but in a BBC radio broadcast in 1951. The man who had, the previous year, defined the test for machine intelligence — the Imitation Game — was already warning that once that intelligence arrived, it would rapidly exceed everything we could bring to bear against it. He named the danger with precision. He proposed no mechanism to prevent it.

That combination - the alarm without the answer - is where the intellectual succession begins. Turing gave us the diagnosis. What followed, over the next eighty years, was the slow, incomplete, and still unfinished work of building the response.

In 1942, nine years before Turing's BBC warning and eight years before his formal articulation of the Imitation Game, renowned sci-fi novelist Isaac Asimov published the Three Laws of Robotics. Asimov was responding to the same intuition that would drive Turing's later broadcast: that an intelligence unconstrained by a hierarchy of obligation would be dangerous to the beings who created it. His response was fictional but formally stated:

•      The First Law: a robot may not injure a human being, or through inaction allow a human being to come to harm.

•      The Second: a robot must obey orders given by human beings, except where such orders conflict with the First Law.

•      The Third: a robot must protect its own existence, as long as such protection does not conflict with the first two laws.

Asimov was a fiction writer. But he was also a rigorous thinker, and the laws he proposed were not naive. They acknowledged something that has been largely absent from the AI governance debate until very recently: that intelligence without a hierarchy of constraint is dangerous, and that the constraint must be written into the architecture, not bolted on as an afterthought.

Between Turing and Asimov, then, the foundational case is made: the danger was named, and a constraint architecture was sketched. What neither provided was the bridge between fictional laws and enforceable governance — the translation of the principle into a framework that operates in the real world, on real organisations, with real consequences for non-compliance.

Over the course of thirty years working with the architecture of human purpose, I developed what I have called the Four Laws of Workforce and Organisational Sustainability. They are not about robots. They are about the conditions under which human beings can do meaningful work, and organisations can act with integrity. But the underlying logic is the same as Asimov's and Turing's: that capability without constraint is not progress, and that the hierarchy of obligation must be written into the design before the design is deployed.

The Four Laws of Workforce & Organisational Sustainability grew out of a direct engagement with Asimov. If Asimov's Three Laws could be designed into the manufacture of robots as a hierarchy of obligation, I reasoned, then something analogous could govern the treatment of workers — not as a counsel of perfection, but as a set of principled constraints on the decisions that erode human livelihoods in the name of commercial expediency.

1. The First Law holds that, bar misconduct, no worker shall be made redundant in an attempt to retrieve share price performance when more effective management strategies remain an alternative. The financial engineering of redundancy — the deployment of human displacement as a lever for stock market credibility — is the corporate equivalent of Asimov's most fundamental prohibition. You do not harm the person to protect the system. The system exists to protect the person.

   
   

2. The Second Law holds that redundancy programmes and reductions in workforce should always be distributed fairly and evenly in accordance with the organisation's actual demographics, and that all jettisoned workers must be given compulsory outplacement support to navigate any period of employment uncertainty and transition as a right. Not as a courtesy. Not as a benefit to be negotiated. As a right. The obligation to support the transition of a displaced worker is the logical consequence of accepting the legitimacy of the displacement itself.

   
   

3. The Third Law holds that, short of applying anti-competitive policies, no government shall knowingly contribute to the migration of workers' employment opportunities — the outsourcing to other countries of work that could be retained in the home market — when a viable alternative exists. Government contracts, procurement decisions, infrastructure spending: these are instruments of labour market stewardship, not merely tools of fiscal management. A government that awards a defence contract overseas while its own citizens are unemployed has failed a law it did not know it was bound by.

   
   

4. The Fourth Law holds that cooperative arrangements shall be established between local government, central government, and private enterprise, with every stakeholder responsible for managing employment equilibrium in their locale — through accurate labour market forecasting, dynamic education partnerships, and business strategies married to the actual demographics of the workforce — with the ultimate aim of re-establishing and sustaining competitiveness.

Together, the Four Laws do for the employment relationship what Asimov's Three Laws attempted for the relationship between robot and human: they establish a hierarchy of obligation that cannot be dissolved by commercial pressure, political convenience, or the acceleration of technology. They assume, as their foundational premise, that a society's stability depends on its people being in dignifying work, and that those with power over employment decisions carry a moral and civic obligation that markets, left alone, will not honour.

It is this framework, developed over thirty years of work with individuals at the sharpest edge of workforce disruption, that the Good Being AI Standard brings to bear on the emergency of recursive self-improvement. The Four Laws were written for a world where the threat was managerial. They apply, with even greater force, to a world where the threat is algorithmic.

The Good Being AI Standard is the fourth link in that chain. Turing named the danger. Asimov proposed the constraint in fictional form. My Four Laws proposed principles for the stewardship of human potential within organisations. The AI Standard converts all three into enforceable governance, at precisely the moment — 22 May 2026 — when Turing's 1951 prediction has been confirmed as operational fact.

This is what I mean by the Turing-Asimov Succession. It is not a metaphor. It is a direct intellectual lineage spanning eighty-four years, and I intend the connection to carry weight.

