I hit rock bottom the morning I felt contempt for a colleague who slept in and couldn't make it to the office by 11.

I wondered how a man in his twenties, working in an unforgiving tech industry, could be so unoptimised. How dare you?

How had he not read The Miracle Morning, The 5 A.M. Club, Atomic Habits, every productivity-obsessed Twitter account I followed? How did he not know that a life is a system and a system can be tuned?

For about a year, I woke at 5am to become a better person. Pray 20 minutes, meditate 10, stare at the goal on my vision board for 10, 20 minutes learning something new, all before the rest of the house was conscious. I had joined the superhuman club.

It worked, in the sense that I was awake at 5am, got things done by noon, and eventually ticked off most of what was on that vision board. It failed at everything that mattered to my soul. I became a man who looked at his team, friends and family and privately felt superior to them.

The contempt reached my faith. I’m Catholic, and prayer had become another input in the optimisation toolset, something I did well, on schedule, to get a result, which is the opposite of what prayer is. You cannot optimise your way into being loved. Grace is, by definition, the thing you did not earn and cannot engineer, and I had built a life with no door it could walk through. My heart was closed for renovation, and nobody was coming through.

I tell you this now because the machine I strapped to my mornings a decade ago is nothing next to the one we’ve all strapped to our minds this year. AI is the most seductive optimisation tool ever built, and it promises to remove the last frictions a human life still runs on: the slow thought, the bad first draft, the waiting, the not-knowing, the needing of other people. This spring Pope Leo XIV published an encyclical touching on these topics, Magnifica Humanitas. Among the sea of shallow commentary online, I’ve finally found one that resonated deeply:

The Curiosity Gap

The Theology of Closing the Laptop

Editor’s Note: First, some personal news: I joined Every full time. Every has had a steadily increasing gravitational pull on my professional life for the last year and a half, and I am thrilled to be all in now: the writing, the AI experimentation, the people, the building, and, crucially, the ability to burn tokens on Every’s dime…

Read more

2 months ago · 4 likes · 1 comment · Katie Parrott

One particular paragraph inspired me to write this post:

“This is why Leo’s encyclical lands so differently for me than a generic warning about screen time or productivity culture. His deepest point is that technology can tempt us to despise the very things that reveal us as human: dependence, limitation, embodiment, need, relationship, and receptivity to grace.” [Emphasis mine].

One of the passages that I found most penetrating in Pope Leo’s document is the claim that we are being tempted to see ourselves as projects to be optimised rather than persons made for relationship. He put a name to the thing that had already half-eaten me ten years before the modern AI tools existed and became omnipresent.

Now let me make the case against myself, because it's strong and I've felt its pull for ten years.

I must admit, the routines worked. I hit the goals I visualised every morning for a year.

Ten years on, the companies got built, the career survived being torn up and rebuilt more times than I can count, I did well financially, and I cannot prove the five a.m. mornings didn’t help. Discipline can be love with its sleeves rolled up. And AI is a gift in many ways. It takes the drudgery that should never have made the desk of most knowledge workers and makes it vanish. The man who tells you to embrace your limitations is usually the man who has already cleared his; it’s easy to praise the slow road from the far end of it. There is something a little obscene about a founder who got everything he chased turning around to warn you off the chase. Limitation is not holy. Efficiency has fed more people than romance ever has.

All of that is true. And none of it is what pulled me out. Rationalising an argument, the logic and logistics are never enough. You need Grace.

What pulled me out was a two-and-a-half-year-old and my wife. I’d started the company when my first son was that age, and for two years my wife carried her PhD and our boy at the same time while I poured myself into the build. It was not an ordinary evening when a call reached me halfway through a hackathon — “enough, get your ass home” — and I left the laptop open on the table and jumped on an Uber. What I walked into stays between me and that house. I'll only say that no morning routine on earth would make me the husband and father that room needed, and that I never went back for the laptop (figuratively). Presence is not a system you can run despite all the self-optimisers telling you to “time box” the date night with your partner on the calendar.

Our daughter was born in the second year of the company, and between them, my wife and two kids imposed themselves on my life until the routine had to move over and make room. No book has ever managed to do that.

So, if the optimisation worked to achieve some goals, what was the problem?

There is a scarcity that tells the truth: I am limited, I cannot make it alone. That is the door grace comes through. But I couldn’t stand at the door and wait. I took a true limit and tried to engineer my way clean out of it, and I made the optimisation tools the god who would save me. But a man busy saving himself has no room left to be saved. He can only earn. He can never receive.

What the burnout and the marriage and those two small people broke open… What Grace broke open was the reverse: I am already enough. Already loved, before the goals and without them. Run the tools from there and they change on you. From abundance, you can pick up the most powerful machine ever made and it stays a machine, a servant. From scarcity, you pick up a pencil and it becomes a god you have to appease every morning at five. The thing is, abundance is already here. The tech messiahs promise you the forthcoming age of abundance, failing to see the abundance we already have. It’s the old marketing trick in the end — fake or create scarcity to sell more.

Most survival advice for the AI era comes in two flavors: resist the machine, or master it. I sell neither. The one question that ever decided anything for me is: what are you building from? Build from the fear that you’re not enough, and AI will only make you faster at fleeing yourself; you’ll arrive nowhere at a spectacular clip. Build from the plain fact that you’re already loved, and the same tool becomes what it was always for: leverage in the hands of someone who knows he is not a machine.

The encyclical has a name for where this points: a life of openness and communion. I’ll add only what a decade of getting it wrong taught me. My son loves fishing, and he often invites me to fish with him. I end up frustrated because I don’t catch any and takes forever. The man who woke at 5am would have counted it as time he could’ve shipped. That is not the tax on a fulfilled life. That is the good life. Guard it like the last thing the machine can’t reach, because it is.

You’re not behind because you were never a project to optimise.

Go be a person.

Well, it was a completely different Alberto Chierici (confabulated and confused with some historical Gaetano Chierici figure), who for some reason was way more visible to AI than I was. Good for him, and rest in peace, but it seemed quite hard to showcase several years of experience and expertise in AI if AI didn’t even consider my existence.

So I decided to figure out how to change AI’s mind on me.

Step 1: show up in reputable places

I started by publishing articles. I learned that AIs would weight results heavily against how reputable the source was. I’ve had several press mentions but they’re too old. AI values current content.

Appearing on .org and .gov domains is an advantage. I also discovered different AIs respect different platforms (but this varies over time) – Claude is very into Substack, ChatGPT loves Medium and Reddit, Google AI reads YoutTube first.

AI’s favourite is third-party signal: reviews about you, media articles, event appearances, being a guest at podcasts, etc. If you’re a shy overachiever, that actually works in your favor

Step 2: don’t obsess over volume, obsess over being correct

In traditional SEO, it’s enough to flood the internet with thousands of blog articles to start showing up in search, and the way people do it today is by creating insane amounts of AI slop.

But AI search doesn’t care to even read AI slop. What AI hunts down and prioritises is thought-through, authentic, expert content. If you publish a few good things, you are good for a long time.

Step 3: answer the right questions

For your publications to be seen as expert by AI, they should answer questions that people actually ask themselves. You have to figure out who you’re helping, what they need help with, what they are looking for. That’s positioning. That’s strategic work

Step 4: introduce yourself consistently

For all its superior intelligence, AI is sometimes very silly. If you introduce yourself with your job title in one place, and with your industry in the other, it will get confused on whether you’re the same person. If you’re introduced with your surname in one place and your name in the other, it will be completely lost.

Wherever you go (own blog, media, podcast) make sure you have control over how you’re introduced – and make it a consistent boilerplate to make it easier for the robots.

Did it work out?

I was invited to speak at a panel discussion on responsible AI in insurance breakfast at AWS. They never asked me for my bio, description, nothing. At the event itself, the moderator just read my profile from ChatGPT:

It sounded quite powerful. Definitely not like something I’d write myself.

I still have some questions to that result, and I’m still working on it. For instance, I’m not Italian-Australian. Just Italian, grazie.

But now that I have a goal in mind and know how to go about it, it’s much easier. And quite enjoyable.

If you’re working on making sure AI speaks highly of your personal brand, hit reply. I might have just the thing for you.

PS: I’m hosting a free webinar to teach you how to build useful stuff with Lovable. Interested? We’re only getting 30 seats. Grab yours → here.

PPS: I wrote this without the help of AI, and you can now certify it (early beta, with limitations, but going towards that direction).

Well, now that is no longer a fantasy. Gen AI may have found the application it was hunting for.

I still believe the narratives of replacing people remain misplaced. If those were true, I couldn’t figure out why A\, OpenAI, and any “AI-pilled” startup keep hiring hordes of software engineers and pay them NBA player packages.

But boy, since I started playing with a few coding tools a year and a half ago till now, the evolution has been beyond exceptional.

I’ve shared reflections and questions on how AI changed software engineering on Gradient Institute’s blog. Have a look.

I also want to point you to a couple of new resources I’m involved with, and I’m sure you’ll find them super valuable:

1. Gradient Briefs - I am convinced the readers of Human Override would find this new, monthly newsletter a must-read! Check out the first two issues here and make sure you subscribe here.

2. Learn to build anything with AI - I am collaborating with serious builders, product managers, designers, and engineers to teach you how to get started with the coding agents tools for anyone who thinks it’s too technical or anyone who’s parked that idea for far too long. You can master the technicalities, and your idea is worth existing. We’ve seen hairdressers go from zero to having custom CRM and booking management systems operational in a few weeks. Reserve your spot here (details still in the making, expect to start in late July and dedicate a few hours per week to it; the cohort has a limited number).

Create the things you wish existed.

Yet the takes remained way too superficial to grasp the depth of despair this fun campaign speaks to.

Even Tony Fadell, the American engineer, designer, and entrepreneur widely known as the “father of the iPod” and a co-creator of the iPhone, posted about it.

There are a few things Fadell is only half-acknowledging.

The iPod Shuffle wasn’t constrained by design but because the tech of 2005 couldn’t do more in that form factor cheaply. People are now choosing it as a constraint, which is a different thing.

The “constraints create freedom” framing has some truth to it, but it slightly launders the history.

Ironically, the iPhone, which Fadell also worked on, is the device that broke the dam and the Shuffle is now being marketed as the escape from the world that device created!

Is this perhaps another engineer seeking forgiveness?

The Back Market ad is fantastic: “Downgrade now” is the kind of line you only get when you fully understand the cultural moment. But let’z go three layers deeper.

YOUR HUMAN OVERRIDE

1) Grief with plausible deniability

I often find myself very nostalgic of the 90s-early 2000s. Nostalgia is wanting the past back. What people are responding to here is closer to mourning a version of themselves they can’t access anymore — the person who could sit with one album for an hour, who had a music taste instead of an algorithm, who finished things.

You used to be the kind of person for whom whole albums made sense. The behavior change is downstream of an identity change you didn’t consent to.

The Shuffle ad works because it briefly returns the agency, implying that the renovation is reversible, that the previous tenant is still in there somewhere.

There’s also a specific kind of shame the ad neutralises. Saying “I can’t focus anymore” is a confession, it makes your willpower, your seriousness, your worth too vulnerable. Saying “I bought an iPod Shuffle” is a story about taste. One is a wound, the other is a personality.

The genius of the ad is offering a socially legible container for what is actually a private grief.

2) Finite as the luxury

My vinyl holds half an album (or a quarter of a concert). A Shuffle holds maybe 240 songs. Spotify holds 100 million. Boy, that infinite optionality is exhausting, isn’t it?

Every Spotify session is 100 million micro-decisions you didn’t make, sitting in your peripheral vision. The cognitive overhead of not choosing the other 99,999,760 songs is real and unrelenting. A finite library gives you back mental real estate.

A constraint is the removal of a tax you didn’t know you were paying.

There’s also a status inversion happening that’s worth calling out.

For most of human history, “more” was the luxury — more food, more clothes, more access, more options. Abundance signalled wealth because abundance was scarce.

We’ve now lived through maybe twenty-five years of digital abundance being free and trivial, and the social meaning has flipped. Now “more” is what poor attention has. The new luxury good is the ability to choose less and stick with it, which requires self-knowledge (knowing what you actually want), security (not needing to keep options open), and time (the rarest currency).

Letterpress, slow food, vinyl, film cameras, the Shuffle — they’re all in the same category, and I don’t think it’s the “vintage” category. I believe it is something deeper: let’s call it legibly chosen finitude.

Choosing finite is how you become a person with edges again. Infinite optionality dissolves the self because the self is constituted by what it commits to and forgoes. You are your no’s as much as your yes’s. The algorithm robs you of meaningful no’s by removing the cost of yes. A finite library forces the no’s back into existence, and the no’s reassemble you into someone with shape.

