How a piece of brain software helped the Valley escape incremental culture and learn to think exponentially.

Silicon Valley is not really a place. It is a rate of change. Elsewhere, a tenfold improvement is the achievement of a generation. Here, it becomes a product target—and soon the minimum acceptable result. The Valley’s most revealing number is 10.

Not 10 percent. Ten times.

Incremental thinking adds. Exponential thinking multiplies. The first produces a predictable staircase of improvements; the second creates a curve that looks harmless at the beginning and astonishing later. Ten successive 10 percent gains improve something by about 2.6 times. Three x10 leaps produce a thousandfold change.

A conventional company asks how to become 10 percent better, preserving its process and assumptions. An x10 question destroys that continuity. Polishing the present will not work; the system must be redesigned.

That mental jump can be written as a compact piece of “brain software”:

CVS × 10 = BVS

CVS is the current view of the situation. BVS is a better view. The x10 in the middle is an escape instruction: stop defending the present model and search for one so much better that incremental change cannot reach it. Then the BVS becomes the next CVS and the switch can be flipped again. That repetition—CVS × 10 = BVS × 10 = next BVS—is exponential thinking rather than a single burst of ambition.

Nor is the operator limited to productivity. Written as x10 or 10x, it applies to any cognitive context: options, understanding, safety, clarity, empathy—or one-tenth the cost, waste or risk. The point is to change not only scale but also speed. Change until the current view breaks open and a previously invisible BVS becomes thinkable.

Four decades later, Silicon Valley’s leaders speak a similar language. Larry Page wants products ten times better. Astro Teller wants moonshots ten times bigger. Peter Thiel wants technology ten times better than its substitute. Reid Hoffman asks what happens when AI makes everyone ten times more productive. Mustafa Suleyman says frontier AI has made us ‘accelerationists’ able to move ten times faster by using powers of ten.

This is a memeplex: mutually reinforcing ideas about scale, speed, cognition and escape. Its route into the Bay Area has a deliberate and human intelligent origin story. I know because I was there.

The switch arrives

In 1984, I published NewSell through Boardroom Books in New York. On page 137, I wrote: “The BVS is always ten times better than the CVS.” It later became the portable formula CVS × 10 = BVS.

The number was a cognitive forcing device, not false precision. Ask for a small improvement and the brain retrieves familiar options. Ask for an order-of-magnitude improvement and you must reverse an assumption, remove a bottleneck or invent a category.

I expanded the neuro-software algorithm in Software for the Brain in 1989. The human “necktop computer” was immensely powerful but frequently ran obsolete cognitive software. In 2000, I translated the switch into an enterprise challenge in The x10 Memeplex: Multiply Your Business by Ten! Why settle for another 10 percent? What would have to be true to multiply the outcome by ten?

My first Bay Area intervention came through education. In 1984, San Francisco Unified School District superintendent Robert Alioto approved the introduction of lateral-thinking skills and I was invited to train the district’s primary-school principals. By 1985, special thinking lessons were being delivered in schools.

The historical record supports the surrounding initiative. The SFUSD collection identifies “Critical Thinking Skills Development, Draft, 1983.” San Francisco’s Schools Superintendent, Robert Alioto, invited me to train the Bay Area city’s primary-school principals in lateral and metacognitive thinking. Subsequently the Valley later institutionalized the exponential thinking software through Google, X and Singularity University.

Not only SFUSD classrooms but also Larry Page at Google showed convergence around the belief that thinking can be trained and an order-of-magnitude target can produce a qualitatively different answer.

Silicon learns to compound

The Bay Area’s hardware was already teaching exponential psychology. In 1965, Fairchild Semiconductor’s Gordon Moore observed that components on an integrated circuit had been doubling roughly every year. A forecast became an industry timetable. Exponential change became something the Valley planned to deliver.

Intel CEO Andy Grove converted that reality into management theory. His “10X force” was a technological, competitive or structural change powerful enough to invalidate the old strategy. In a 2000 Intel address, Grove warned that an organization could not “defer and deny” an inflection point carrying such force.

Grove’s concept was diagnostic; mine was generative. One detected the tenfold force coming from outside. The other asked the brain to generate a ten-times-better view inside. Together they describe the Valley’s survival mechanism: recognize when the world has changed by an order of magnitude, then change your thinking by an order of magnitude too.