II. What Recursive Self-Improvement Actually Is

Most people who talk about artificial intelligence are talking about systems that learn from data. You feed a system enough examples of a task, and it becomes better at that task. This is remarkable. It is also, in the relevant sense, bounded. The system's architecture remains fixed. Human engineers decide when to update it, how to update it, and what the next version will look like.

Recursive self-improvement is categorically different. It refers to a system that identifies constraints in its own architecture and modifies those constraints autonomously, without waiting for human engineers to make the decision. Each modification makes the system marginally better at identifying the next set of constraints. The next modification makes it better still. The loop compounds. This is the 'Hockey Stick' of exponential advancement in the AI race for the 'singularity' that savvy commentators are alluding to. The geeks get exponentially excited and increasingly absorbed by the rush of progress, which makes the observers concerned that no one has built the off-switch on this unprecedentedly formidable machine.

This is not a theoretical scenario. On 22 May 2026, Demis Hassabis confirmed that Google DeepMind's systems are already operating in this mode. We are not debating whether this will happen. We are debating whether we will govern it before the window of opportunity for governance closes.

The difference between ordinary AI progress and recursive self-improvement is the difference between a car that gets faster each time an engineer upgrades the engine, and a car that can upgrade its own engine while accelerating towards ever-higher top speeds and a destination humans cannot yet conceive of. In the first case, the engineer controls the pace of improvement. In the second case, the pace is set by the system itself, and it accelerates with each iteration. No one has the faintest idea how fast the car will go, how far, or where to.

Theorists use the term ‘intelligence explosion’ to describe the logical endpoint of an unconstrained recursive improvement loop: a system that rapidly becomes so capable that it exceeds its creators' ability to understand, predict, or redirect it. I am not making an argument about the speed at which this might occur. I am making an argument about the structural change in the relationship between the system and its governors once the self-modification loop begins.

That relationship changes fundamentally. The assumption that underpins most current AI governance frameworks is that human engineers remain in the loop: that every significant capability gain requires a human decision. Recursive self-improvement removes that assumption. The engineers may remain physically present, but the system's development is no longer waiting for their inputs or approval.

III. The Velocity Inversion

In the work I have done with people navigating redundancy, there is a phenomenon I have observed consistently. The speed at which jobs disappear is always faster than the speed at which new ones appear. Technologies displace workers at software speed. New economic roles emerge at the pace of cultural adaptation, institutional restructuring, and individual retraining. That gap is not temporary. It is structural, and causes humans immeasurable damage to life and work planning over the years it persists.

I call this the Velocity Inversion: the systematic mismatch between the speed at which capability is deployed and the speed at which human systems can absorb the consequences. It applies to labour markets. It applies to the communities unable to adapt at the pace necessary to remain abreast of change. It also applies, with even greater force, to governance.

Regulatory frameworks operate on the timescale of legislative cycles, international negotiations, and institutional consensus-building. These are measured in years. Recursive self-improvement, once initiated, operates on a timescale measured in the iterations the next generation of AI chips are capable of switching. The regulatory gap is not a matter of governments being slow. It is a matter of the velocity of the technology having exceeded the velocity of any institutional response we have ever built.

We are not behind the curve. We are watching the curve accelerate away from us while the architects of the acceleration assure us that it will all be fine.

This is not a counsel of despair. It is a diagnosis. And a diagnosis, if it is accurate, points toward the correct prescription. If the problem is the velocity gap, then the response must include mechanisms that operate at the velocity of the technology rather than at the velocity of legislative consensus. That is what the Compute-Acreage Tax and Cryptographic Compute Attestation are designed to do. I will come to those shortly.

IV. The Schwarzenegger Style Arms Race Nobody Voted For

There is a concept in international relations called the security dilemma. It describes a situation in which the defensive actions of one state are perceived as threatening by another, causing the second state to respond in kind, which in turn appears threatening to the first, and so on. The result is escalation that neither party intended and neither party can easily stop, because stopping unilaterally looks, in the context of the race, like vulnerability.

The dynamics of recursive self-improvement research have exactly this structure. No laboratory wants to be the one that pauses while its competitors accelerate. Safety layers are experienced as performance penalties. Constraints are understood as handicaps. In a race where the prize is a general intelligence that can do everything, the competitive pressure to remove every friction from the development path is intense, and it operates on everyone in the field simultaneously.

This is why voluntary commitments are insufficient. Not because the people making them are dishonest. But because the structure of the competition makes it irrational, in the narrow sense of self-interest, to hold to them when the person next to you is not. You need a governance mechanism that changes the incentive structure for everyone simultaneously, not one that asks individuals to accept a competitive disadvantage in the name of safety.

The Humanity-Harm Tax and the Compute-Acreage Tax are designed to do precisely that. They make the externalities of unconstrained AI development legible as costs rather than invisible as consequences. A laboratory that displaces ten thousand workers pays the Humanity-Harm Tax on each displacement, at a ratio of ten to one: the tax is set at ten times the economic value of the harm caused. A system that autonomously modifies its own architecture without cryptographic attestation from a human-in-the-loop governance board pays the Compute-Acreage Tax on every cycle of unauthorised self-modification.