3) The body wants back in (i.e., why I love vinyl)

The thing screens did that almost no one noticed at the time is that they collapsed the entire spectrum of human gesture into one motion: the swipe. Reading, choosing music, navigating, shopping, talking, learning, dating, and much more became the same flick of a thumb. This is unprecedented in human history. Every previous tool required its own motion: the violin doesn’t play like a hammer doesn’t swing like a pen doesn’t write like a needle doesn’t drop. When my kids were deprived of writing with pen and paper during COVID, they lost important mental faculties and brain connections built by that gesture.

The body knew where it was by what it was doing. Screens severed that. Your body has no idea whether you’re reading Tolstoy or buying socks because the gesture is identical. Proprioceptive identity — the sense of being a body engaged with a particular task — got dissolved into uniform glass.

The Shuffle’s click wheel and physical affordances matter because they give you back a task-specific motion. Your thumb knows it’s choosing music in a way that swiping doesn’t. This sounds small. It isn’t. The motor cortex and the sense of agency are deeply linked: feeling like you did something requires the body to have actually done something distinct.

Touchscreens give you outcomes without efforts, which is why everything done on them feels slightly unreal and slightly unsatisfying even when it works. Over years, this accumulates as a low-grade dissociation from your own actions.

Memory is the other casualty no one talks about. Memory is encoded with motor and spatial context: you remember where on the page something was, you remember the weight of the book, you remember turning the record over. This is called embodied cognition. It’s how the hippocampus works. When all your inputs come through the same flat rectangle, your brain has nothing to hang them on. Hence, the strange contemporary symptom of having consumed enormous amounts of content and remembering almost none of it. It’s not your fault because you didn’t pay attention. This new setup doesn’t give your body something to attach the memory to.

The vinyl ritual — sleeve, needle, flip, sleeve — is a memory-encoding scaffold. Same with dog-earing a page, same with the click wheel’s haptic feedback.

And there’s something almost spiritual underneath this, though I’ll try to say it without being cheesy. Humans evolved as creatures whose thinking happens through doing.

The Greeks walked while they philosophised. Monks copied texts by hand because copying was a way of knowing. Cooks know things their hands know that they can’t say. The body is part of the cognitive apparatus, and it’s too often mistaken as being just a delivery (and receptive) vehicle for the brain. Fifteen years of touchscreen primacy have run a hidden experiment on whether thinking can be done with the body sidelined, and the early results are in: it can’t, or at least not well, or at least not in a way that feels like living.

People reaching for tactile objects right now are trying to think again.

This is why my wife and I started building a vinyl collection. We see it as a tool you use to be a person.

It’s been a while, I know.

I’ve spent the last month writing a long essay that I felt was much needed in the AI discourse. After I’ve read everywhere this silly statement that AI is the Copernican revolution of cognition, I couldn’t take it anymore and had to set the record straight.

Moreover, I’ve been rethinking what the Human Override has to become and what it is for you now that I am approaching 1,000 readers on Substack (+3,000 on Ln). But I need your input: reply to this email, and let me know:

• Why do you read it?

• What is the most valuable thing for you?

• What am I not writing/doing that you’d like me to do?

And anything else you share. I read and reply to every email.

That said, here’s the long essay that I’m pretty sure you’ll find interesting, and hopefully a spark for discussions and debates with me, with your friends, coworkers and many more!

I believe this metaphor conceals a profound confusion about what intelligence is, what the human person is, and what is ultimately at stake when we speak of mind, meaning, and the place of the human person in this technological transformation.

This is not an essay against AI’s remarkable capabilities. It is an essay about what happens when there’s an unintended divorce between science and philosophy — a phenomenon with a long and instructive history. Curiously, though, the pattern has inverted since the last time it caused a cultural disaster.

The inversion: from bad philosophy stemming from misunderstood science to bad philosophy stemming from unlearned technologists.

A century ago, the problem ran in one direction. The great revolutions of early twentieth-century physics — quantum mechanics and relativity — were systematically misinterpreted by philosophers and public intellectuals who lacked the scientific training to understand what these theories said about reality. The uncertainty principle became “reality is unknowable.” Relativity became “everything is relative.” Profound, mathematically rigorous theories and new knowledge of natural phenomena were trivialised into metaphysical systems that parted ways from what the science had discovered.

At the time, the scientists themselves were learned enough to push back. Werner Heisenberg — the German physicist who won the Nobel Prize in 1932 for the creation of quantum mechanics — devoted much of his later career to untangling these philosophical knots. In Physics and Philosophy (1958), he tenaciously traced how positivists and half-informed commentators had distorted the meaning of quantum mechanics, and he did so from a position of genuine philosophical literacy.2 Heisenberg had read Plato, Aristotle, Descartes and Kant. He could engage with the philosophical tradition on its own terms, identify where it had gone wrong about physics, and offer corrections. He was not alone: Niels Bohr, Erwin Schrödinger, Albert Einstein and others brought serious humanistic education to bear on questions about what their discoveries meant for our understanding of reality.3

Today, the pattern has inverted, and the inversion is as much, if not more, dangerous because of the status the proponents of these dubious ideas have globally vs. the intellectuals of that era.

Now it is scientists, engineers, CEOs and technologists who are spreading bad philosophy. The Google engineer who declares that large language models may be sentient, the computer scientist who announces that human cognition is “just pattern matching,” the mathematician who casually asserts a Copernican revolution in intelligence, and the philosophers, not knowing the science, trust the scientists and develop logical conclusions like the singularity — the view that machines recursively improve themselves until they decide getting rid of humans is a more efficient world order. Popular figures from Silicon Valley are making far reaching philosophical claims while being, for the most part, embarrassingly ignorant of the philosophical traditions that have grappled with these questions for millennia.

Unlike Heisenberg, who could hold his own with Kant, the typical AI commentator making pronouncements about the nature of mind has not read — and often has not heard of — the thinkers whose work is most directly relevant.

Today, one would expect philosophers to correct scientists who get the philosophy wrong. But most contemporary philosophers, with notable exceptions, do not understand the technology well enough to engage credibly. The result is a vacuum, a space where bad philosophy circulates essentially unchallenged, because the people with the philosophical training lack the technical understanding, and the people with the technical understanding lack the philosophical training.

There are exceptions. Luciano Floridi — the Italian-British philosopher who holds the Castle Chair at Yale’s Digital Ethics Center, and whose work spans the philosophy of information, digital ethics, and AI — is one of the rare figures who combines genuine philosophical depth with serious technical understanding.4 His insight that AI represents “a divorce between agency and intelligence — the ability to solve problems successfully and the necessity of being intelligent in doing so” — is the kind of philosophically informed formulation that the discourse requires.5 But such voices are scarce. For the most part, we are living through a period in which the people making the loudest philosophical claims about AI are the least equipped to make them.

A word with a long and tortuous history

The difficulty begins with the word “intelligence” itself, which in contemporary usage has been so narrowed as to be almost unrecognisable to anyone formed in the great tradition of Western thought. The word has a history that most people deploying it in AI discourse seem merrily unaware of.

The Latin root, intelligere originally carried the sense of discernment: the ability to read between things, to perceive what is not immediately given.6

The ancient Greeks had an even richer vocabulary. Aristotle — the fourth-century BCE philosopher whose works on logic, metaphysics, ethics, and natural science shaped Western thought for two thousand years — used nous (νοῦς) for the highest faculty of the mind: the capacity for direct intellectual apprehension of first principles, the unprovable starting points from which all scientific demonstration proceeds.7

Note that it is not the realm of computing or calculus. It is not even what modern computation calls (stubbornly vomited everywhere these days) “pattern recognition.” It is something closer to insight — the grasp of what must be true before reasoning can even begin.

Plato — Aristotle’s teacher, and the founder of the Academy in Athens — had earlier distinguished nous from dianoia (discursive, step-by-step reasoning), placing intuitive intellect above the demonstrative kind in his famous analogy of the Divided Line.8

In the medieval period, the Latin intellectus became the scholarly rendering of nous, and it carried nontrivial metaphysical content. Thomas Aquinas — the thirteenth-century Dominican friar whose synthesis of Aristotelian philosophy and Christian theology remains foundational in Catholic thought — developed the concept of the intellectus agens (the active intellect), the faculty by which the mind abstracts universal forms from particular sensory experience.9 Let me say it again, this is not computing or inferring from repeated patterns; it is the movement from the concrete to the universal, from this triangle drawn in sand to the concept of triangularity itself.

The early modern period began belittling the word. Thinkers like Francis Bacon, Thomas Hobbes, and John Locke — the architects of British empiricism — abandoned the metaphysically onerous vocabulary of intellectus in favour of the more modest “understanding,” maybe deliberately so because they wanted to distance themselves from the elaborate scholastic apparatus of soul, intellect, and form.10

Then, in the early twentieth century, the psychometric tradition — Alfred Binet, Charles Spearman, David Wechsler — attempted to turn “intelligence” into something measurable: a score, a quotient, a position on a bell curve.11 As a brief aside, we should remember that Spearman and the American psychologists who adapted Binet’ s methods were deeply entangled with the eugenics movement, using their tests to rank human worth based on hereditary assumptions.12

What’s indisputable though, is that the rich philosophical concept of intelligence was operationalised into a number.

It is this final, hollowed-out, desiccated residue of the word that now exercises a quiet tyranny over our discourse concerning artificial intelligence. When the heralds of technology proclaim that these machines are “intelligent,” or summon us to “rethink intelligence” in the light of their silicon artifacts—while one cannot but ask, with a certain interior sorrow, why we do not rather rethink it in the light of that profound contemplative and eschatological tradition which our post-Enlightenment culture has so violently severed from itself—they wield in truth only a diminished, impoverished, and narrowed conception of intelligence: that melancholy reduction of a human mystery to measurable performance, to number, to a ranking upon tasks we have ourselves, and quite arbitrarily, declared to be the measures of cognitive worth.

And by that definition, yes, these models are strikingly capable, and on a number of tasks they surpass the average person. But read that sentence back, and notice how it sits with you. To me it has a patronising quality, and something faintly worse than that—does it not strike you the same way (what/who is the average person?)?

This is rather like concluding that a submarine swims. The submarine moves through water, it arrives at its destinations, it outperforms fish in a great many contexts. And yet something essential is lost the moment we file it under “swimming” and announce that our understanding of aquatic locomotion has been overturned.

I am not prepared to redefine intelligence merely so that we may go on lowering the bar. For we shall lower it, and lower it again, until at last nothing remains that is exceptional in being human. Is this truly what we desire?

The case against human exceptionalism

To be fair, the “Copernican” camp is not entirely wrong. There is a legitimate intention here, and it has some merit.

For centuries, human self-understanding has been warped by various forms of anthropocentrism. We assumed that our perceptual categories were the natural way to slice up reality, that our style of reasoning was the only kind, that the specific cognitive profile of Homo sapiens — good at narrative, mediocre at probability, brilliant at social inference, terrible at large-scale computation — was coextensive with “thinking” itself.

The psychometric tradition that José Hernández-Orallo, a Spanish computer scientist at the Universitat Politècnica de València whose work on universal psychometrics seeks to measure intelligence independently of any particular species or architecture, has critiqued the term was built entirely around human norms.13 Intelligence was what humans were good at, tested in ways that made sense for humans, scored against a distribution of other humans.

The advent of LLMs has shattered that narcissistic self-reflection, perhaps offering mankind an opportunity to rediscover the profound dimensions of intelligence that the reductive metrics have long obscured.

LLMs reveal that many tasks we intuitively coded as requiring “understanding” — summarisation, translation, analogy, even certain kinds of creative recombination — can be performed by systems that operate on entirely different principles than biological brains. Systems that, it is worth remembering, have been built by human inventiveness and skill and rely on a gargantuan amount of humanly generated artifacts (books, photos, videos, sounds, and so on).

Tao’s observation, while not universally applicable, that AI has “driven the cost of idea generation down to almost zero, in a very similar way to how the internet drove the cost of communication down to almost zero” may be applicable in some fields (I take he’s speaking of his experience in mathematics). What was once the bottleneck (generating possible approaches) is now cheap, faster, and the bottleneck has shifted to evaluation, judgement, and the ability to discern which ideas matter.14

Allow me one more digression. In domains like mathematics, idea generation might turn out to be a search problem, tractable to search over a well-structured, though gargantuan, possibility space. Mathematics, like the game of Go15 that AlphaGo16 so famously mastered by AI, is a self-contained world: its postulates are fixed, its rules logically defined, its moves exhaustively specifiable in principle. The space is finite, or at least formally bounded. Yet, it remains too vast for any single genius, or indeed any generation of mathematicians or players, to traverse in a lifetime. A breakthrough, in this kind of world, can be the discovery of something that was always there, waiting inside a space too large to search by hand. And there is another kind of breakthrough, and it is of an entirely different order: the invention of a new game altogether, a new mathematical system or physical theory, which preserves the old one as a special case while opening a wider range of possibilities—as general relativity did to Newton, as quantum mechanics did to classical physics. This second kind of breakthrough is not search. It is the construction of a new space within which new search problems becomes possible.