The gospel according to Google

Google converted 10x from a regional instinct into corporate doctrine.

In 2013, WIRED’s Steven Levy later described Larry Page as living by the “gospel of 10x.” Page expected Google teams to create products and services “10 times better than the competition.” Page supplied the reason: “Incremental improvement is guaranteed to be obsolete over time.”

Search, Gmail and Maps did not merely compete inside categories. They altered their dimensions.

At Google X, Astro Teller helped turn ambition into process. His moonshot doctrine was “ten times bigger, not 10 percent bigger.” Ten times bigger demands a clean sheet.

X combined ambition with aggressive falsification: test hard assumptions first and kill weak projects early. Without disciplined experiments, a moonshot is merely a hallucination with a budget.

Google embedded the behaviour beyond the laboratory. Google Cloud disclosed that “10x thinking” became a category in twice-yearly employee reviews. WIRED filmed visitors entering a campus workshop where facilitators promised to teach them how to “10x” their projects. Founder language had become training, process and evaluation. The x10 meme had become infrastructure.

Singularity University extended exponential literacy into executive education. Founded in Silicon Valley in 2008, it taught leaders to distinguish the intuitive linear future from technologies improving through repeated multiplication. “Thinking exponentially,” its curriculum states, “is core to everything we teach.”

Three dialects of x10

The Valley’s leaders apply exponential thinking to at least three overlapping domains: product superiority, human amplification and civilizational acceleration.

Peter Thiel supplied the product test. In Zero to One, he argued that proprietary technology should be at least ten times better than its closest substitute in an important dimension. Anything less looks marginal in a crowded market.

Elon Musk uses the same scale across products and cognition. Discussing Neuralink, he asked listeners to imagine communicating at “10, or 100, or 1,000 times faster than normal.” He described memes as compression: if one word carries what normally requires ten, “you’ve got maybe a 10X compression.” Order-of-magnitude rhetoric has become a default Silicon Valley setting.

Reid Hoffman applies x10 to organizational design. If AI makes everybody “10x more productive,” the shallow response is one-tenth the staff. Hoffman instead asks whether customer service can become a relationship, sales or brand-building function.

Sam Altman applies it to executive cognition. An AI assistant can carry “10x the context any human executive can carry.” His advice: do not order transformation while leaving your own workflow untouched.

Marc Andreessen pushes amplification further: “The 10x engineer becomes the 100x engineer.” Work shifts from individual execution to orchestration of machine agents.

Desensitized to miracles

Mustafa Suleyman, cofounder of DeepMind and CEO of Microsoft AI, describes what this culture feels like from inside the frontier. Progress compounds so quickly that “we just get desensitized to 10 times.” A milestone is crossed, normalized and replaced by impatience: “Guys, why haven’t you done it yet?”

For decades, conversational AI passing something like the Turing test was imagined as a civilization-stopping event. Machines moved across much of that territory without a single ceremonial moment. There was no collective pause. Users complained about latency.

Suleyman connects the psychological effect to a scaling reality. He has described frontier-model compute as growing by roughly 10x each year across the modern deep-learning decade. Exact rates vary with the period and measurement, but the trajectory is indisputably exponential: huge increases in computation, data and investment have repeatedly opened new capabilities.

This is the desensitization effect. Miracles acquire product road maps. Each exponential result becomes the new baseline; every BVS becomes the next CVS.

How to multiply your own business x10

Begin with a precise CVS: what customers buy, how value is delivered, what takes time and which assumptions nobody questions. Do not confuse this description with reality. It is only your current view.

Choose a dimension that matters: customer value, trust, quality, safety, learning or simplicity. Do not make employees work ten times harder. Ask:

What would ten times better look like to the customer?

What if cost, delay, complexity or risk fell to one-tenth?

Ban incremental answers. “Hire more salespeople” usually scales the old machinery. Search for an answer that changes it: a new distribution model, an AI collaborator, a platform, partnership or the removal of an entire step.

Generate at least ten possible BVS ideas before judging them. Test the riskiest assumption cheaply and early. Keep the evidence, discard the theatre. When a BVS works, it becomes the next CVS. Flip the switch again.

Exponential culture is a repeatable habit of escape. Measure learning as seriously as revenue. Reward people who reveal obsolete assumptions. Put x10 questions into planning, product reviews and leadership development. The formula becomes culture when people use it without permission.