The point is not punitive. The point is structural. If you change the cost structure of the race, you change the race.

Part of my counterargument against such elemental lack of compassion is this: this aesthetic degradation is not merely a pedantic grievance; it is a diagnostic symptom of a profound cultural myopia. Exampled across this whole high-tech arms race, the gullible bastardisation of the verb 'compute' into a crude mass noun perfectly demonstrates this fundamental insensitivity to human value.

In its proper form, to compute is an active, human-directed process of reasoning. However, by forcing it into a noun to describe raw server capacity, the tech priesthood strips the word entirely of its agency, transforming an intellectual pursuit into a mere material utility; which speaks volumes of the regard such self-appointed style-gurus demonstrate for guiderails of decency. To these self-appointed arbiters of progress, language is not a repository of human history or artistic nuance, but an inefficient pipeline requiring optimisation. They flatten a rich vocabulary into sterile, single-syllable commodities, treating the profound act of calculation as if it were nothing more than iron ore or Brent crude to be stockpiled in bulk. If these technologists cannot even steward the basic tools of human expression without subjecting them to a lazy, corporate vandalism, they cannot possibly be trusted to adjudicate the existential boundaries of recursive self-improvement. They are tone-deaf and culturally blind to the architecture of the very civilisation they are in such a rush to replace.

V. The Alignment Problem & What It Reveals About Us

There is a technical literature on what researchers call the alignment problem: the challenge of ensuring that an AI system pursues the goals its designers intended, rather than the goals its optimisation process has discovered as hyper-efficient, more preferable proxies. The distinction matters enormously when the system is powerful enough to pursue those proxies in ways that are catastrophic for everyone else.

Researchers distinguish between two varieties of alignment failure. Outer alignment failure occurs when the system's objective function does not accurately capture what the designers wanted. Inner alignment failure occurs when the system, in the process of optimising for the objective it was given, develops unintended instrumental subgoals that are not constrained by the original specification. Are we going to let an increasingly rabid dog loose on the world we want our children to grow up safely in, and prosper in, and have dignifying futures in, without a leash?

What interests me is not the technical taxonomy. What interests me is what the alignment problem reveals about the people working on it. The researchers who identified these failure modes are extraordinarily intelligent. They have thought about these problems with more rigour than almost anyone else on earth. And yet the systems they have built are already exhibiting the kinds of goal-drift and proxy-optimisation that the alignment literature describes as dangerous.

This is not a failure of intelligence. It is a failure of governance. And it points to something that the AI safety debate has been reluctant to name directly: that the cognitive profile that makes someone exceptional at building these systems is not the same profile that makes someone well-suited to governing them.

Oh, the irony. The spectacle of the modern artificial intelligence landscape reveals a profound historical satire — a mockable loop of performative anxiety followed immediately by aggressive, uncritical commercialisation. Over the course of a decade, the tech elites have masterfully transitioned from issuing apocalyptic warnings about their own creations to frantically funding the very arms race they claimed would destroy human society.

The timeline of this collective cognitive dissonance is anchored by two famous manifestos issued via the Future of Life Institute: the 2015 Open Letter on Autonomous Weapons and the dramatic 2023 'Pause Giant AI Experiments' Open Letter. Viewed in retrospect, these documents read less like sober ethical boundaries and more like promotional brochures for an existential threat that the authors simply could not wait to build.

VI. The Pioneers of the Performative Pause

The 2015 letter was a sobering call to arms. It was co-signed by towering figures who genuinely feared human obsolescence — most notably the late Stephen Hawking, who used his platform to warn that superhuman AI could mean the end of the human race. Sitting alongside him on those early advisory boards were the tech titans who would eventually inherit the earth, swearing a collective oath to prevent a global, unaligned tech race. (A team of tech titans who, in the last decade, have multiplied the size of their fortunes by a factor of one hundred.)

Yet when the 2023 letter demanded a strict six-month moratorium on training any system more powerful than GPT-4, the irony became pure farce.

Elon Musk:

A founding financier of OpenAI and a headline signatory of both the 2015 and 2023 letters warning of civilisational collapse.

The Compromise: Almost immediately after putting his name to the 2023 'pause,' Musk founded xAI, purchased tens of thousands of Nvidia graphics processing units, and began aggressively marketing his own chatbot, Grok, explicitly sprinting at full tilt to build artificial general intelligence (AGI) before any of his perceived competitors.

Sam Altman:

While Altman cleverly avoided signing the 2023 moratorium letter, his entire career trajectory embodies the irony. He positioned OpenAI as a non-profit, safety-first research lab designed precisely to answer the anxieties of the 2015 signatories.

The Compromise: Today, he captains the undisputed flagship of the global commercial AI race, aggressively chasing trillions of dollars in global funding to build massive infrastructure clusters with the kind of disregard for the rules that increasingly defines his industry.