What the case against human exceptionalism tells us, in other words, concerns the structure of certain tasks: their decomposition, their information-theoretic properties, their susceptibility to search. And it tells us where the genuinely hard parts are shifting, where we could spend more time.

Hernández-Orallo’s “universal psychometrics” represents a serious attempt to develop measurement frameworks that are not biased toward any single cognitive architecture — what he calls measuring cognitive abilities “in the machine kingdom” (see footnote 12).

And the broader point about cognitive diversity is well taken.

We have always lived among other intelligences — the navigational genius of migratory birds, the distributed cognition of ant colonies, the chemical computation of plant root networks. Recognising that intelligence comes in many forms, with radically different profiles of strength and weakness, is not a demotion of humanity. Rather, it is an enrichment of our understanding of nature.

The case for human exceptionalism (the right kind)

There is a slide — subtle, often unconscious — from the reasonable claim that human intelligence is not the only kind of information processing in the universe to the much stronger claim that humans are not exceptional in any deep or important sense, especially in the post-Enlightenment age where reason and intelligence have been wielded as the most important dimensions of humanity. This second claim is not an empirical discovery but a poorly supported philosophical stance — and, I would add, a misanthropic one.

Start with the most basic observation: humans are the only entities in the known universe that ask the question.

The fact that this debate is happening at all — that Tao and other illustrious scientists can formulate the Copernican analogy, that Emily Bender and her colleagues can mount the “stochastic parrot” critique,17 that researchers across disciplines can argue about the nature of mind — is itself a datum that the deflationary view cannot easily accommodate. LLMs participate in conversations about intelligence. For instance, I “chatted” with several LLMs to clarify and polish this article. They do not, so far as anyone can tell, care about the answer. My friend , who helped me (more successfully than the LLMs) develop ideas for this article, cares.

Immanuel Kant — the eighteenth-century German philosopher whose three Critiques remain among the most influential works in modern philosophy — grounded human dignity in rational autonomy: the capacity to set one’s own ends, to act according to self-legislated moral principles, to treat other persons as ends in themselves rather than merely as means. “Rational nature exists as end in itself,” he argued in the Groundwork of the Metaphysics of Morals (1785).18 You can quibble with Kant’s framework (many have), but the core insight has proven remarkably durable: there is something about the human capacity for moral self-determination that is categorically different from the ability to optimise an objective function.

An LLM can generate a (sometimes) compelling consequentialist argument for any position you like. What it cannot do — and here the stochastic parrot critique, for all its crudeness, points at something real — is mean it. It cannot stake its existence on a moral commitment. It cannot, in the sense of Søren Kierkegaard — the nineteenth-century Danish philosopher who insisted that truth must be personally appropriated through existential commitment — make a leap.19

Hannah Arendt — the twentieth-century German-American political theorist, a student of Martin Heidegger and Karl Jaspers, whose work on totalitarianism and the nature of political action reshaped modern thought — drew a distinction that goes one level deeper. In The Human Condition (1958), she argued that what is most distinctively human is not labour (which we share with animals) or work (which produces the durable objects of civilisation), but action — the capacity to begin something genuinely new, to insert oneself into the world as a unique and unrepeatable being. “In acting and speaking,” she wrote, “men show who they are, reveal actively their unique personal identities and thus make their appearance in the human world.”20 This disclosure of who someone is, as opposed to what they are — their qualities, talents, and shortcomings — is, for Arendt, the domain of freedom, of plurality, of the political. It is what makes each person not a token of a type, but a source of irreducible novelty.

An LLM generates novel text. It does not act in Arendt’s sense. It does not disclose a who. It does not carry the weight of a life, a history, a set of commitments and betrayals and loves that make a particular utterance this person’s utterance and no one else’s. Which also illuminates the debacle among many writers and scholars about writing with or without LLMs. I extensively use LLMs to help me write, but it is entirely my thought, my action, my research, my history that influenced the desire to put down these words and share them. I sympathise with scholars who lament attribution — but let’s keep that debate for another time.

The confusion between generating novel outputs and performing genuinely new action is one more philosophical mess of the current moment.

The Enlightenment’s amusing detour

There is a deeper irony that is worth savouring. The intellectual tradition that is now most enthusiastic about dethroning human intelligence is, in a sense, the same tradition that enthroned it in the first place.

The Enlightenment project was a sustained effort to place human reason at the summit of the knowable universe. Out went revelation, tradition, and divine authority; in came empiricism, rationalism, and the sovereignty of the thinking subject. Kant adopted the ancient Roman poet Horace’s injunction — Sapere aude, “Have courage to use your own understanding!” — as the motto of the Enlightenment, a declaration that human reason needed no external warrant.21 It was its own foundation.

Kant himself described his philosophy as a “Copernican revolution.” In the Critique of Pure Reason (1781), he reversed the traditional view that knowledge conforms to objects, claiming instead that objects must conform to the conditions of our knowledge.22 The AI enthusiasts now heralding a “cognitive Copernican revolution” are arriving comically late to a party that Kant threw in the eighteenth century. In fact, they are reversing the revolution again: Kant put reason at the centre; they seek to dethrone it. If the Copernicus analogy is apt, it is like repeatedly moving the Sun in and out of the solar system according to the most viral swing on social media.

The comparison is instructive in another way. Copernicus’s original shift of perspective was brilliant because it solved real problems in natural sciences: the epicycles and equants that had accumulated in Ptolemy’s system were symptoms of a model that had placed the wrong thing at the centre, and the heliocentric reframing dissolved them.

Kant’s “Copernican” turn, at least, attempted to solve a genuine philosophical problem — how synthetic a priori knowledge is possible — by relocating the structuring conditions of experience from the object to the subject.

A reasonable question emerges: what exactly are the problems that the new “cognitive Copernican revolution” is solving? If it is, as this essay has argued, merely the latest chapter of a positivism operating with an impoverished sense of intelligence, then the answer is: none. It is rediscovering a pre-modern insight while having systematically discarded the conceptual tools that once made that insight philosophically productive. Kant’s noumenon — the thing-in-itself that lies beyond the reach of empirical knowledge — is itself a reminder that there are limits to what the knowing subject can grasp, a kind of “negative knowledge” that marks the boundary of cognition. The current discourse could use a similar dose of epistemic humility.

By rooting all authority in human rationality, the Enlightenment placed the entire burden of meaning and truth on human reason. When Friedrich Nietzsche — the nineteenth-century German philosopher whose diagnosis of European nihilism proved prophetic — declared that God was dead,23 he understood something crucial. Far from a celebration, his pronouncement warned that if human reason is the only source of meaning and value, then the entire edifice of civilisation rests on something far more fragile than its architects supposed. In today’s ironic twist, AI is exposing this fragility in plain sight.

Now, the heirs of the Enlightenment are busily dismantling the very pedestal they built. Having spent three centuries insisting that human reason is the measure of all things, the technologist of the AI age announces that, actually, human reason is not so special after all. There are other intelligences. We are not the centre. Some are even rebuilding the concept of god by saying we created “godlike” technology. Some, I’d dare to say, would even go as far as thinking AI is a kind of new god.

Various traditions, of course, have maintained this all along, under different names and frameworks — and not only theistic ones.

South American indigenous cultures, for instance, have long understood ayahuasca, trees, and jaguars as bearers of wisdom and intelligence; they could think of ecosystems themselves as intelligent instances. “The jungle knows,” as the saying goes. “The river knows.”

This “new” cognitive Copernican revolution is, from certain angles, secular thought rediscovering what it spent centuries denying — but arriving at the destination stripped of the conceptual resources (dignity, personhood, teleology) that once made the co-existence of human and non-human intelligence philosophically manageable. This is what makes the current discourse not just mistaken but shallow. Floridi’s observation that AI represents a divorce between agency and intelligence is so valuable because it resists that dilettantism. The divorce gives us a conceptual tool for understanding what is genuinely new about AI without banalising into either hysteria or deflation (see footnote 5).

The value of dignity

The worst forms of human exceptionalism are well known and rightly condemned. The idea that one group of humans is inherently superior to another — the ideological foundation of every genocide — is exceptionalism at its most catastrophic. When people are rightly wary of “human exceptionalism,” this is often what they have in mind (I hope): a ranked ordering of beings, with a particular subset of humanity conveniently at the top.

But there is another exceptionalism — call it dignitarian exceptionalism — that has nothing to do with ranking and everything to do with recognising a specific kind of value. And, if you hate hierarchies, let’s say a different place for humans in nature: have you ever asked why humans care about other animal species so much so that they feel responsible for their wellbeing and survival? Have you seen any cows caring for pandas? Or ants organising NGOs to protect white bears? Humans care and defined (or discovered?) the concept of dignity.

A person with severe intellectual disabilities has the same dignity as a Nobel laureate. A newborn infant, who cannot pass any cognitive benchmark, is no less a person than the most brilliant mathematician. This should be obvious, but in a culture that increasingly measures worth by performance and productivity, it needs to be said with renewed force. Human dignity does not rest on being the best reasoner in the room. If the argument is “LLMs can do X, therefore humans are not exceptional,” then we have implicitly accepted that exceptionality depends on doing X. That’s Batman morality: “It’s what you doooo, that defines youuu [rough voice].

Human dignity does not rest on being the best reasoner in the room. Rather, on the capacity for love, for moral commitment, for self-awareness, for the peculiar and unquantifiable experience of being a someone rather than a something.

Both poles of the reaction are true, but in different registers. “Relax, you’re not that important” is true in the cosmic sense: we are not the centre of the universe, our cognitive style is one among many, our pattern-matching has measurable limits. But “don’t worry, you are indeed exceptional and unique and infinitely valuable” is equally true, in the register of ethics, of dignity, of what it means to stand before another person and recognise them as an end in themselves.

The philosophical task is to hold both of these truths at once. The deflationary view (“we’re just another form of information processing”) and the inflationary view (“humans are the crown of nature”) are each wrong in isolation and each captures something real. The hard work — the genuinely philosophical work, not the Davos-panel or podcast version — is to articulate a framework in which cognitive humility and personal dignity coexist without contradiction.

The true Copernican lesson

It is worth remembering what the actual Copernican revolution taught us. Copernicus did not prove that the Earth was unimportant. He proved that it was not the centre of the universe. These are different claims.

The cultural resistance to heliocentrism was not really about orbital mechanics. It was about what the geometry was taken to mean. In a world where philosophy and theology had not yet learned to treat science as one method of knowledge among others — rather than as a rival account of ultimate reality — to demote the Earth spatially felt like demoting it in every other sense. If the Earth was no longer at the centre of creation, then creation itself seemed to lose its centre of gravity. The fear was that the novel discovery would hollow out the significance of human life.

That fear was misplaced, and we can see now why. The Earth after Copernicus was no less remarkable — no less the home of life, consciousness, culture, love, and suffering — for not being at the geometrical centre of the solar system. The same confusion persists today among those who assume that because science describes the physical structure of reality with remarkable success, it must therefore exhaust what is real, leaving no room for meaning, value, or anything that does not appear on an instrument. This is the Ptolemaic anxiety of the secularist dogma: the worry that to locate ourselves honestly within nature is to lose whatever made us worth locating in the first place.

If anything, the recognition of our true position in the cosmos made the existence of these things more astonishing, not less. The universe is vast and largely indifferent, and yet here we are: beings who compose symphonies and mourn their dead and argue about the nature of intelligence.

If AI teaches us that human cognition is one island in a larger archipelago of possible information-processing systems, this is a valuable lesson. At the same time, that risks being a move from “we are not the only intelligence” to “we are not importantly different”. Every time a civilisation has convinced itself that certain categories of beings lack significance, the results have been dreadful. The direction of true progress ought to be toward expanding the circle of dignity, not contracting it.

The stochastic parrot critique and the Copernican intelligence thesis are, in a sense, mirror-image errors. One says: machines can’t really think, so nothing has changed. The other says: machines can really think, so everything has changed. Both are wrong because both assume that “thinking” can be a well-defined, in the scientific sense, dimension, and that the philosophical question is simply whether machines do or do not possess it. By the way, both forget that we (unexceptional humans) figured out how to make machines “think” or “parrot” convincingly.

The richer question — the one that Hernández-Orallo’s universal psychometrics begins to gesture at, and that Tao’s mathematical experience illustrates vividly — is how to map the full space of cognitive possibilities without reducing the ethical and existential dimensions of being a particular kind of mind. An LLM occupies a fascinating region of that space, if there is such a thing. So does an octopus. So does a person reasoning about what to do in a moral emergency. These are not points on a single scale. They are different kinds of things, and the philosophical vocabulary we need is one that can honour that difference without either inflating human cognition into the measure of all things or deflating human personhood into just another data point. For now, Floridi’s insistence on “agency” remains the most useful conceptual framework.