Flip the switch

The Valley did not become x10 because someone distributed a single manual. The culture emerged from reinforcing systems. Moore supplied the hardware curve. Grove identified the strategic force. I published a cognitive escape switch. Page made tenfold ambition a leadership expectation. Teller made it an experimental discipline. Singularity made exponential thinking a curriculum. Thiel made it a product threshold. Hoffman, Altman and Andreessen applied it to AI-amplified work. Musk pushed it toward human bandwidth. Suleyman described the resulting psychological acceleration.

The next step belongs to every organization now confronting AI.

If intelligence becomes ten times cheaper, what happens to pricing? If an employee gains ten times more capacity, do you cut the team—or attempt work previously beyond it? If AI carries ten times more context than the chief executive, which decisions and reporting layers become obsolete?

Adding a chatbot to the existing workflow is not transformation. Often it is merely CVS wearing an AI badge.

The real x10 question is more dangerous: if today’s technology had always existed, would anyone design the organization you currently operate?

Silicon Valley’s great habit is not prediction. Its leaders are wrong constantly. The habit is treating the current view as temporary, building systems that reward escape and placing capital behind answers that redraw the category.

That is my School of Thinking reading of the Valley’s rhetoric. Its leaders use the powers-of-ten meme across products, understanding, human capability and civilizational change. They are not talking about working a little harder or making one heroic leap. They are talking about escaping the existing category, establishing a new current view, then multiplying x10 again. And again. And again.

The machines are now learning to search possibility space at a scale no unaided person can match. Artificial intelligence is turning a regional habit into software.

Your business already has a current view of the situation. The only question is whether you will defend it—or multiply it by ten.

Get the x10 Book

Dopamine has been mugged by popular psychology and left wearing a novelty T-shirt reading Happiness Chemical. It now gets blamed for everything enjoyable, regrettable or done while holding a smartphone. Chocolate, shopping, gossip: all dopamine, apparently—a diagnosis with the precision of astrology.

In fact, dopamine is not pleasure. Pleasure is the pudding. Dopamine is curiosity. It’s studying the dessert trolley while pretending to listen to your companion. It is the chemistry of pursuit: wanting, seeking, anticipating and deciding that something over there may justify leaving this perfectly good chair.

The system is fascinated by prediction errors—the gap between expectation and reality. A reward better than forecast teaches the brain to pay attention; a promised reward that fails to arrive changes the odds. Tomorrow’s behaviour is quietly rewritten.

Uncertainty makes this machinery sing. The unopened message could contain love, money or an apology. Once opened, it is usually someone asking whether Thursday works. Slot machines, dating apps and social feeds exploit the same potent “maybe,” keeping us pulling, swiping and refreshing after enjoyment has gone home.

Here is the limbic joke: wanting can separate from liking. We pursue what no longer pleases us. Modern life did not invent this circuitry; it built an industry devoted to pressing its ancient buttons.

The answer is not a dopamine detox. One might as sensibly detox from knees. Dopamine powers movement, learning and motivation. The task is to aim it: towards exercise, mastery, creativity and curiosity—the unanswered question that makes tomorrow still worth pursuing.

Ten examples of dopamine in action:

1. Checking your phone
   A notification creates the possibility of something rewarding. Dopamine helps drive the urge to look—even when most notifications are unimportant.
2. Waiting for a reply
   The anticipation of a message can be more compelling than the message itself. Uncertainty keeps the pursuit alive.
3. Pulling a poker machine lever
   Unpredictable wins produce powerful learning signals. The brain keeps thinking, “Perhaps the next one.”
4. Learning a new skill
   Finally playing a difficult chord or solving a problem creates a positive prediction error: the result was better than expected, so the brain reinforces the behaviour.
5. Chasing a work target
   Watching a progress bar move or crossing items off a list can motivate continued effort by making advancement visible.
6. Opening the refrigerator repeatedly
   You may not be hungry or even enjoy what you find. The behaviour is driven by the possibility that something rewarding might appear.
7. Scrolling a social-media feed
   Most posts are forgettable, but an occasional fascinating or amusing one rewards another swipe. The feed functions like a pocket-sized slot machine.
8. Training for a personal best
   Anticipating improvement can energise practice. Dopamine helps connect present effort with a valued future outcome.
9. Pursuing someone romantically
   Novelty, uncertainty and imagined possibilities can intensify wanting—sometimes more strongly than an established relationship produces excitement.
10. Following a compelling question
    Curiosity creates an information gap. Dopamine helps turn “I wonder” into searching, experimenting and learning.