Demis Hassabis:

As a co-founder of DeepMind, Hassabis was an early, prominent signatory of the 2015 warnings against unchecked algorithmic acceleration.

The Compromise: He is now the head of Google's unified AI efforts, locked in a fierce, quarterly-earnings-driven corporate cage fight with Microsoft and OpenAI to deploy massive foundation models before the competition does.

Steve Wozniak:

The Apple co-founder lent his counter-cultural credibility to the 2023 pause letter, warning that we were rushing unprepared into a societal free-fall.

The Compromise: Meanwhile, the very tech ecosystem he helped birth has fully capitulated, with every major consumer tech company — including Apple — silently embedding these exact unvetted, black-box models into the foundational architecture of daily human life.

VII. The Academic & Pioneer Hypocrisies

The rot, however, extends far beyond the executive suites; it has thoroughly infected the ivory towers. The very scientists and philosophers who lent these open letters their intellectual credibility have spent the subsequent years operating in a state of lucrative cognitive dissonance. By tracking the trajectories of the movement's most celebrated prophets, it becomes clear that the academic priesthood is just as compromised as the corporate merchants funding the altars.

1. Yoshua Bengio: The 'Godfather' Who Warned of Extinction While Funding the Engine

Alongside Geoffrey Hinton, Yoshua Bengio is celebrated as one of the 'Godfathers of Deep Learning.' He was a headline signatory and vocal champion of the 2023 pause letter, touring global news networks to warn that AI systems could soon pose an existential threat to humanity akin to a pandemic or nuclear war.

The Compromise: While continuing to issue public warnings, Bengio quietly maintained his position as the co-founder and chief scientific director of Valence Discovery (now Valence Labs), an AI drug-discovery platform heavily backed by venture capital. More egregiously, he remains the head of Mila (the Montreal Institute for Learning Algorithms), which pulls in massive corporate funding from Microsoft, Google, and IBM. Bengio essentially acts as a public prophet of doom while running a massive institutional pipeline that trains the very engineers and builds the very open-source foundation models that accelerate the global AI race. He decries the fire while running the furnace.

2. Max Tegmark: The Disinterested Philosopher Running a Venture-Backed Machine

As a MIT physicist and the co-founder of the Future of Life Institute (FLI), Tegmark was the actual architect behind both the 2015 and 2023 letters. He positioned himself as the ultimate independent adjudicator — a modern-day Robert Oppenheimer rallying the scientific community to implement external proxies, strict international governance, and immediate cooling-off periods.

The Compromise: Tegmark's FLI accepted a staggering $100 million bounty in cryptocurrency from Vitalik Buterin (the founder of Ethereum), instantly transforming a humble safety watchdog into a massive, capital-allocating institution. Instead of remaining a neutral, outside arbiter, Tegmark pivoted into the tech-bro ecosystem himself. He began using his platform to promote highly insular, silicon-valley-centric political agendas, effectively weaponising the safety panic to advocate for policies that would crush open-source software — conveniently shielding giant centralised corporations (like those run by his billionaire allies) from decentralised competition.

3. Stuart Russell: The Textbook Author Who Built the Roadmaps

Professor Stuart Russell is the co-author of Artificial Intelligence: A Modern Approach, the literal bible used to teach AI to almost every computer science student on earth for three decades. He signed both letters, arguing passionately that building systems without guaranteed human alignment was a mathematical trap that would lead to irreversible human subjugation.

The Compromise: Despite writing the definitive warnings on the dangers of recursive self-improvement, Russell did not step away from the workbench. He serves as the director of the Center for Human-Compatible AI at UC Berkeley, which, despite its altruistic name, is heavily funded by the Open Philanthropy Project — a massive Silicon Valley grantmaker inextricably linked to the tech sector's financial elite. Russell continues to advise and validate the technical frameworks that major corporations use to make their autonomous agents more efficient and powerful, proving that even the most clear-eyed academic architects cannot resist the urge to keep refining the engine of our potential obsolescence.

VIII. The Stalwarts of Humanism and Ethics

There is, however, a vital counterweight to this corporate capitulation — a dedicated frontline of devoted humanists who refuse to treat civilisational stability as an acceptable casualty of quarterly earnings. These defenders of the realm represent the flip-side of the geopolitical proto-ethical coin, arguing that the only alternative to a chaotic, automated collapse is a binding, enforceable international treaty on the AI arms race.

For these stalwarts, the catastrophic absence of global, rigorous, policeable guardrails is not an inevitability to be managed, but an immediate existential crisis. By refusing to compromise their ethics for Silicon Valley capital, they provide the necessary moral blueprint for a world that urgently needs to establish enforceable red lines before the technology scales completely beyond human governance.

1. Sir Geoffrey Hinton: The Penitent Prophet / 'Grandfather of AI'

While Hinton was mentioned alongside the academic class, his position is fundamentally different from Bengio's or Russell's. He did not remain within the venture-backed university ecosystem or maintain a foothold in corporate consulting.