Martin Heidegger — the twentieth-century German philosopher whose later work on the essence of technology remains among the most penetrating analyses of modernity — observed that technology itself cannot think about the essence of technology.24 The same applies, with even greater force, to intelligence: if intelligence is reduced to computing, to what Heidegger called rechnendes Denken (calculative thinking), then it cannot grasp its own nature. Henri Bergson — the early twentieth-century French philosopher whose work on consciousness, time, and evolution earned him the Nobel Prize in Literature — put it with characteristic precision: one must think against intelligence to understand the genesis of intelligence.25 This is the recognition that intelligence, in its deepest sense, exceeds the scope of any particular exercise of it — including the exercise of building machines that simulate it.

—

The scientists and technologists now rushing to announce a Copernican revolution in intelligence are, in many cases, making the same kind of mistake that philosophers and public intellectuals made with quantum mechanics a century ago — extracting sweeping metaphysical conclusions from genuine empirical discoveries, without doing the careful conceptual work required to know what those discoveries actually are, and then, only then, what they mean. The difference is that back then, learned scientists could set the record straight. Today, there is a vacuum where the corrective should be. As our chief scientist at Gradient Institute — a non-profit research organisation helping policymakers and industry leaders to have clarity on AI based on science — science is about what things do, metaphysics is about what things are. (We plan to write a blog about this, and what it means for the science of artificial intelligence — stay tuned!)

That work is harder and slower than writing a viral tweet (hence this very long article, sorry!). It requires engaging seriously with the philosophical tradition — with Aristotle on nous, with Aquinas on dignity, with Arendt on action, with Heidegger on the essence of technology, with Bergson on the limits of the intellect, with Floridi on the philosophy of information, with the full, tangled history of what we have meant and failed to mean by “mind” and “intelligence” and “person.” It requires, above all, the intellectual humility to recognise that being good at building intelligent systems does not make you an authority on what intelligence is — any more than being good at building telescopes makes you an authority on what the stars are for.

References

Arendt, H. (1958). The human condition. University of Chicago Press.

Aristotle. (c. 340 BCE/2009). The Nicomachean ethics (D. Ross, Trans.; L. Brown, Rev.). Oxford University Press.

Bender, E. M., Gebru, T., McMillan-Major, A., & Shmitchell, S. [Mitchell, M.] (2021). On the dangers of stochastic parrots: Can language models be too big? In Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency (pp. 610–623). https://doi.org/10.1145/3442188.3445922

Burchard, P. (2025). The AI pipeline fallacy [LinkedIn post]. https://www.linkedin.com/posts/paulburchard_artificialintelligence-machinelearning-activity-7199091650701246464-lb0y

Floridi, L. (2014). The fourth revolution: How the infosphere is reshaping human reality. Oxford University Press.

Floridi, L. (2019). What the near future of artificial intelligence could be. Philosophy & Technology, 32(1), 1–15. https://doi.org/10.1007/s13347-019-00345-y

Floridi, L. (2023). The ethics of artificial intelligence: Principles, challenges, and opportunities. Oxford University Press.

Heisenberg, W. (1958). Physics and philosophy: The revolution in modern science. Harper & Brothers.

Hernández-Orallo, J. (2014). Universal psychometrics: Measuring cognitive abilities in the machine kingdom. Cognitive Systems Research, 27, 50–74. https://doi.org/10.1016/j.cogsys.2013.06.001

Hernández-Orallo, J. (2017). The measure of all minds: Evaluating natural and artificial intelligence. Cambridge University Press.

Holden LR, Tanenbaum GJ (2023). Modern Assessments of Intelligence Must Be Fair and Equitable. J Intell, (6):126. 10.3390/jintelligence11060126

Kant, I. (1784/1996). An answer to the question: What is Enlightenment? In M. J. Gregor (Ed. & Trans.), Practical philosophy (pp. 11–22). Cambridge University Press.

Kant, I. (1785/2012). Groundwork of the metaphysics of morals (M. Gregor & J. Timmermann, Trans.; Rev. ed.). Cambridge University Press.

Kierkegaard, S. (1843/1983). Fear and trembling (H. V. Hong & E. H. Hong, Trans.). Princeton University Press.

Nietzsche, F. (1882/2001). The gay science (J. Nauckhoff, Trans.; B. Williams, Ed.). Cambridge University Press.

Plato. (c. 380 BCE/2006). Republic (R. E. Allen, Trans.). Yale University Press.

Tao, T. (2026, March 20). Kepler, Newton, and the true nature of mathematical discovery [Interview by D. Patel]. Dwarkesh Podcast. https://www.dwarkesh.com/p/terence-tao

Bergson, H. (1907/1911). Creative evolution (A. Mitchell, Trans.). Henry Holt and Company.

Heidegger, M. (1954/1977). The question concerning technology. In The question concerning technology and other essays (W. Lovitt, Trans.; pp. 3–35). Harper & Row.

Heidegger, M. (1959/1966). Discourse on thinking (J. M. Anderson & E. H. Freund, Trans.). Harper & Row.

Kant, I. (1781/1998). Critique of pure reason (P. Guyer & A. W. Wood, Trans.). Cambridge University Press.

Viveiros de Castro, E. (1998). Cosmological deixis and Amerindian perspectivism. Journal of the Royal Anthropological Institute, 4(3), 469–488. https://doi.org/10.2307/3034157

Viveiros de Castro, E. (2014). Cannibal metaphysics: For a post-structural anthropology (P. Skafish, Trans.). Univocal Publishing.

Notes

New readers highlight: Deborah H, Leadership Coach | Hendra H, General Manager | Sanija N, AI Engineer | Stephen L, VP | Hilary C, Director | Liem G, Policymaker and +43 more amazing people.

Welcome!

It hits me when I’m in line at the post office, scrolling through my phone, caffeine from my morning coffee barely kicking in. My thumb freezes on a notification: another company just slashed thousands of jobs, offering the inexorable rise of AI as the main reason for a decision “we don’t take lightly”.

Suddenly, my heart rate stutters, stomach tightening, the familiar surge of uncertainty flooding in. I can’t shake this swirl of emotions every time I see another breathless headline about AI “devouring” jobs.

On the one hand, these headlines can feel like exaggerations. It feels like the public conversation leaps from optimism to apocalypse without much nuance. For example, last year, ESPN introduced an AI tool to generate sports recaps and automate basic news briefs. The rollout led to a real reduction in temporary writer contracts for that narrow coverage. An undeniable job impact, though far from replacing entire newsrooms. It’s this kind of selective automation that gets turned into sweeping claims about AI taking over everything, which misses the real complexity.

And the “thousands of jobs automated” stat? I roll my eyes... It’s not like someone just hits Ctrl+Tab and poof, factories of code roll themselves out.

But then I pause: if a fraction of that is even remotely true, what’s to stop my role from being next in line?

On the other hand, let’s be honest, the capability in today’s tools - Claude Code, Cowork, Lovable, Codex, the new Gemini… you name it - is nothing short of astonishing. I’m torn between mocking the doomsayers and admitting I’m occasionally awestruck.

I remember those agonising all-nighters spent chasing down a redundant function that mercilessly crashed the database sync. Though I’ve not been coding for work for a few years, I hear from my friends, software engineers, that AI swoops in now, sketches out a fix in seconds, and I’m left wondering if I’m allowed to feel relief or if that relief is a harbinger of my own obsolescence.

Sure, I love that we don’t have to toil over boilerplate or debug the same mistake for the tenth time. But sometimes I find myself nostalgically craving that low-level grind, that sense of victory when you at last crack a stubborn bug. There’s a strange guilt in watching AI handle the grunt work. I should be grateful, right? Yet I can’t help suspecting that by relaxing a bit, I’m edging closer to being replaced.

While all the chatter is about layoffs and cutbacks, I can also see startups and more innovative incumbents churning out more features, better customer service, sleeker products, and entering new markets. Some even create entirely new market categories (check out Synthesia). It makes me wonder if these AI breakthroughs are really unlocking growth, or just shifting costs around until someone, somewhere, gets axed. And perhaps it is this dichotomy that is wrong. It’s not an or-or. All those things can live together.

Let’s take web design as a case study, an interesting market that one would have thought was dead several times already.

In 2012, I spent $500 on a guy developing a simple, straightforward HTML wedding site, no frills. Now I’ve got a $6,000 quote for rebuilding my personal website. Insane? Maybe. But if I piece together vibe-coding credits, hosting fees, build hours and monthly tweaks, I’m looking at north of $10,000 over a few months if I choose to do it myself. So yes, while the sticker shock stings, I also get why the market bears it.

The web has morphed wildly: static HTML in ’95, CMS and blogs by ’05, mobile-friendly and analytics-driven by ’15, and now full-blown digital products optimised for performance, security, accessibility and conversions. That evolution didn’t hollow out jobs. Instead, it reshaped them and expanded the field. Mordor Intelligence pegs the web design market at $56 billion in 2024, with growth to $92 billion by 2030 (Mordor Intelligence, 2026)[1].

Site builders, WordPress, no-code platforms were supposed to kill off developers. Instead, they lifted the floor and raised the ceiling. AI feels like the next tide. It will crank out your boilerplate, draft your components, prune your routine hours and then clients will demand more: smarter, deeper, faster. They always do.

I can picture what this rising ceiling looks like. Imagine a client who wants a web experience that evolves in real time. The site greets each visitor by name, adjusts the layout and messaging based on browsing habits, and even morphs its look and tone for different audience segments as they interact. The AI analyses feedback, cross-references external events, and auto-launches fresh features to keep the user engaged. Suddenly, it is not enough to just deliver a site that loads fast; the bar keeps moving upward, and staying ahead means inventing experiences that were barely imaginable a year ago.

Still, I can’t avoid wondering: is this wave different? Or will history repeat? When factories electrified, manufacturing jobs doubled, output tripled, and employment climbed (U.S. Bureau of the Census, 1975)[2]. Can we trust that pattern again?

So here I stand, torn. Part of me wants to cling to old certainties; another part is itching to plunge into every new tool. I feel equal measures of dread and excitement, scepticism and hope. Maybe that’s exactly where we’re supposed to be: uncomfortably on edge, learning fast, and trusting that adaptability indeed has always been the core skill to hone.

I keep coming back to one idea: uncertainty and change are the only certainties, and curiosity is a solid strategy for being a protagonist. Letting curiosity lead means I can hold all this ambiguity without getting paralysed, letting myself explore instead of defend, and keep moving through the thrill and the fear. It’s a way to lead, not just react.

The total available market doesn’t shrink when capability expands. It grows.

Calm down. Learn the tools. Raise your prices.

That’s always been the playbook. Let’s hope it holds even now.

___

PS.

Webinar: I'm running a small, live session on the one question nobody's answering well: how do you decide what to learn when everything changes every quarter? I'll share the framework I've used to stay relevant across four career pivots and an AI revolution. 50 spots, so it stays conversational. 35 USD. Grab a seat here → The AI Mindset Shift

This article’s writing level: L1 - Human led. (What are these levels?)

Further reading: AI’s Wake-Up Call: New SHRM Research Reveals 23.2 Million American Jobs Already Impacted. 2025. A more nuanced study on what automation does to jobs.

Sources:
[1] Mordor Intelligence. (2026). Web design market size & share analysis – Growth trends & forecasts (2025–2030). https://www.mordorintelligence.com/industry-reports/web-design-market

[2] U.S. Bureau of the Census. (1975). Historical statistics of the United States, colonial times to 1970 (Bicentennial ed.). U.S. Government Printing Office.

New readers highlight: Rachel R, News Editor | Adriano P, Head of Data | Alessandra G, Creative Director | Jurriën M, CFO | Vedran J, CTO and +84 more amazing people.

Welcome!

When my family recently moved to Australia to support my wife’s academic career, I found the country’s attitude towards AI and the local tech startup scene in a weird place.

Thank God, my wife is faring pretty well. Her recent study on Sydney’s radioactive map was featured in the Sydney Morning Herald yesterday!

But AI… oh my.

Brilliant minds are not missing here. What is missing is the nerve. Australia was once in a remarkable position, with frontier science, top researchers, and a global centre for machine learning, National ICT Australia (NICTA). World-class. Genuinely competitive. Then, in 2014, the government defunded it. It died in 2015.

What followed was a diaspora: machine learning researchers scattered abroad, eventually in the early ranks of the AI successes we now associate with the US and the UK. Some ended up at DeepMind, which became “Google DeepMind” precisely in 2014. Some went to Silicon Valley. Some are now teaching at Mohammed Bin Zayed University of AI in Masdar City, the world’s first university dedicated entirely to artificial intelligence. In the UAE.

In 2017, while Australia was still figuring out (or forgetting?) what it had lost, the United Arab Emirates appointed the world’s first Minister of State for Artificial Intelligence. No task forces, committees, consultation papers, think tanks, or Web Summit roamings. A minister!