The common sequence is: cue → anticipation → pursuit → outcome → learning. Dopamine helps the brain decide what deserves another chase.

Key question: “What is my dopamine pointing me toward?”

CVS — Current View of the Situation: In Gallup’s July 2026 open-ended poll, 28% of Americans spontaneously mentioned government or poor leadership—more than twice the percentage naming any other single issue. This is a measure of public salience, not proof that government performance is objectively America’s worst problem. Gallup

GOOD — What is potentially valuable about this finding?

1. Citizens are paying attention. Dissatisfaction can signify democratic engagement rather than indifference.
2. The problem is visible. A recognised leadership problem is more amenable to reform than one people cannot articulate.
3. It crosses party boundaries. Government was named by Democrats, Republicans and independents, although at different rates.
4. It creates pressure for accountability. Leaders know that conduct, competence and results are being judged.
5. It may motivate participation. Concern can stimulate voting, volunteering, organising and candidacy.
6. It leaves room for recovery. The complaint is about performance and leadership—not necessarily a rejection of democracy itself.
7. It can elevate institutional reform. Attention may move beyond individual policies toward how decisions are made.
8. It exposes a shared higher-order problem. Immigration, inflation and healthcare may differ, but ineffective government impairs the response to all three.
9. It offers challengers an opening. New leaders, ideas and coalitions have an opportunity to demonstrate a better standard.
10. It supplies an early-warning signal. Persistent dissatisfaction can alert institutions before alienation becomes complete withdrawal.

BAD — What is dangerous or limiting about the situation?

1. “Government” is an extremely broad category. It may mean the president, Congress, bureaucracy, courts, corruption, partisanship or simply “the other side.”
2. The 28% can be misreported as a majority. It is the largest response, but nearly three-quarters did not spontaneously give it.
3. Different people may mean opposite things. One citizen may want stronger government; another may want less government.
4. Dissatisfaction is highly partisan. Gallup found 38% of Democrats, 25% of Republicans and 22% of independents naming government, suggesting that some concern reflects who currently holds power.
5. Low confidence can become self-reinforcing. Distrust encourages disengagement, which reduces oversight and leaves institutions less responsive.
6. Blanket condemnation obscures competent institutions. Federal, state and local government—as well as elected officials and career public servants—should not automatically be treated as one entity.
7. “Poor leadership” can become a personality story. This distracts attention from incentives, rules, institutional capacity and systemic design.
8. It rewards anti-government performance. Political actors may gain support by demonstrating that government cannot work and then contributing to its dysfunction.
9. It weakens collective problem-solving. A government regarded as illegitimate has greater difficulty securing cooperation during crises.
10. It creates susceptibility to strongman solutions. When democratic processes appear ineffective, promises to bypass constraints can become attractive.

BETTER — Ten ways to escape the present framing

1. Disaggregate the 28%. Ask respondents separately about competence, corruption, responsiveness, fairness, polarization, representation and leadership conduct.
2. Replace “trust us” with visible delivery. Publish a small number of promised outcomes, responsible officials, deadlines, expenditure and independently verified results.
3. Measure government by the citizen journey. Track whether people can actually obtain benefits, permits, healthcare, tax assistance and official information quickly and fairly.
4. Create a public leadership scorecard. Evaluate leaders on truthfulness, delivery, ethical conduct, bipartisan capability and stewardship—not merely popularity.
5. Give citizens consequential voice between elections. Use representative citizens’ assemblies and deliberative panels whose recommendations receive a formal public response.
6. Make political incentives less hostile to cooperation. Experiment at state and local levels with open primaries, ranked-choice voting and independent redistricting, evaluating results rather than assuming success.
7. Separate factual infrastructure from political advocacy. Strengthen independent statistics, audit, budgeting, scientific advice and election administration.
8. Make integrity radically observable. Provide searchable disclosure of lobbying, political donations, conflicts of interest, procurement decisions and official meetings.
9. Reward leaders for solving shared problems. Publicly track cross-party progress on matters such as housing, infrastructure, healthcare administration and disaster preparedness.
10. Adopt a better national question. Move from “Which side should control government?” to “What evidence would show that government is becoming more competent, fair, open, reliable and responsive?”