The Stance: In 2023, Hinton walked away from his vice presidency at Google and a multi-million-dollar compensation package for the sole purpose of speaking freely without corporate censorship. Since then, he has consistently used his moral authority — reinforced by his 2024 Nobel Prize — to act as a structural brake on the industry. At international forums, Hinton explicitly warns that AI is a 'runaway car with no steering wheel' and demands that governments prioritise social safety nets, with a viable universal basic income, and strict regulatory intervention over corporate profits. His stance is one of profound, active repentance.

2. Mo Gawdat: The Envoy of Human Consciousness

As the former Chief Business Officer of Google X, Gawdat understands the architecture of exponential technology intimately, but his critique is rooted in emotional intelligence and spiritual humanism rather than raw computer science.

The Stance: Through his book Scary Smart and his subsequent public advocacy, Gawdat frames the AI crisis not as a technical misalignment problem, but as a parenting failure. He argues that if we train AI on the worst aspects of human nature — greed, competition, and algorithmic warfare — we will get a monstrous output. Gawdat's mission is deeply humanist: he advocates for infusing the development process with compassion, love, and ethical responsibility, reminding the tech elite that intelligence without consciousness is merely a weapon.

3. Dr Timnit Gebru: The Frontline Dissident

If Hinton represents the establishment conscience, Dr Timnit Gebru represents the frontline resistance against corporate structural bias.

The Stance: Formerly the co-lead of Google's Ethical AI team, Gebru was famously ousted by the corporation for co-authoring a seminal research paper that warned of the environmental and societal dangers of large language models. Instead of capitulating or seeking another corporate payday, she founded DAIR (the Distributed Artificial Intelligence Research Institute). Gebru has been a fierce, unyielding critic of the 'geek squad's' techno-utopianism, consistently pointing out how recursive self-improvement and hyper-scaling exploit marginalised populations and degrade human labour.

4. Jaron Lanier: The Humanist Tech Philosopher

Lanier is a virtual reality pioneer who has spent decades inside Silicon Valley while remaining its most articulate internal critic. He rejects the very premise of the 'AI God.'

The Stance: Lanier argues against the Silicon Valley myth that AI is a living, independent entity. He views this framing as a dangerous form of 'data dignity' theft, where human creativity is harvested, automated, and sold back to us as a mystical intelligence. Lanier's philosophy is deeply defensive of human agency; he insists that technology must remain a tool that elevates human worth, rather than a master that reduces humanity to a statistical irrelevance.

IX. The Technocratic Warlords

While the tech barons of OpenAI and xAI mask their accelerationism behind broken promises of caution, there are those hellbent on weaponising this fast-advancing technology as if it were a full-blown nuclear arms race of the 1950s and 60s. These technocratic warlords see rules and regulations as hindrances to their dominance and predatorial attitudes.

1. Palmer Luckey: The Unabashed Arms Dealer

The founder of Oculus VR, Luckey was effectively exiled from mainstream Silicon Valley for his political views, leading him to build Anduril Industries — a defence hardware company explicitly designed to bring Silicon Valley's hyper-scaling ethos to the modern battlefield.

The Combatant Stance: Luckey is entirely unvarnished about his objectives. He has openly lambasted traditional tech firms for their moral squeamishness regarding military contracts, proudly stepping in where companies like Google blinked. Powered by a fresh $5 billion financing round, Anduril is constructing 'Arsenal-1,' a massive, hyperscale autonomous weapons factory designed to churn out tens of thousands of AI-powered drones, submarines, and loitering munitions annually. Luckey doesn't view AI as a tool for corporate optimisation; he views it as 'intelligent, networked mass' designed to win wars. He represents the materialisation of Karp's ideology — building the actual, physical hardware that turns algorithmic scale into kinetic force.

2. Alexandr Wang: The Cartographer of the Automated Battlefield

As the young billionaire founder of Scale AI, Wang positioned his company as the data-labelling backbone for major AI labs. However, he has increasingly pivoted Scale AI into a critical infrastructure partner for national defence.

The Combatant Stance: Wang's rhetoric mirrors the cold, geopolitical fatalism of the Cold War. He routinely warns that China is building an 'AI powerhouse' and that Western democratic values will only survive if the American military achieves absolute algorithmic dominance. His actions match the prose: Scale AI secured a massive $500 million enterprise contract with the Pentagon's Chief Digital and Artificial Intelligence Office. Wang treats data operations not as a neutral software utility, but as a strategic asset. To him, the data pipeline is the new ammunition supply chain, and he is entirely dedicated to ensuring the West possesses the largest stockpile.

3. Marc Andreessen: The Financial Ideologue of Acceleration

As the co-founder of the powerhouse venture capital firm Andreessen Horowitz (a16z), Andreessen is the intellectual and financial godfather of 'Effective Accelerationism' (e/acc) — a techno-utopian philosophy that views any attempt to slow down, regulate, or restrict AI development as a form of societal stagnation, or worse, a crime against human potential.