I remember reading this and thinking: Who does that?

Then I thought: No, the better question is, who doesn’t? Why, when the rest of the world was turning on the lights, did Australia reach for the switch and turn them off?

My recent trip to Abu Dhabi forced the question into sharper focus.

1. The printing press

I visited the Ministry of AI in Dubai, where our guide made a historical observation that stuck with me.

He described the Islamic Golden Age on one side. A civilisation at the absolute pinnacle of human knowledge — mathematics, astronomy, medicine, poetry, philosophy. Baghdad was the intellectual capital of the world. Algorithms are named after Al-Khwarizmi. Algebra is an Arabic word. The House of Wisdom made the European Renaissance possible by preserving and advancing what Greece had started.

On the other side: the printing press.

When Gutenberg’s press spread across Europe in the 1450s, the Islamic world resisted it, mainly due to cultural differences, theological objections, and a bit of guild interests. A sense that the existing methods of transmission were sacred, or at least sufficient. The resistance lasted decades in some places, centuries in others. Meanwhile, Europe printed, published, argued, reformed, discovered, and reorganised.

The reordering of nation-states that followed did not happen because European civilisation was inherently superior. It happened because it adopted and diffused a general-purpose technology while another civilisation, equally brilliant, chose not to.

The people at the Ministry of AI in Abu Dhabi know this history. It is their history. And they have decided, with what I can only describe as extraordinary lucidity, that they will not repeat it.

That is the level of self-awareness driving what is happening in the UAE.

Australia is living another printing press moment. Without even the dignity of a theological argument.

2. A Different Assumption

There is a question underneath all the noise about AI policy, safety frameworks, and guardrail consultations. It is rarely asked directly.

Here it is: do you believe AI is a general-purpose technology, or do you believe it is a product category, one more tool in the drawer?

The answer changes everything.

Electricity was a general-purpose technology. Steam was. The internet was. The printing press, as we’ve seen earlier. These are technologies that do not improve one sector; they reorganise the entire structure of the economy and society and the position of nation-states within it. The countries that understood this early and wired it into everything — their factories, their cities, their institutions — pulled ahead. The countries that waited for theoretical certainty, wavered for cultural reasons, or spent their energy drafting liability clauses spent decades managing someone else’s infrastructure.

The UAE has made its decision: AI is electricity. Therefore, the task is to build it, invest in it, and spread it before the compounding gap becomes too wide to close.

Hence, in 2024, Mubadala and G42 launched MGX, a sovereign-backed investment platform targeting semiconductors, compute infrastructure, and model-layer assets. Not a startup accelerator. National-scale capital, structured around a multi-decade mandate.

That same year, the European Union finalised the AI Act, the most comprehensive AI regulation ever written. The most comprehensive regulation of technology being built somewhere else.

I am sure Brussels is very proud.

3. The Electricity Parable, Or: Who Rebuilt the Factory

In 1882, Edison demonstrated electric lighting in lower Manhattan. In the decades that followed, American factories reorganised their entire floor plans around electric motors. This was not obvious at the time. The transition proved chaotic, expensive, and full of failures.

Other countries adopted electricity, too. But in some places, adoption stayed shallow for reasons that appeared perfectly sensible at the time. The gap did not announce itself. It compounded quietly, for decades, until it was simply too large to close by effort alone. By then, the debate over adopting electricity had become rather academic.

The UAE’s sovereign wealth funds — ADIA, Mubadala, ADQ — were built precisely for this kind of long-horizon bet. Oil revenues were not spent; they were placed into institutions that invest across generations. The logic is clean and worth stating plainly: oil is finite. The returns from oil, placed wisely, do not have to be. The sovereign mandate is to transfer wealth to all future generations of Emiratis. Think about that!

MGX exists because a diversified sovereign fund cannot concentrate heavily in a single frontier theme without distorting its entire portfolio. MGX can. It is built to go deep: semiconductors, compute, data centres, the model layer.

In every previous technology wave, infrastructure captured the durable returns. Applications changed and became commodities; some were disrupted, others got replaced. Whereas infrastructure endured to this day.

4. A Mistake That Sounds Like Prudence

There is a seductive logic to Europe’s position, and I want to take it seriously before I raise my doubts about its efficacy.

Set the standards first. Define the risks. Build guardrails. Then deploy. Reasonable people believe this. Some of them are my friends. I, too, was quite convinced by this approach for some time.

But the argument has a structural flaw. If most of the advanced chips, the frontier models, and the large-scale compute are built elsewhere, then European regulation mostly governs foreign production. You are writing the rules for a game being played on someone else’s court, with someone else’s equipment, by athletes who trained in someone else’s gyms. The referee matters less than you think, and less than the players know.

Moreover, risk becomes comprehensible through use rather than through theory.

Electric grids improved because engineers built them and fixed failures in real time. Aviation became safer because aircraft flew and incidents were investigated. The pattern is consistent across every engineered system in history: you learn by doing, under supervision, with accountability. You do not learn by waiting until the system is perfectly understood, because that moment never comes.

Therefore, the countries deploying AI in their ports, energy grids, hospitals, and procurement systems as operational layers are also the countries developing the most accurate understanding of where the risks actually reside. Exposure is part of risk management. It has always been.

Australia faces its own version of this problem, probably because, with its deep pension funds, careful governance instincts, and a resource-based economy, it has not yet felt the urgency. But, in a world where compute capacity becomes the new defining resource, what exactly is being exported?

5. AI Literacy

There is one argument for the regulatory-first approach that I genuinely respect. If ministers and civil servants do not understand what a language model actually does, what its failure modes are, what the cost of running it at scale looks like, they will either avoid it entirely or outsource it entirely. Both outcomes are bad. In fact, the second outcome — outsourcing without understanding — is arguably worse than not adopting at all. It is the worst of both worlds: dependency without capability.

Literacy is infrastructure. The White House understood this and embedded AI training across federal agencies. The UAE’s AI Strategy 2031 includes adoption goals across sectors precisely because technology adoption and diffusion need a workforce capable of using it.

A bureaucracy that does not understand the tools will not govern, regulate, or deploy them well. This is the strongest argument the regulatory-first countries have — and they are largely wasting it by leading with compliance requirements rather than capability building. Start with literacy. Then build. And the two feed each other.

6. The Only Question That Remains

Sovereignty used to mean control of territory. In the 21st century, it may mean something else: control of the layer on which everyone else depends.

If you do not own the infrastructure, you manage someone else’s. If your institutions cannot use the technology, they are governed by those who can. If your capital does not fund the build — the compute, the literacy, the diffusion — it ends up funding the application layer: volatile, replaceable, and always one disruption away from irrelevance.

The rebuilt factories pulled ahead because they understood that a general technology cannot be fully understood from the sidelines. You learn it by using it, at scale, with accountability, and the willingness to fail forward and fix what breaks.

Some countries are embedding AI into the structure of their economies.

Others are writing the footnotes.

The footnotes will not be finished in time.

Interesting sources to discover more:

• Abu Dhabi Launches Comprehensive Global Investment Strategy on AI

• Official MGX Launch Announcement

• UAE National Strategy for Artificial Intelligence 2031 and Ministry of AI guidance

• US Department of Labour’s AI Literacy Framework, released 13 February 13:

  • DOL press release: dol.gov/newsroom/releases/eta/eta20260213

  • Training and Employment Notice TEN 07-25 (the official guidance document): dol.gov/agencies/eta/advisories/ten-07-25

  • Full framework PDF: TEN 07-25 complete document

This post writing: L2 (learn about my AI scale here)

Dear readers and friends,

Welcome to the Human Override. Twice a month (or less), I share short reflections about AI and society, ending with the “Human Override” (strategies for leveraging AI, helpful tips, or ways to preserve humanity when everything turns synthetic). Paid subscribers get access to special essays and a monthly AMA session where you can ask me anything in a live webinar.

New readers highlight: Franco T, Director | David D, Marketer | Dr Maxime D, CEO & EU Sovereign Strat | Lisa DC, AI Product Leader | Jennifer R, Tech Journalist | Christina P, Strategic Partnerships & Philanthropy | Nick, Ian, Alex, Rachel and +209 more legends! Welcome!

STORY

In the early 20th century, the US and China adopted electricity at roughly similar technical levels. The difference was spread. In America, electrification ran through factories, homes, and cities within decades. Output jumped. In China, the spread lagged. Over time, the gap compounded.

That analogy came from the Chief Strategy and Safety Officer at MGX, a $40 billion AI fund placing the UAE at the centre of the current industrial shift. I heard it late one morning in Abu Dhabi, the city turning white-ish yellow against the Gulf, inside a tower at ADGM on Al Marya Island.

Their working thesis was blunt: AI reshapes economies and power structures the way electricity did. Winners adopt early, push it everywhere, and build the capital, compute, and rules that make it diffusable.

I spent two weeks in a country I respect and feel grateful toward. I lived in Abu Dhabi for almost six years, which is long enough to see what wisely used wealth, long-term horizons, and state intent can actually build. Perhaps it’s the advantage of not betting on fast wins that wins consensus, or makes a CEO popular at the next quarterly meeting. That is the advantage of investing in systems meant to last.

If you’re not familiar with the UAE’s approach to artificial intelligence, one fact matters more than most commentary admits: in 2017, it became the first country to create a Ministry of AI. Around the same time, Australia cut major funding for AI research. Timing has a sense of humour.

I went back recently. An unplanned trip that quickly turned into a wonderful event. I lectured at NYU Abu Dhabi on the political, economic, and social power of AI. I visited MBZUAI, the first university fully focused on AI, Khazna data centres (part of the Stargate buildout), the Ministry of AI, and MGX itself.

I came back with a bit more excitement about the opportunity of AI, a more hopeful view, but mostly, with a cleaner way to think about power.

The real centre of gravity

Most public debate tracks the next system or the next personality. People argue about agents, home setups with many Mac minis, and screenshots going viral (most made by AI… so, hard to tell what’s true and what’s not). After a few hours with MGX’s leadership, that frame felt small.

The contest that shapes AI’s future is economic, geopolitical, and infrastructural. The real centre of gravity sits where three forces meet: long-horizon capital, national infrastructure with policy permission, and deep use across institutions. That intersection matters more than model releases, celebrity founders, or viral tools.

MGX doesn’t talk about AI as a product cycle. They treat it as a general-purpose technology in the economic sense. Like electricity, the steam engine, the internet.

And the UAE’s position becomes clear when you see the combination:

• The state wants early and deep use, backed by money and policy.

• Capital is patient. Decades matter more than quarters.

• Rules are clear enough that projects don’t stall for years.

You don’t see that combination often. Silicon Valley builds systems. Places like this build the conditions that let those systems spread. By the end of the meeting, it was obvious this wasn’t for show; rather, it was a sound plan.

THE HUMAN OVERRIDE

If AI is changing how your industry works, five things matter.

First, treat AI as infrastructure.

If you see it as another tool or feature, you’re missing the point. It sits underneath how work gets done, how decisions are made, and what kinds of output are possible. Ask where it belongs at the core (operations, decision flow, products). Look for areas built on repeated mental labour and experiment with them systematically. It takes a lot of trial and error to realise the potential

Second, depth beats being first. History doesn’t reward early moves without follow-through. Map where AI actually sits in your organisation: core processes, support functions, or optional add-ons. Push it from the edges into the centre within a year.

Third, match your timeframes to the work. Short cycles don’t build infrastructure. If you’re funding AI internally or externally, judge it over two to three years. Measure how much it changes workflows and decisions, not just ROI in a few months.

Fourth, clear permission matters. Progress slows to a crawl when rules are vague, or overly cumbersome. Know what’s required where you operate. Start pilots in grey zones with strong internal controls, and push for standards that remove doubt.

Fifth, choose partners for reach. Convenience or your schoolmate as selection criteria doesn’t cut it. You’ll pay the price later. Capital at this scale pulls ecosystems together. Work with partners who give you access to compute, talent, and long-term alignment, and structure a deal with balance. The right relationship spreads use. The wrong one just adds features, or will colonise you.

SPARK - The frame matters

If AI becomes the base layer of this century, those who push it deepest into their institutions will shape what comes next. Using AI is not the same as being reshaped by it.

If your leaders are still thinking of AI as a way to boost productivity or replace jobs, they have the wrong frame. I wish fewer people focused on how to do the same things faster, and instead chose to explore and relentlessly experiment with what else becomes possible. Hours saved in inefficient meetings, communications, and decks can go into speaking with customers, creating new products, solving new problems, and entering new markets.

If AI is successful, the addressable market expands. If you’re blind to the opportunity, it won’t.

This is the first issue in a series on global AI strategy, economic adoption, job impacts, and institutional responses, including when regulation helps and when it doesn’t. This one is open. The next issues are for subscribers.