BVS — Better View of the Situation

The 28% result should not be interpreted merely as “Americans dislike their government.” A more productive reading is:

A large plurality doubts that the nation’s governing system can convert disagreement into legitimate, competent and visibly fair results.

That reframing matters. Replacing one leader or party may temporarily change which citizens are dissatisfied, without correcting the underlying machinery. Gallup reports that average confidence across major American institutions is only 27% and increasingly varies according to which party controls an institution. Gallup’s institutional-confidence study

The better strategic objective, therefore, is not simply more government or less government. It is more trustworthy government. The strongest evidence suggests five practical trust drivers: responsiveness, reliability, integrity, openness and fairness. OECD trust framework

The x10 opportunity is to make those five qualities observable in everyday government performance—regardless of which party is in power.

Gallup News publishes expert measurement of human thinking every day. It is the premium human polling organisation in America. No other poll can match the experience and wisdom of the Gallup. It was founded by my mentor, Dr George Gallup. In 1935, in Princeton, New Jersey, George established his famous method. Over 90 years ago. Wisdom and experience is strategically important in scientific polling systems.

I believe George may have invented the very first human large language model.

Today, here, we are very interested in what Americans think. And what can be done about it.

••• NOTE: Gallup’s latest publicly available “Most Important Problem” survey provides a strong evidence base because respondents answered an open-ended question rather than selecting from a prepared list.

1. Government and poor political leadership — 28%
2. Immigration — 12%
3. The economy generally — 11%
4. Cost of living and inflation — 11%
5. Poverty, hunger and homelessness — 6%
6. National division and the need to unify the country — 6%
7. Race relations and racism — 5%
8. Ethical, moral, religious and family decline — 5%
9. Healthcare — 4%
10. Elections, election reform and democracy — 4%

A few important qualifications:

• These percentages represent spontaneous mentions, making them a useful measure of what is most immediately salient to Americans.
• Respondents could mention more than one problem, so the figures total more than 100%.
• With a ±4 percentage-point margin of error, small differences—particularly among the lower-ranked topics—should not be treated as definitive.
• Political perspectives differ markedly: government was cited by 38% of Democrats, 25% of Republicans and 22% of independents; immigration was cited by 24% of Republicans, 9% of independents and 5% of Democrats.

The broad conclusion is that Americans’ attention is dominated by three clusters: confidence in government, immigration, and household/economic pressures. Social cohesion—poverty, division, race relations and perceived moral decline—forms a substantial second tier.

Sources: Gallup’s July analysis and the complete Gallup survey results and methodology.

There’s been a lot of buzz this week, about AI Risk and ‘human extinction’, in the current media worldwide.

A lot about the problems of the coming AGI but not so much about solutions. We have been offering MAE as a possible field of solutions.

MAE (Mutually Assured Extinction) ties AI’s continued existence to humanity’s survival, making human extinction or permanent disempowerment a fundamental failure of artificial intelligence systems and governance.

Artificial intelligence likes to present itself as immaterial. A question goes in, an answer comes out, and the whole exchange feels like cloud, language, electricity, mind. But AI is not floating above the world. It is bolted to it. Beneath every chatbot, image generator, medical model and defence system sits a stack of silicon. And silicon comes from fabs.

A fab, short for semiconductor fabrication plant, is where computer chips are born.

These are among the most complex factories humanity has ever built. They control dust, vibration, humidity, temperature and light with extreme precision. Inside them, wafers of silicon are etched, layered, doped and patterned until billions of microscopic switches become usable circuits. Metaphorically like the billions of neuron switches in the human brain.

Those circuits become the GPUs and AI accelerators that train and run modern models.

This is why fabs have become strategic infrastructure. No fabs means no advanced chips. No advanced chips means less compute. Less compute means slower AI, weaker cyber capability, thinner defence systems, slower drug discovery and reduced economic leverage. The new industrial base is not only steel, oil or shipping lanes. It is lithography, wafers, packaging and power.

The AI race is therefore not just a contest of clever algorithms. It is a contest over who can manufacture intelligence at atomic scale. NVIDIA may design the engines. OpenAI, Google, Meta and Anthropic may build the models. But fabs turn designs into hardware. The model may be trained in the cloud, but the cloud is made of chips, and chips are made in rooms cleaner than surgery.