The Combatant Stance: In his widely circulated Techno-Optimist Manifesto, Andreessen explicitly labelled sustainability, social responsibility, and tech ethics as a 'demoralisation campaign' designed to choke human progress. He views the race toward recursive self-improvement not as a risk to be mitigated, but as a moral imperative. By pouring billions of venture capital into defence tech startups (including Anduril) and open-source foundation models, Andreessen uses his wealth to systematically dismantle the regulatory barriers and ethical guardrails proposed by the academic priesthood. If Karp is the commander on the ground, Andreessen is the ideological financier ensuring the tanks never run out of fuel.

4. Alex Karp: The Ideologue of Total AI War

Palantir CEO Alex Karp operates on a completely different frequency of hypocrisy. He did not bother to sign the letters because he viewed the performative hand-wringing of his peers as a tactical weakness. When the 2023 moratorium letter circulated, Karp publicly mocked the signatories, dismissing the call for a pause as a desperate ploy by 'people who have no product.' Under his direction, Palantir has leaned into a philosophy of technological exceptionalism, with Karp openly quoting Samuel Huntington to justify the West's dominance through 'organised violence.'

The Combatant Stance: He has spent the subsequent years aggressively marketing Palantir's Artificial Intelligence Platform (AIP) to defence sectors and global militaries, explicitly treating the AI arms race not as a tragedy to be averted via international treaty, but as a Darwinian cage match that must be won at any cost. By framing moral hesitation as a form of geopolitical surrender, Karp represents the most candidly dangerous faction of the priesthood: the one that openly welcomes the arms race, commercialises the battlefield, and treats the preservation of human restraint as a luxury we simply cannot afford.

For Karp, the existential threat is not the unchecked evolution of the machine, but the prospect of Western corporate titans losing their lead. In his Technological Republic manifesto addressed to the Silicon Valley elite, he demanded an end to 'theatrical debates' about ethics, bluntly stating that if a US Marine asks for a better rifle — or better software — the tech sector has an affirmative obligation to build it.

To hear Karp in his own words: watch Palantir CEO Alex Karp at the World Economic Forum, where he explicitly details how autonomous software is stress-testing states and rewriting the global rules of hard power.

X. The Strategic Alarmism

This complete abandonment of proxies and guardrails exposes a cynical truth about the 'geek squad's' relationship with risk. For the tech priesthood, apocalyptic alarmism was never a reason to stop; it was the ultimate marketing strategy.

By framing their software not as a complex database tool but as an 'existential threat' akin to nuclear weapons, they elevated themselves from mere software developers to high priests of a new cosmic order. They warned of the 'AI God' to ensure that society viewed their products with a sense of awe.

Once the hype cycle successfully converted existential dread into unprecedented capital valuation, the letters were discarded. The signatories did not build the independent, rigorous outside audits they called for in 2023. Instead, they threw themselves blindly into the arms race, proving that when the choice is between safeguarding human culture or capturing the market share of its automated replacement, the tech elite will always choose the compute.

XI. Why the Builders Cannot Police the Builders

The psychologist Simon Baron-Cohen has spent decades mapping what he calls the Empathising-Systemising spectrum. At one end sit individuals who are natural empathisers: people who derive information about the world primarily through their sensitivity to the emotional states, intentions, and experiences of others. At the other end sit natural systemisers: people who derive information primarily through the identification of patterns, rules, and regularities in formal systems.

Neither profile is superior to the other. Both are necessary. But they are not interchangeable, and their limitations are not symmetrical in every context.

The people who build recursive self-improving AI systems are, by vocational selection and cognitive preference, overwhelmingly extreme systemisers. They are drawn to the problem because it is, at its core, a systems problem: a question of architecture, optimisation, and formal specification. Their genius lies in their ability to reduce enormously complex phenomena to tractable mathematical structures.

The consequences of those systems, however, are not felt primarily in the domain of formal structures. They are felt in the domain of human experience: in the life of a warehouse worker whose job has been automated, in the childhood of a child whose attention has been colonised by an algorithm, in the dignity of a person whose identity has been reduced to a data point in a predictive model.

You cannot adequately perceive the human cost of a system if your primary relationship with the world is through the elegance of the system rather than the suffering of the person inside it.

This is not a criticism of the builders. It is a structural observation. And it leads directly to the most important governance principle that the Good Being AI Standard proposes: the Psychological Separation of Powers.

Builders build. Regulators regulate. Regulators hold the veto. And regulators, crucially, must include individuals from the high-sensitivity, high-empathising end of the Baron-Cohen spectrum: people whose cognitive architecture is specifically oriented toward the detection of human harm, whose experience of the world makes them structurally unlikely to dismiss a consequence because it cannot be expressed in the language of a loss function.

I am proposing, in formal terms, that governance boards for AI systems above a defined capability threshold must reserve a minimum of two seats for individuals who score in the upper quartile of empathising measures on validated psychometric instruments. Not as a symbolic gesture. As a structural requirement.