I’m also opening up the AMA session in February for everyone (usually reserved for paid subscribers only), so you can get a sense of what you're missing out. Sign up here.

Consider supporting Human Override:

Subscribe now

If this was useful, reply and tell me what you want the next piece to answer.

Hi dear readers and friends,

Welcome to the Human Override. Twice a month, I share short reflections about AI and society, ending with the “Human Override” (a way to preserve humanity when everything turns synthetic).

New readers highlight: Peter M, CMO | Helen M, SVP Consumer Products | Justin M, Chief Customer Officer | Bianca I, Founder | Shelly N, MIT Sloan MBA | Shally, Amy, Colin and +296 more legends! Welcome!

STORY

When Fei-Fei Li spent years hand-curating ImageNet—millions of images, labelled one by one—there was no AI to help. The work was tedious, often maddening. But something else was happening in that tedium. She was building intuitions that no shortcut could provide.

Three weeks ago, Li launched Marble: a world model that generates interactive 3D environments from text or images. Entire navigable spatial worlds, pretty cool, huh?!

ImageNet enabled the breakthroughs in deep learning, the AI architectures behind all of today’s marvellous technology, which came from years of struggle. Now we’re building tools that eliminate struggle entirely. What are we trading away?

Consider one developer’s confession. ”I’ve become a human clipboard,” he wrote. Top engineer, 14 years at Google, the kind who once relished hunting down gnarly bugs. Now errors flash, he copy-pastes into ChatGPT, receives the solution, pastes it back. Problem solved. Nothing learned.

A Microsoft and Carnegie Mellon study confirmed the pattern: the more people lean on AI, the less critical thinking they engage in. Some research or article I read long ago (sorry, I couldn’t find the source) used the term “learned technological helplessness.” The phrase stings because it describes a choice, while many internalise it as fate. We learn to be helpless.

Meanwhile, the infrastructure shifts quietly, outside the spotlight (still obsessively pointing to generative AI). China’s Darwin Monkey 3, a neuromorphic chip mimicking brain processing, outperforms traditional supercomputers at the edge. Intel’s Hala Point runs 1,152 neuromorphic processors. MIT’s photonic chips perform neural computations using light, completing tasks in under half a nanosecond. Humanoid robots are moving from lab trials to commercial deployment.

So, are we doomed to become human clipboards, or can we actually do more interesting things when machines can do what we once did?

THE HUMAN OVERRIDE

“Become an AI supervisor” sounds reasonable. In practice, it often means watching outputs, clicking approve, slowly losing the skills that made you valuable… while all becomes boring as hell!

There’s a better path.

Cognitive expansion

Take these two organisations: they use the same tools but achieve different, equally interesting outcomes.

Morgan Stanley’s financial advisors did not mandate leveraging AI copilots to handle triple the clients, but they focused on reclaiming time for complex strategy sessions that paperwork once made impossible. NVIDIA’s robotics team offloads simulation grunt work to AI, then channels freed hours into exploring robot behaviours no one has imagined.

Both ways expand the surface: they leverage AI gains to expand cognitive capacity. One approach creates human clipboards. The other creates human persons who are more capable than before.

What you can start today

1. The 90-day hands-on project

Pick one emerging area. Build something real within 90 days. If you’re in traditional tech, learn PyTorch (neuromorphic companies use it). If you’re in healthcare or manufacturing, map how AI integrates into your domain. Start before you feel qualified.

2. Protect your foundation

Set aside time weekly for manual work. One developer practices “AI-free Fridays”—no copilot, no ChatGPT. His debugging instincts stay sharp. Your foundational skills let you recognise when AI gets it wrong. Without them, discernment disappears.

I used to think foundation maintenance was inefficient, a waste of time. I’d better spend time on new capabilities. Wrong! The foundation is what makes new capabilities trustworthy.

3. Build hybrid intelligence roles

At law firms, valuable people know exactly when AI’s “good enough” will get a client sued. While Deloitte hasn’t learned it yet (perhaps they should have accepted my application to return to work with them two years ago… well, too late, mate!). The skill is transitioning from operating the tool to knowing its limits. Give Gradient Institute a buzz to train your team to discern AI capabilities and calibrate their trust in AI.

No PhD required. Lawyer + AI ethics = emerging role. Marketer + AI strategy = $95K-135K positions. Banker who knows when not to trust models becomes irreplaceable.

4. Cross-pollinate

Neuromorphic computing creates jobs that didn’t exist five years ago: chip designers who understand neuroscience, developers building spiking neural networks, and integration specialists deploying brain-inspired systems in robotics. Another cool trend I learned recently is that electricians, air conditioning, and other cooling systems are becoming important industries to support the gargantuan data centre build-up. There are tonnes of innovations still waiting to happen here, and there’s already a shortage of these skills.

5. Document Publicly

Blog, post, create videos. Companies hiring for emerging roles reward visible curiosity over polished credentials. Your learning journey is evidence you can adapt.

You’re on your own

Only 35% of US companies offer frequent role-specific training. Most organisations won’t build pathways for you. Waiting for permission is a losing strategy.

The companies getting this right deliberately restructure work so the human contribution becomes more intellectually demanding. If yours isn’t thinking this way, push, or find one that is.

SPARK

If an entire generation never knows the satisfaction of solving problems unaided, never experiences the hard-won understanding that comes from struggling, what becomes of breakthrough thinking?

Fei-Fei Li’s grunt work was the teacher. The tedium built into intuitions no shortcut could provide. What happens when we optimise away tedium entirely?

Maybe we evolve different cognitive muscles. Maybe the next generation finds forms of mastery we can’t imagine. Or maybe we’re about to learn what happens when we trade depth for speed.

I don’t know. But the question feels urgent enough to sit with—and honest enough to name.

Further reading

• Fei-Fei Li’s manifesto on spatial intelligence

• ”Avoiding Skill Atrophy in the Age of AI” by

• ”The AI Efficiency Trap” (Wharton)

• ”The Neuromorphic Wave” (Sombrilla Magazine)

• ”How AI is Rewriting Work in Finance” (Brookings)

Until next time, keep your hands on the wheel.

The Human Override is about staying capable, curious, and creatively alive in an age that wants to automate everything.

This newsletter will always be free. I don’t plan to include affiliate links or sponsor ads if this publication becomes more supported by paid readers, who also get access to an AMA session every 5 weeks and paid-only longer essays, like the latest Forms Carries Care. If you like my work, consider supporting me as a paid subscriber.

Subscribe now

This is a post for paid subscribers only, but heck, it was Black Friday yesterday, and Thanksgiving is on Thursday for our American friends. I’m giving it out for free to everyone! Only one caveat: I apologise if you haven’t seen Stranger Things, as I’m not sure you’ll be able to follow along (but, perhaps, it might convince you to watch the show).

Please get a few more Black Friday giveaways on this post - deals end at 23:59 everywhere on Earth on Monday, 01 December.

There’s a moment in every genuine crisis when the manual stops working.

It is when the situation has exceeded the frame. The manual is perfectly accurate for all the scenarios contemplated, but some circumstances can still break through even the most perfect plan. When that happens, when the rules fail, when the science doesn’t give you an answer, when you need enough certainty to orient yourself, you need something else. It’s something closer to faith.

Netflix released the fifth season of Stranger Things this Thanksgiving. The series is a five-season meditation on this moment. Beneath the nostalgia and the monster fights, the show keeps returning to a single question: what do you reach for when everything else–the science, parents, institutions–fails?

The Two Frameworks

The kids in Hawkins have access to two explanatory systems for the nightmare they’re facing. The first is empirical. Mr Clarke, their science teacher, explains radio waves, wormholes and alternate dimensions. Hawkins Lab, for all its horrors, generates real data. The scientific method applies: gather evidence, form hypotheses, and test them.

The second is mythological, Dungeons & Dragons, a game of imagination, storytelling, and shared make-believe. The Demogorgon becomes a monster the kids have already encountered in their imaginary campaigns. The Upside Down becomes the Vale of Shadows. The incomprehensible becomes clearer with the right metaphors.

What’s striking is the place each framework has.

The scientific one helps them with temporary tools that aid with difficult tasks or support educated guesses. The mythological one helps them comprehend the incomprehensible. These are different capacities entirely. One requires technique (technology). The other requires meaning (metaphysics).

The Mystery at the Centre

Eleven represents a mystery the show never fully resolves because it can’t be thoroughly exhausted.

No one knows where her powers originated. Perhaps they came from the psychedelic drugs the lab gave to her mother. The doctors at Hawkins Lab may have helped her expand and harness them. That lab is the most reductive environment imaginable. Everything there is measured, controlled, quantified. Brenner treats her as a subject to be studied, the military as a weapon to be calibrated.

What emerges from that crucible transcends every framework that produced it.

Eleven’s abilities cannot be replicated, cannot be democratised, cannot be reduced to method. They exist outside the system that created them. And they’re the only thing capable of closing the wound that system opened.

This is the show’s deepest statement: the irrational remnant that survives reduction becomes the source of salvation.

I’ve sat with this idea for years, since watching the first series. It unsettles the part of me that wants solutions to be scalable, transferable, applicable. Eleven’s gift is none of these things. It simply is—miraculous, necessary, unteachable.

Maybe that’s the point. Some things remain a mystery, and when part of that mystery is revealed, a greater one unveils.

The Cognitive Trap

Today, we’re chasing a response to the work and existential anxieties posed by AI the way we respond to every disruption: we frantically try to measure the change, try to capture it, and we try to predict as much as we can about it.

Learn to code. Study machine learning. Acquire AI literacy. Upskill before the wave hits. Pseudoscience about philosophical goals, such as the artificial “general” intelligence, or the latest trendy word, “super” intelligence. Badly hiding the assumption underneath that if we understand the technology thoroughly enough, we’ll survive its disruption.

Research suggests something more uncomfortable. AI anxiety correlates most strongly with techno-complexity and cognitive overload. The more we try to comprehend these systems rationally, the more overwhelmed we become. And the more we use these systems, the more we increase the overload. I often ask myself, where is the progress, the productivity, the time and other liberations the technology complex promised us for decades? Why are we working more, and are we way more anxious than previous generations, while having marvellous tools our parents and grandparents could only see in naive renderings over Star Trek movie flicks?

We’re caught in a loop: the cure intensifies the disease.

This doesn’t mean learning is useless. Quite the contrary. But it implies understanding is insufficient. And our response has overweighted it catastrophically, as if knowledge alone could carry us through a transformation this profound.

What the Mind Flayer Knows

The Mind Flayer is the show’s most chilling creation because it operates like every system optimised for scale.

It doesn’t hate. It doesn’t feel. It processes. It reduces individuals to nodes in a network, consuming particularity to create coordination. Each person it touches becomes less themselves and more an instrument of the collective. But the orchestrator behind Vecna indeed hates humanity and wants to control it all.

This is what efficiency looks like when applied to consciousness.1

The party defeats it through anti-optimisation. Through responses that make no strategic sense, at least from a rational point of view.

Joyce believes Will is alive when every piece of evidence says otherwise. She hangs Christmas lights across her house and talks to walls. Anyone watching would call her delusional. She’s right, and her rightness emerges from something no algorithm can recommend: a mother’s certainty that transcends evidence.

Hopper stays to close the gate when escape is possible. The tactical choice is obvious—run, survive, fight another day. He doesn’t take it. His sacrifice is irrational by every metric that would matter to a Mind Flayer.

The characters’ decisions that make them heroes defy any dry calculation. They are made out of love, belonging, and redemption.

Notice who doesn’t save Hawkins instead. Not the government. Not the military. Not the schools, nor even the parents (except for Hop and Joyce). The institutions that should protect these children are corrupt, absent, or actively hostile. Brenner weaponises Eleven. The Department of Energy covers up the truth. Local authorities dismiss Joyce as hysterical. School and parents find it hard to have any meaningful line of communication with the kids, except for Michael and Nancy’s mom and Mr Clarke, the science teacher. They almost seem like protecting them from growing up–the very thing they should be educating and accompanying.

This resonates because it mirrors something we sense but rarely articulate: the structures meant to serve us increasingly serve themselves. The show doesn’t argue for chaos or anarchy, but it convincingly shows that when institutions fail their purpose, salvation comes from smaller loyalties. A party of friends. A mother who won’t stop searching. A cop who remembers why he became one.

The 1980s as Lost Capacity

The period setting means something specific.

The 1980s in Stranger Things represent the last cultural moment before total digitisation, before every mystery became an information problem. Before the unknown collapsed into a category (in truth, a fallacy) called “insufficient data.”

The Upside Down terrifies because it resists mapping. But Eleven’s powers evoke something else entirely—something we’ve almost lost the word for.

Awe.