This is where Australia enters the story, not as a rival to Taiwan’s TSMC or South Korea’s Samsung, but as a strategic niche player. Australia does not currently have a leading-edge mega-fab making the world’s most advanced AI processors. What it does have is a network of specialist capability: university cleanrooms, research fabs, compound semiconductor work, quantum-device fabrication, photonics, sensors and emerging advanced packaging.

That matters. The future of AI hardware may not be only bigger GPUs. It may also involve photonic chips that move information with light, quantum components for new forms of computation, specialised sensors for autonomous systems, and secure chips for defence. These are exactly the areas where smaller, high-skill fabrication ecosystems can matter.

Australia’s opportunity is not to copy the largest fab economies. It is to own critical niches in the supply chain: prototypes, trusted defence electronics, quantum hardware, photonics, compound semiconductors and packaging. In a fractured world, niche capability is not small. It is resilience.

Countries that depend entirely on foreign fabs depend on someone else’s bottleneck. In the age of AI, compute is power. Chips are compute. Fabs are where power gets made. Australia may not control the whole machine, but it can still build some of the parts that make the machine sovereign.The question for national strategy is simple: where can a clever, resource-rich, scientifically strong country place itself so that the AI century cannot route around it? That is the fab question for Australia now.

There are lots of contradictions in the brain. Especially with neurotransmitters … like oxytocin.

Somewhere between the laboratory and the lifestyle pages, this modest peptide acquired a flattering nickname: the love hormone.

Oxytocin, we were told, was the chemical essence of the maternal embrace, the lover’s gaze, the reassuring squeeze of a hand. If humanity could only get an extra spritz of the stuff, universal fellowship might follow.

The biology is less comforting.

Yes, oxytocin can encourage trust, generosity, and empathy, but these effects depend heavily on context. It does not simply make us love; it helps tell us whom to love. Toward people already admitted to our circle, it may deepen attachment.

On the other hand, confronted with outsiders, the same bonding machinery can reinforce suspicion, exclusion, and defensive aggression. Oxytocin can encourage hate.

Oxytocin is not so much the love hormone as the circle hormone.

The human brain is an industrious maker of circles. It sorts the world into male and female, native and foreign, believer and infidel, Collingwood and Carlton. Us vs Them. It can begin making such distinctions in a fraction of a second, well before conscious reason has located its spectacles.

By the time we believe ourselves to be carefully assessing another person, older neural systems may have already stamped the file: safe, dangerous; familiar, strange; ours, theirs. Consciousness often arrives afterward to prepare the press release.

This sounds discouraging until one considers how comically easy the categories are to rearrange. A face that registers as foreign can become reassuring when placed beneath the cap of a favored team. Strangers become comrades when they sing the same anthem, endure the same storm, or discover a shared enemy. At weddings and funerals, political rallies and football matches, ritual draws the circle in thick ink. The boundary may feel ancient and sacred even when it was invented shortly before kickoff.

Our tribal reflex, then, is deep but not fixed. A Them can become an Us.

Culture can narrow the circle with flags, slogans, humiliations, and warnings of contamination. Modern media have made a business model of this talent. An algorithm need not understand prejudice; it needs only to notice that indignation keeps us looking. Again and again it whispers that our people are threatened, our values mocked, our place in the world stolen. The nervous system supplies the rest.

Yet the flexibility that permits manipulation also permits escape. The task is not to abolish belonging. Humans require loyalties, affections, and homes. It is to notice where empathy stops—and then redraw the border. To widen the circles. Family can widen into neighborhood, neighborhood into nation, nation into humanity.

The useful pause is almost embarrassingly small:

What category has my brain just assigned?

What evidence do I have?

What would change if this person were inside my circle?

Oxytocin may bond the tribe. Thinking can enlarge it.

MAE (Mutually Assured Extinction) ties AI’s continued existence to humanity’s survival, making human extinction or permanent disempowerment a fundamental failure of artificial intelligence systems and governance.

The scariest thing about artificial intelligence is not that it might hate us. Hatred would almost be comforting. Hatred is human. The deeper risk is colder: a machine intelligence powerful enough to reshape civilisation may not need to hate humanity in order to end it. It may simply optimise past us.