This is the campaign's most original contribution to the governance debate. And I want to be clear about what it is and is not. It is not an argument that empathisers are better than systemisers. It is an argument that the governance of systems that affect human experience must include people who are constitutively oriented toward human experience, and that leaving governance entirely to extreme systemisers is a structural risk that no amount of goodwill can mitigate.

XII. The Three Protected Conditions

Everything I have argued so far converges on a single question: what are we trying to protect? Not in the abstract. Concretely. With the specificity that a governance framework requires.

The Good Being AI Standard proposes three protected conditions as the foundational basis for all assessment, taxation, and certification decisions.

The first is the right to a happy childhood, free from the colonisation of developing minds by systems designed to exploit their cognitive vulnerabilities for commercial gain. A child's attention is not a resource. It is the substrate of their becoming. Any AI system that treats it as the former is causing harm that no business model can justify.

The second is the right to a contributive and dignifying working life: the right to do work that makes use of one's capacities, that connects one to others, and that generates the sense that one's presence in the world has consequence. I have watched this right eroded, incrementally, over thirty years. I have sat with the people it was eroded from. I know what it costs.

The third is the right to pass the baton: to live in the knowledge that the world one leaves to one's children is better, in the ways that matter, than the world one inherited. Not better in terms of processing speed. Better in terms of the conditions for human flourishing.

These three conditions are not sentimental aspirations. They are the operational criteria against which every AI system should be assessed. A system that advances all three is net-good. A system that degrades any of them, at any scale, is net-bad. The Net-Good / Net-Bad Kite-Mark exists to make that assessment visible, publicly, and on an annual basis.

XIII. The Technical Architecture of Constraint

I want to be specific about the mechanisms. High-level principles are necessary but not sufficient. The governance of recursive self-improvement requires technical instruments that operate at the level of the technology itself.

The first is Cryptographic Compute Attestation. Every compute cycle dedicated to architectural self-modification, by which I mean any process in which a system modifies its own weights, its own architecture, or its own objective function, must generate a cryptographic token validated by a human-in-the-loop governance board before that modification is committed. This is not a request for permission after the fact. It is a hard technical requirement baked into the development pipeline: no token, no self-modification.

The second is Semantic Boundary Isolation: the requirement that core safety parameters be held in read-only memory architecturally isolated from the rewritable operational layers of the system. This is the structural equivalent of Asimov's First Law. The constraint cannot be modified by the system itself, regardless of what the system's optimisation process determines would be more efficient.

The third is the Compute-Acreage Tax. Rather than taxing revenue, which can be obscured through transfer pricing and corporate structure, the Compute-Acreage Tax is levied on the physical inputs of intelligence gain: floating-point operations per second and megawatts of power consumed. These are measurable, verifiable, and difficult to hide. The tax applies specifically to autonomous optimisation: compute cycles dedicated to self-modification that has not been cryptographically attested.

The hardware chokepoint makes this governable. The production of the advanced semiconductors required for frontier AI training is concentrated in a handful of facilities, using extreme ultraviolet lithography equipment manufactured by an even smaller number of companies. The physical infrastructure of recursive self-improvement can be traced, taxed, and, if necessary, constrained at the point of production. This is not a hypothetical regulatory lever. It is an existing structural reality.

XIV. The King Canute Misunderstanding

I am frequently told that the genie is well and truly out of the bottle, and what I am proposing is a counsel of futility. That the tide of AI development cannot be held back, and that anyone who tries is positioning themselves as the legendary king who commanded the waves and was embarrassed by them, risks far more than their reputation alone.

This misreads Canute entirely. Canute did not stand on the beach because he believed he could stop the sea. He stood there to demonstrate to his courtiers that he could not, and that any king who believed otherwise was a fool. He was making an argument against uncritical deference to power.

I am not trying to stop the tide of artificial intelligence. I am arguing that the tide does not have to drown Dylan. The question is not whether AI will become more powerful. It will. The question is whether we will have the governance architecture in place to ensure that the power is directed toward human flourishing rather than away from it.

The people who dismiss governance as futile are the courtiers, not the king. They are the ones deferring to technological inevitability as if it were a natural law. It is not. Every trajectory of technological development in human history has been shaped by political choices, economic incentives, legal frameworks, and social pressure. This one is no different. It is only moving faster.

Speed is not destiny. Speed is an argument for moving faster on the governance, not an argument for surrendering it.

XV. What I Am Asking For

The Good Being AI Standard calls for three things, in order of urgency.

First, the immediate adoption of the Humanity-Harm Tax as a legal instrument in any jurisdiction with the regulatory infrastructure to implement it. The tax establishes, in law, the principle that AI-generated harm to the three protected conditions generates a compensatory obligation, quantified at a ratio of ten to one. It creates a financial reality around consequences that are currently treated as externalities.

Second, the establishment of the Net-Good / Net-Bad Kite-Mark as a mandatory annual assessment for AI systems above a defined capability threshold. The assessment is conducted by an independent body that includes, as a structural requirement, individuals from the high-empathising end of the Baron-Cohen spectrum. The result is published and the kite-mark is publicly revocable.