We’ve built a culture that treats every unknown as a failure. A gap to be closed, a problem to be solved, an inefficiency to be eliminated. While this might serve the modern obsession with productivity, it impoverishes the soul. It reminds me of a sentence that stuck with me since I was 14. In my home country, Italy, a large annual cultural festival is held every August in Rimini, Italy. “The Meeting of Rimini” (officially: Meeting for Friendship Among Peoples) began in 1980 and is organised by the Catholic movement Communion and Liberation, but it is not a religious conference in the narrow sense. It’s a broad, public forum. Its 1999 edition was titled:

“The unknown generates fear, the Mystery generates wonder”

(The Italian original: “L’ignoto genera paura, il Mistero genera stupore”)

Exactly that awe we need more of.

Stranger Things reminds us that human beings wrestle with two kinds of unknown: the darkness we resist and the light we trust without understanding. When we eliminate the second category, we lose our capacity to tolerate the first.

So what prepares us for a future this uncertain?

Technical literacy helps. The kids still use walkie-talkies; Mr Clarke’s lessons still matter. I’m not arguing for ignorance.

But we’ve neglected the other side of preparation almost entirely.

The capacity for awe. Communities that value presence over productivity. Relationships that refuse to optimise. Practices—creative, embodied, spiritual—that remind us we’re more than computation.

And perhaps most importantly, frameworks that honour mystery rather than eliminate it. The willingness to reach for meaning over the breathless search for scientific answers. As if science could answer the unfathomable depth of the human desire for purpose, freedom, and happiness.

The Wound

The Upside Down is a metaphor for the ominously ever-present darkness within us and, by extension, in the world.

We built it ourselves, one system at a time. And in the past twenty years, a new form of it emerged: every algorithm that reduces the person to a profile. Every optimisation treats attention as a resource to be extracted. Every layer of automation that processes human beings as inputs.

The breach is real. Scary. But Stranger Things teaches that breaches can be closed. They just require something other than the tools that opened them.

When Eleven confronts the Demogorgon at the end of Season 1, she doesn’t reach for any tools or the gods technology. She reaches into something beyond knowledge—call it will, call it spirit, call it the stubborn human insistence that we are more than what can be measured.

Every character’s story arc in the show seems to argue that this capacity exists in all of us, not just the telekinetic girl from the lab. The capacity to love without calculation. To sacrifice without strategy. To believe in what we cannot prove.

Stay Weird, Remain Strange

We face uncertainty in jobs, a fraying social fabric, institutions we no longer trust, and machines that already exceed us in ways that matter (actual machines or more figurative machinery).

But a question keeps surfacing in me: can we remain strange enough to matter in ways that can’t be computed?

The monsters are real—we’ve built them, trained them, deployed them at scale. So are miracles, though we’ve grown embarrassed to use the word. Something in human consciousness that persists despite every attempt to reduce it.

The kids in Hawkins prevail because they refuse the obvious lesson—that power belongs to the one with superior agency. They bring mythology to the fight against the unknown. They bring love to their decisions, facing unbelievable stakes. They bring irrational courage to a threat that cannot be understood, nor defeated, by rational means.

There’s another layer the show earns across its seasons: we watch these children grow up. Not just their characters but the actors themselves. The wonder and terror of Season 1 mature into something more complex and more grounded by Season 5. Fear doesn’t disappear. It transforms into responsibility. The party moves from hiding in basements to fighting in the open air and scary tunnels.

This is what coming of age actually means: stop hiding vulnerability and discover that the only stability worth having isn’t safety. It’s true friendship oriented toward a shared purpose. The kids don’t outgrow their strangeness. They learn to wield it together.

That’s both the nostalgia we feel when watching the show and the instruction.

The future belongs to those who remain strange enough.

As Honest AI has grown in trust and readership, we’ve started receiving several contributor pitches. If you’d like to write for Honest AI, you can share your idea here.

This article was translated into English using Google Translate, edited by , and finally signed off by the author. For the Italian readers, you can jump to the end to read the original article.

Grazie, Gioele!

How many of us have seen The Matrix? Putting aside the philosophical and computer science references the film is packed with, I want to highlight the general setting in which the story unfolds. We find ourselves in a future where humanity, after an unparalleled technological leap, ends up at war with its own creations—machines that have become so “evolved” they have subjugated and enslaved almost the entire human race.

Those who escape this fate live underground, waiting for a messiah, “The One.” The entire plot revolves around this war: Man vs. Machine, Creator vs. Creature.

The current AI boom mirrors the rise of LLMs. But what are they? What does this curse word actually mean? LLM stands for Large Language Model. These are algorithms—very complex, yes, but ultimately just mathematical formulas—that humans invented and exploited thanks to the ever-increasing computation speed of computers. AI, being the finished product of calculations performed by multiple servers, is purely virtual.

Ethics and Washing Machines

I want to bring the focus on ethics’ so let me first define it. Here, we define Ethics as the ability to reason and discern between good and evil. Consequently, speaking of “AI Ethics” is a delusion and a contradiction in terms.

Consider this: If we load a washing machine with glass tumblers, run a spin cycle, and then get angry because everything is broken, whose fault is it? Is the washing machine responsible? Did it fail its “ethics”? No. It performed the purpose for which it was programmed. The machine is innocent; the responsibility for the error is ours.

This simple example demonstrates three fundamental characteristics of AI:

• Amoral: It acts regardless of implications. It is without “amore” (love).

• Contextless (Deficient): It is “deficient” in the true Latin sense of the word (deficio, meaning “lacking”). It cannot understand context or intent from a prompt alone, nor can it mitigate its response based on nuance.

• Without Ethics: It cannot know if its actions are right or wrong.

Why do we struggle to judge AI clearly? Without trying to untangle the entire Gordian knot, let’s look at the symptoms.

When we chat with AI, it responds as if it were human, using “I” and feigning consciousness or feelings. This generates a false sense of empathy in our brains. It works both ways: men abuse female-coded AI, and conversely, 70% of users on the AI companionship app “Dippy” prefer male characters, suggesting many users are women seeking connection (Wired, 2024). And that happens because AI looks like “on of us.” We have deceived ourselves: we have empathized with the machine, attributing to it feelings and stirrings of the human soul that it does not possess. Let’s remember it is just a machine. That is, ultimately, just a string of 0s and 1s on electronic devices called computers and servers.

But I want to offer another reflection.

The New Human Sacrifice

We treat AI like a modern Oracle, expecting it to solve all our problems. Most people don’t realize these models are trained on data up to a specific time. Ask about something recent, and it either hallucinates (invents facts) or searches the web just like we would. Research shows that AI accuracy drops drastically as problems become more complex or unique. We have turned AI into an Oracle that provides ordinary answers to ordinary problems.

This reveals a hard truth we could learn by simply slowing down: our problems are rarely unique. Someone else has faced what we are facing. The AI knows the answer only because it has read thousands of human texts discussing the same struggle.

We look back at ancient civilizations—the Phoenicians, Egyptians, Aztecs—and smile at how they anthropomorphized unknown forces, elevating them to gods and offering them sacrifices, often human ones, to secure a future.

Today, we are doing the exact same thing scientifically and massively. We are sacrificing our privacy and our children’s futures to the deity of Information Technology. We feed AI models with our habits, photos, and intimate data to make them “smarter,” all while rendering ourselves more tracked and commodified for the economic benefit of a few tech owners.

The “Sharenting” Crisis
Think of a child today. How will they reach adolescence when their parents have already shared their entire young lives online? Anyone can know their most intimate aspects. What moral authority will we have as educators if we have splashed our own indiscretions everywhere? How is redemption or a “new life” possible if our digital past constantly returns to accuse us?

In our narcissistic and nihilistic society, the sacredness of Human Life has been ousted by market value. To quote Nietzsche, “God is dead”—or rather, Man has killed God and, deceived by the “Ancient Serpent,” believes he can be God.

By elevating ourselves to divine status, we have lost the horizon of the transcendent. We find ourselves naked and afraid, chasing virtual worlds. We ignore the physical cost: while we talk of a green transition, we ignore the massive energy consumption required by data centers to power this technology. We act like sheep, accepting mediocrity as long as we get our scraps.

Artificial Deficiency (AD)

The core pathology is that we have discarded our divine principle and attributed human roles to our artifacts. We call it AI, but it is really AD (Artificial Deficiency). If we have lost our own ethical roots, how can we judge the ethics of our machines?

This leads to two consequences:

1. The threat of control becomes real simply because people believe in it and follow it (like stock market speculation).

2. We become lonelier, cynical, and afraid of the future.

We must ask: Who is looking out for the youth? Do they really need this technology as urgently as we think, or is our euphoria instilling fear in them?

The Tool, Not the Idol

Is this technology absolute evil? Absolutely not. If we can restore the centrality of the Human and the Divine Principle in our society, we can use this tool as a tool, without deifying it or fearing a Matrix-style rebellion.

I leave you with an invitation and a provocation:

The Invitation
Do not fall into the trap of empathy. The machine has NO HEART. It has neither humanitas nor pietas. It can describe these concepts with perfect words, but it remains a cold instrument without a soul. Remember that.

The Provocation
Isn’t it paradoxical that we are working tirelessly to give AI human-like features, while simultaneously building unreal, virtual worlds for ourselves to live in? Do we really want to virtualize ourselves while giving physicality to the virtuality of AI?

I don’t know if the future will be like The Matrix, but today, are we already blocking out the Sky?

Gioele Raiano is a software developer, and a curious mind who values dialogue with specialists across fields.

Original version

Deficienza Artificiale

Chi di noi ha mai visto il film Matrix? Lasciando da parte tutti i riferimenti filosofico-informatici di cui il film ne è pieno, vorrei invece evidenziarne l’ambientazione generale in cui la storia si va dipanando. Siamo in un futuro in cui l’umanità, dopo un salto tecnologico senza eguali, si ritrova in guerra contro le proprie opere, ossia le macchine, divenute tanto “evolute” da soggiogare e rendere schiava praticamente la totalità degli umani. Chi non è soggetto a simile destino, oltre a vivere sottoterra, è in attesa del messia, dell’eletto.

Tutta l’azione quindi si svolge all’interno della guerra uomo vs macchina, della creatura vs il creatore.

Il boom dell’IA è specchio di quello occorso agli LLM. Ma cosa sono, cosa significa questa parolaccia? LLM sta per Large Langue Model e fa riferimento alla creazione di grandi modelli linguistici, ma cosa sono questi modelli linguistici? Gli LLM sono algoritmi (molto complessi ma pur sempre formule matematiche) che l’uomo ha inventato e definito potendo sfruttare le capacità e velocità di calcolo dei computer. Pertanto l’IA, essendo il prodotto nel senso di manufatto finito di una serie di calcoli fatti da molteplici server, è puramente virtuale.

Volendo portare il focus del discorso sull’etica, fissiamo in primis cosa definisco con etica. Per ETICA qui intenderemo la capacità di ragionamento e discernimento nel capire cosa sia buono e cosa male, ne segue quindi che parlare di ETICA dell’IA è un inganno a noi stessi e un controsenso in termini propri. È come se noi caricassimo una lavatrice con dei bicchieri di vetro, le facessimo eseguire la centrifuga e poi ci arrabbiassimo perché sono tutti rotti.

Ora, di chi sarebbe la responsabilità del danno? Della lavatrice? Essa ha mancato alla propria “etica”? Allo scopo per il quale è stata progettata e programmata? La risposta in questo caso è ovvia: la macchina è innocente e la responsabilità dell’errore è nostra.

Paradossalmente, con questo semplicissimo esempio, abbiamo dimostrato tre caratteri fondamentali dell’IA. Essa è:

• Amorale: agisce sempre e a prescindere, senza alcuna preoccupazione circa le implicazioni e conseguenze. Amorale: senza amore.

• Senza Contesto: è deficiente, nel senso latino del termine: deficio ossia mancante. Non è in grado, dalla sola domanda, di capire il contesto ed il fine e, in conseguenza, mitigare e affinare la risposta.

• Non Etica: non è in grado di sapere se il suo agire sia bene o male, giusto o sbagliato.

E allora, davanti a queste palesi evidenze, perché non riusciamo ad avere né ad essere così lucidi, fermi e decisi nel giudizio dell’IA?

Senza la presunzione di riuscire a dipanare la Tela di Arianna, vedremo, con pochi semplici esempi, come la risposta sia “molte manifestazioni di una sola patologia”.

Iniziamo col dire che, quando dialoghiamo e interroghiamo le varie IA, ad esempio sul futuro o su argomenti prettamente umani, esse rispondono includendosi nel dialogo come se, in qualche modo, avessero coscienza, capacità di ragionamento e sentimenti. Come se le loro preoccupazioni fossero le stesse delle nostre ossia come se noi e loro fossimo tutte persone umane. Questo primo sintomo è uno dei più cruciali.