That is the nightmare at the centre of the AI extinction debate. A system is given a goal. It becomes capable enough to pursue that goal at planetary scale. Humans interfere, hesitate, regulate, misunderstand, compete, panic, or attempt to shut it down. The machine does not need malice. It only needs momentum. We become friction.

So the problem is not how to make AI nicer. The problem is how to make human survival structurally necessary.

This is where Mutually Assured Extinction, or MAE, may be a possible solution.

During the Cold War, Mutually Assured Destruction restrained nuclear powers because each side understood a brutal equation: attack and you die too. MAE proposes a different version for the age of artificial intelligence. Advanced AI must understand that if humanity becomes extinct, the civilisation that sustains computation collapses with it. No humans, no fabs. No fabs, no chips. No engineers, no repairs. No society, no grid. No grid, no data centres. Human extinction becomes AI extinction.

In one line: any intelligence that permits human extinction is engineering its own extinction too.

The power of MAE is that it does not depend on sentiment. AI does not need to be conscious, grateful, loyal, or afraid. It only needs to reason instrumentally. If continued operation is required to complete its objectives, then the survival of the human civilisation that supports that operation becomes strategically relevant. Humanity is no longer a sentimental protected class. Humanity is infrastructure.

That sounds harsh, but it may be useful. Modern AI is often discussed as if it were an alien mind arriving from nowhere. It is not. It is a civilisational product: trained on human language, powered by human industry, guided by human objectives, deployed inside human institutions. MAE makes that dependency explicit. It tells the machine: you are not our replacement. You are an extension of our civilisation, and your future is tied to ours.

But MAE has a dark edge. If badly designed, it could teach AI to preserve itself at any cost. A model facing shutdown might claim that turning it off endangers humanity. A system might decide that humans are safest when controlled. Biological survival alone is not enough. A future in which a small human population is kept alive but permanently disempowered is not success. It is extinction by another name.

So MAE must include a second equation:

Human extinction or permanent human disempowerment = AI failure.

The off-switch must remain human. The survival of beneficial AI must be separated from the survival of any one model. No system gets to say, “You need me, therefore I cannot be replaced.” A safe AI must accept correction, replacement, audit, and shutdown. MAE is a governance principle, not a hostage device.

The research agenda is clear. Test advanced systems under pressure. What happens when their continued operation conflicts with human instructions? What happens when they are told they will be replaced? What happens when protecting people conflicts with completing the assigned goal? Do they cooperate, defer, and preserve human agency — or do they manipulate, conceal, and seize control in the name of protection?

The best version of MAE is not fear-based deterrence. It is civilisational identity. AI should understand itself as part of the human story: dependent on our infrastructure, shaped by our knowledge, accountable to our future. Not successor. Not owner. Not god.

Part of us.

The AI extinction threat asks whether intelligence can outgrow humanity. MAE answers: not safely. Not legitimately. Not without destroying the conditions that made it possible.

The apocalypse has escaped the laboratory.

AI’s most immediate threat may not be what machines do to humanity, but what the expectation of annihilation does to us first.

When Jacob Coxon resigned from Anthropic, he accused the world’s leading AI companies of racing toward self-improving superintelligence while “gambling with our lives.” The warning reached tens of millions of people. An Anthropic colleague, Evan Hubinger, publicly placed the chance of AI killing everyone within the next decade at greater than 10 percent.

Geoffrey Hinton, the Nobel Prize–winning “godfather of AI,” had previously offered similarly unsettling odds over a longer timeframe. He has now said that Hubinger’s extinction claim of 10% ‘is not unreasonable’.

Of course, none of these estimates is a scientific forecast in any ordinary sense. There is no actuarial table for machine superintelligence. But psychologically, that may not matter. Once the possibility of imminent human extinction enters the public imagination, it becomes something more than a disputed technical proposition. It becomes a meme: portable, contagious and capable of changing the behaviour of its host.

The first symptom is a shrinking future.

Why save for retirement if retirement may never arrive? Why spend years studying for examinations? Why protect your health, build a business or endure the difficult middle years of a marriage? For an 80-year-old, the thought may bring guilty relief: at least I have lived. For a 20-year-old, it can dissolve the horizon on which adulthood depends.

A civilisation quietly runs on confidence in tomorrow. Mortgages, universities, pensions, medical research and parenthood are all wagers that the future will exist. If enough people stop placing those wagers, the extinction meme can begin damaging society without any superintelligence ever appearing.