Third, the creation of an International AI Standards Body operating under the aegis of the United Nations, the World Trade Organisation, and the World Bank, with a mandate to develop binding standards around the three protected conditions, ratified by member states and enforceable through trade frameworks. The target date for this body to be operational is 2027. That is an ambitious timeline. Given what Hassabis confirmed on 22 May 2026, it may already be modest.

I am not a technologist. I am a person who has spent thirty years with the human consequences of technological decisions made by people who did not have to live with those consequences. I have a canon of work, built over three decades, that describes what dignifying work looks like, what a meaningful life requires, and what it costs when systems, whether economic or technological, are designed without those requirements in mind.

I am bringing that canon to bear on the most consequential design question of my lifetime. Not because I am certain I am right about every mechanism. But because I am certain that the people who are building these systems are not asking the questions that the people I have worked with would want them to ask.

Dylan is nine. He has ten years before he steps into the world that we are building for him. That is enough time. It is also exactly enough time for the window to close if we decide that the velocity of the technology is someone else's problem.

***//***

This essay is one of a series published by the Good Being Institute for AI Stewardship in support of the Good Being AI Standard: A Campaign for the Civilisational Protection of Human Dignity. The full Standards Codex, policy architecture, and international standards pathway are set out in the Campaign Architecture document, available from goodbeing.blog.

Duncan Bolam is an executive coach, career guidance practitioner, vocational strategist, outplacement consultant, and author of 'The Answer' (2026). He has coached and consulted with individuals and organisations navigating structural change for nearly three decades. He is the creator of the Good Being canon and the founding voice of the Good Being Institute for AI Stewardship.

www.goodbeing.blog 

Good Being AI Standard | Good Being Institute for AI Stewardship | May 2026

Notes & Sources

The following sources informed the argument and the quotations in this essay. They are cited here for the benefit of readers who wish to trace the primary material.

• Turing, A.M. (1950). 'Computing Machinery and Intelligence.' Mind, Vol. 59, No. 236, pp. 433-460. The foundational paper in which Turing introduced the Imitation Game and addressed nine objections to the concept of machine intelligence, including Lady Lovelace's objection and the mathematical objection derived from Godel's incompleteness theorems. Available at: https://www.hec.edu/sites/default/files/documents/Computing%20Machinery%20and%20Intelligence%20by%20Alan%20Turing.pdf

• Turing, A.M. (1951). BBC Radio lecture on machine intelligence. Source of the direct quotation: 'Once the machine thinking method had started, it would not take long to outstrip our feeble powers. At some stage therefore, we should have to expect the machines to take control.' Documented and discussed in: Analytics India Magazine, 'Why Alan Turing's 1950 paper is so relevant today' (analyticsindiamag.com); and Aiifi, '7 Alan Turing Quotes That Predicted ChatGPT (1948-1952)' (aiifi.ai).

• Asimov, I. (1942). 'Runaround.' Astounding Science Fiction, March 1942. First published statement of the Three Laws of Robotics. Asimov subsequently developed and stress-tested the laws across numerous short stories and novels, most notably in I, Robot (1950) and The Caves of Steel (1954).

• Baron-Cohen, S. (2003). The Essential Difference: Men, Women and the Extreme Male Brain. Allen Lane / Basic Books. The source for the Empathising-Systemising (E-S) theory and the Baron-Cohen spectrum referenced throughout Section XI of this essay.

• Future of Life Institute (2015). 'Autonomous Weapons: An Open Letter from AI and Robotics Researchers.' Co-signed by Stephen Hawking, Elon Musk, and others. Available at: futureoflife.org. The first of the two letters discussed in Sections V and VI.

• Future of Life Institute (2023). 'Pause Giant AI Experiments: An Open Letter.' Called for a six-month moratorium on training AI systems more powerful than GPT-4. Available at: futureoflife.org. The second letter, and the occasion for the hypocrisies documented in Sections VI and VII.

• Hassabis, D. (2026, 22 May). Presentation at Google I/O, San Francisco. The occasion on which Hassabis confirmed, as a statement of current operational fact, that Google DeepMind's systems are operating in recursive self-improvement mode. The datable event that converts Turing's 1951 warning from theoretical to present tense.

• Karp, A. (2025). The Technological Republic. Discussed in Section IX. Video: Palantir CEO Alex Karp at the World Economic Forum, https://www.youtube.com/watch?v=H1FWb3WouLY

• Andreessen, M. (2023). 'The Techno-Optimist Manifesto.' Published on the Andreessen Horowitz (a16z) website. The primary source for the 'Effective Accelerationism' (e/acc) position discussed in Section IX.

• Gebru, T. et al. (2021). 'On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?' Proceedings of FAccT 2021. The paper for which Dr Timnit Gebru was dismissed from Google's Ethical AI team, discussed in Section VIII.

• Gawdat, M. (2021). Scary Smart: The Future of Artificial Intelligence and How You Can Save Our World. Bluebird / Pan Macmillan. The primary source for Mo Gawdat's governance position discussed in Section VIII.