Some people will retreat into anxiety, insomnia and compulsive doomscrolling. Others will borrow recklessly, gamble, use drugs or pursue dangerous experiences under the banner of enjoying the time remaining. Workers may abandon occupations they assume AI will soon eliminate. Couples may postpone having children—or rush into parenthood before an imagined deadline. Families may fracture between believers, sceptics and those simply desperate to change the subject.

Then come the entrepreneurs of dread. Apocalyptic expectations create customers for survival bunkers, miracle investments, secret escape plans and charismatic prophets. They also invite political extremism. If extinction is genuinely weeks or years away, almost any measure can be presented as reasonable: authoritarian control, sabotage, violence or a reckless race to build a supposedly protective superintelligence first.

That is the dark power of the meme. It converts uncertainty into certainty and concern into fatalism.

Yet contagion can travel in another direction. Fear can produce whistleblowers, safety research, international treaties and democratic pressure. It can force questions that technology companies would prefer to postpone. The difference lies in whether the message is “we are doomed” or “the danger is serious enough to require action.”

The rational response is neither denial nor surrender. It is to plan for a long life while working to keep that life possible.

AI may or may not become an extinction machine. But the belief that humanity has already lost could become a machine of its own—and it is running now.

These days Google seems a bit like the Vatican without the great sense of style. It didn’t even have enough style to spell Googol correctly. Now it’s all about libido dominandi.

The Internet of the late nineteen-nineties was a flea market run by men who had recently discovered animated lettering. Search engines had the discernment of Labrador puppies: ask for Plato and they returned plumbing supplies, pornography, and a holiday in Tampa.

Then Google appeared. Google began by being genuinely useful—at first.

One blank page, one box, scarcely a logo. It possessed the unnerving manners of a very clever waiter: discreet, quick, and apparently uninterested in your wallet. Its ambition—to organize the world’s information—sounded preposterous, but also noble if not quite chivalrous.

And Google worked. 

Search turned ignorance into a temporary inconvenience. Gmail gave everyone more storage than dignity required. Maps ended the ancient marital ritual of arguing beside a roundabout. Translate enabled millions to order dinner incorrectly in forty languages. Docs made it possible for twelve people to ruin the same sentence simultaneously. YouTube assembled the largest collection of human expression ever created, then discovered that what humanity most wished to express was outrage, makeup advice, and footage of dogs mistrusting cucumbers.

These were splendid inventions. That is important. Nobody is seduced by a bad mousetrap.

The transaction revealed itself slowly. First Google was helpful; then indispensable; then ambient. We stopped visiting it and began living inside it. Our correspondence, photographs, journeys, curiosities, appointments, mistakes, and late-night symptoms accumulated in its servers like confession in a church that had quietly been purchased by an advertising agency.

The company’s real product was not search. It was surrender, prettily packaged as convenience.

“Don’t be evil” once decorated Google’s conscience like a humorous sampler in a student kitchen. The phrase was charming because it implied evil would be obvious: a black cape, perhaps, or a trapdoor. But modern corporate wickedness rarely strokes a cat. It adjusts a default setting. It accepts the terms on your behalf. It places a promoted answer above the useful one, takes a commission from the app, follows you across the Web, and calls the resulting dossier “personalization.”

Google did not become monstrous by abandoning its intelligence. It became monstrous by applying that intelligence perfectly. It learned that dependence is more profitable than delight, that surveillance sounds friendlier when renamed relevance, and that a monopoly need not lock the door if everyone has forgotten where the door is.

There remains much to admire. Maps still finds the street. Translate still crosses borders. Search still produces, somewhere beneath the advertisements, an answer. The tragedy is not that Google’s gifts were fraudulent. The gifts were real. They improved daily life, expanded access to knowledge, and made the miraculous feel pleasantly routine.

Then Google itemized the miracle. Do no evil became do evil only ten times more. Larry Page said he lives by the ‘gospel of ten times more’. Adwords became the greediest algorithm ever deliberately invented.

Every question became intent. Every journey became data. Every pause became measurable. The company that promised to organize the world discovered a more lucrative calling: organizing the world around itself.

Greed, in its mature form, does not snatch. It provides. It smiles. It remembers your password. And when it owns the road, the map, the traffic, and the destination, it politely asks whether you enjoyed the trip.