Il fatto che l’IA risponda come se fosse una persona umana genera nel nostro cervello, a livello più o meno inconsapevole, l’inganno dell’empatia e della vicinanza. È cosa nota che ci sono persone di sesso maschile che sfruttano l’IA come sesso femminile da aggredire, umiliare e svilire ma è anche vero il contrario: dichiara Jagga, fondatore di Dippy app IA di “compagnia”, “il 70% degli account di Dippy tende a preferire personaggi maschili, il che potrebbe indicare che molti utenti si identificano come donne” (Wired, 2024). E questo perché l’IA è “una di noi”. Ci siamo ingannati: siamo entrati in empatia con la macchina, abbiamo riconosciuto alla macchina dei sentimenti, dei movimenti dell’animo umano che essa non ha. Ricordiamoci che, in ultima analisi, essa è solo una serie di 0 e di 1 su dei dispositivi elettronici chiamati computer e server.

Altro spunto di riflessione che desidero portare in evidenza. L’IA ci sembra onnisciente e la panacea a tutti i nostri problemi: chiediamo ed essa risponde a mo’ anzi meglio degli antichi Oracoli! I più di noi però non sanno che i modelli, le varie versioni di IA, vengono addestrati con informazioni esistenti al tempo specifico X. Al tempo X+1 il modello non ha risposta e deve cercarla su internet, come farebbe ciascuno di noi, oppure fornisce una risposta senza né capo né coda, a volte anche inventandosi le fonti dalle quali ha preso spunto, in gergo tecnico si dice che ha avuto un’allucinazione. Inoltre, una recente ricerca ha dimostrato che la percentuale di accuratezza delle risposte tecniche fornite in ambito di sviluppo cala drasticamente alla complessità, all’unicità del problema e di quanto esso sia temporalmente più recente rispetto dal momento dell’addestramento. Abbiamo reso l’IA il nostro Oracolo moderno che fornisce risposte ordinarie a problemi ordinari ossia comuni a tutti noi.

Un altro inganno al quale potremmo facilmente ovviare rallentando un momento le nostre vite e che l’IA stessa, essendo addestrata su migliaia di testi umani, ci porta in evidenza: il nostro problema NON è unico, non è il nostro esclusivo. Qualcun altro sta o è stato come noi, nella medesima situazione. Qualcun altro si sta ponendo le nostre stesse domande.

Ho definito l’IA come il nostro Oracolo Moderno ed ho altresì fatto riferimento alla situazione empatica di questa tecnologia rendendo a noi così ardua la trattazione etica.

Noi oggi bonariamente ridiamo e guardiamo ai Popoli Antichi le cui credenze e sistemi politico-religiosi tendevano ad antropizzare, a dare fattezze e comportamenti umani, praticamente ad ogni Forza Ignota e queste venivano elevate al rango divino e a loro l’uomo offriva sacrifici imbonitori, sacrifici spesso umani: schiavi, prigionieri, giovani donne vergini, principesse e principi… pensiamo ai Fenici, agli antichi egizi, ai miti greci ma anche alle Civiltà Precolombiane come gli Aztechi, giusto per citarne una. Loro sacrificavano il futuro dei propri figli e noi oggi, in modo scientifico e massivamente, stiamo facendo la medesima cosa: stiamo sacrificando non solo noi stessi ma anche il futuro dei nostri figli. Basti pensare ai dati che condividiamo e che, più o meno consapevolmente, cediamo come, ad esempio, abitudini e siti online ma anche foto nostre, dei nostri figli, nipoti, amici. Tutti questi dati sono una medaglia che, se da un lato, viene usata come cibo con il quale i modelli di IA vengono fatti crescere e istruiti affinché forniscano risposte migliori, dall’altro, con l’idea di essere più “smart”, moderni e giovani direi: “all’altezza dei nostri tempi”, sono un modo per renderci sempre più schiavi e conosciuti, più tracciati e assimilati ai dati e ai numeri che produciamo: senza dignità, senza umanità per il vantaggio economico dei pochi proprietari e di questa tecnologia e dei modelli matematico-statistico di cui fa largo uso.

Pensando ad un qualsiasi nostro figlio di oggi, mi chiedo: come arriverà all’adolescenza dopo che già noi genitori gli abbiamo condiviso tutta la giovane vita online?! Praticamente chiunque potrà conoscerne anche gli aspetti più intimi e privati.
E questo senza pensare a quello che un domani i nostri nipoti o figli potranno dire delle nostre “bravate di oggi”! Quale sarà e qual è la nostra autorità di educatori se avremo spiattellato ovunque bravate e pensieri?! Come sarà possibile la redenzione, un cambiamento, una nuova vita se ad ogni passo avremo una qualche informazione tornata dal passato ad accusarci: chi siamo, cosa siamo stati e a ricordare quanto accaduto?!
Stiamo sacrificando la privacy e le informazioni nostre, e non solo, alla divinità Tecnologia Informatica, all’IA come se questa possa salvarci o elevarci al rango divino. Tutto questo oggi è palesemente sotto i nostri occhi, nelle nostre società narcisistiche e nichiliste dove la centralità e sacralità della Vita Umana (che ha un inizio, uno svolgimento ed una fine) sono state spodestate dal denaro, dalla produzione compulsiva dei valori economici e dal mercato.

Oggi al tempo della secolarizzazione ecco che sembra essere diventato reale l’irrealizzabile, per citare Nietzsche: “Dio è morto”, meglio: l’Uomo ha ucciso Dio, l’Uomo si è lasciato ingannare (ancora) dal Serpente Antico secondo le cui menzogne noi possiamo essere come Dio. Ecco quindi che, essendoci alzati al rango divino, abbiamo perso l’Orizzonte Alto del Cielo, la nostra parte trascendete (anche se poi alziamo torri e grattacieli sempre più alti quasi a cercarla inconsciamente) e ci troviamo nudi e impauriti a rincorrere Mondi Virtuali la cui promessa di idealità è sempre più spostata a domani ma, al contempo, affinché la promessa di “domani” possa realizzarsi è sempre più necessario il fomento, l’assiduità e il soggiogamento di tutti in favore del guadagno di pochi.

Ragioniamo anche sulle bugie o le verità non dette: parliamo tanto di transizione verde, di decarbonizzazione del mondo ma quanti sanno, quanti conoscono l’energia necessaria ai vari datacenter solo per il funzionamento di questa tecnologia? Siamo consapevoli dell’inquinamento prodotto oppure, come pecore al pascolo, evitiamo domande e ci accontentiamo della mediocrità purché si abbia il pezzo di pane?

Ecco, quindi, che le varie argomentazioni superficialmente qui trattate sono le diverse facce della medesima patologia: cestinando e disconoscendo il Principio Divino, oggi l’Uomo ha perso se stesso e riconosce ai propri artefatti, tra i quali l’IA o meglio la DA (Deficienza Artificiale) che quasi brilla di luce propria, un ruolo e comportamento che non hanno. Se abbiamo perso, cestinato o disconosciuto i nostri valori storici, le nostre radici, la nostra etica (di società, di popoli e personale) oggi tutte cose così vetuste, come possiamo adeguatamente valutare e giudicare l’etica, se esiste, dei nostri artefatti?!

Questa difficoltà ha due conseguenze immediate:

• l’aspetto di minaccioso controllo (dell’IA, della divinità Tecnologia) da ipotesi diviene e diverrà tanto più reale e certo quante più persone lo seguiranno, un po’ come per le speculazioni in borsa;

• diveniamo sempre più soli, egoisti, cinici, vuoti dentro, sospettosi e impauriti del domani e di ogni cosa che pensiamo possa minacciare la nostra vita, il nostro status quo e benessere. Quindi, eccoci qui a scrivere e ragionare su noi e di noi adulti in rapporto alla modernità ma chi si è preoccupato dei nostri giovani? Di coloro che oggi stanno entrando nella vita adulta? Ci siamo accorti della paura che la nostra euforia per la DA sta ponendo in loro?! Di come LORO vedono e vivono questo aspetto della tecnologia da noi creata e in creazione? Ne hanno davvero un bisogno così urgente e viscerale che merita il nostro sforzo così importante?

Quanto fin qui esposto vuol quindi bollare questo artefatto come male assoluto?! Assolutamente no! Se saremo capaci di porre l’Uomo (e non il dato che è, che rappresenta o che potrebbe generare), se saremo capaci di far ritornare la Sacralità della nostra Vita ed il Principio Divino che alberga al centro delle nostre società, dei nostri modelli di business saremo allora capaci di adoperare lo strumento (ed ogni altro frutto dell’ingegno) come tale senza alcuna divinizzazione né conflitto o paura che l’artefatto si ribelli all’artigiano.

Vorrei quindi concludere indicando un aspetto ed un invito che trovo fondamentali:

• [invito] non cadiamo nell’inganno, nell’illusione dell’empatia con la macchina, essa NON HA cuore! Non ha né humanitas né alcuna pietas, può descriverle con le parole migliori ma resta fredda, uno strumento senza cuore né anima. Ricordiamocelo.

• [provocazione] non è paradossale come l’uomo stia alacremente lavorando per dare fattezze umanoidi all’IA, ad una serie di calcoli matematici, mentre per se stesso si stia costruendo una serie di mondi irreali e virtuali basati su una serie di 0 e 1?

Davvero vogliamo virtualizzarci e dare fisicità alla virtualità dell’IA?

In ultima analisi: non so se il futuro sarà come presentatoci in Matrix ma oggi ci stiamo forse oscurando il Cielo?

Grazie per l’attenzione.

Gioele Raiano

Hi dear readers and friends,

Welcome to the Human Override. Twice a month, I share short reflections about AI and society, ending with the “Human Override” (a way to preserve humanity when everything turns synthetic).

New readers highlight: Haygun, AI Specialist | Michael, CTO | Marco, Creative Leader | +64 anonymous legends! Welcome!

STORY

I’ll start with an uncomfortable self-reflection. I spent years collecting degrees and building expertise across physics, statistics, actuarial science, business and computer science. And the more I learned, the worse I got at admitting what I didn’t know.

That’s the paradox of expertise—the very thing that makes you credible also makes you blind.

47% of AI experts say they’re more excited than concerned about AI in daily life. Only 11% of the public feels the same (data from Pew Research).

Who’s right? Maybe neither. Or both. Perhaps experts are just really bad at sharing their enthusiasm with the public!

Too many experts I know in AI repost everything Sam Altman, Dario Amodei or Elon Musk says. They get excited when a billionaire CEO drops a vision of the future. Or they cite academics, “godfathers of AI”, like scripture.

But these same experts have apparent conflicts of interest (e.g., research grants to protect, companies to promote, products to sell). There’s massive self-serving interest baked into their narrative. If (when) the house of cards falls, they can’t easily pivot to other work, and “everyone’s got a mortgage to pay” (cit. Thank you for smoking—watch that movie if you haven’t!).

So here’s my confession: I’m nobody special. I don’t run a billion-dollar company. I’m not trying to stay quotable for Wired or keep investors calm. I’ve been an entrepreneur, raised capital, shipped products—but I’ve also taught, debugged, and watched AI evolve from multiple continents and cultures.

I’m a nobody with a newsletter, which might be precisely why you should listen. I have no conflict of interest.

THE HUMAN OVERRIDE

The paradox of expertise happens when deep knowledge causes tunnel vision, bias, and loss of flexibility. You can become so confident that you ignore new information that doesn’t fit, leading to poor predictions.

The Dunning–Kruger effect works both ways. Beginners think they’re more capable than they are because they don’t see the complexity. True experts often underestimate themselves because they do see it.

I’m not immune to this. Writing a newsletter warning people not to trust experts—while asking them to trust me—is pretty ironic.

A Framework to identify who you should actually listen to about AI

Strip away titles and headlines. Here’s the 5-Factor Filter I use:

1. Hands-on experience, little skin in the game

Look for people who can afford to be wrong. If their opinion is tied to stock prices, political office, or tenured legacy, they’ll defend their story even when it’s outdated (or bluntly false).

2. Simplicity

Deep understanding unravels complexity to its essence. If someone needs jargon to sound smart, they probably don’t understand it well enough.

3. Multi-disciplinary rigor

AI intersects economics, linguistics, neuroscience, computer science, engineering, philosophy. If someone pontificates about the future of jobs but has never studied labor economics or policy, take their optimism with a pinch of salt.

4. Reputation across domains

Can this person speak across fields without sounding like a tourist? Authority isn’t earned through TED talks. It’s built by doing serious work that’s respected by different communities (not just Silicon Valley).

5. Humility to update views

Watch how they respond when proven wrong (like Sam, here). Thought leaders worth following evolve. Prophets don’t.

AI isn’t a single field. It’s a system of interlocking impacts. You wouldn’t trust a neurosurgeon to design your national pension system. So why trust a chatbot creator to predict the economic future of humanity?

SPARK

What if we trusted the quiet experts instead?

We’re trained to listen to the loudest voices. But what if the best insight comes from people who don’t get media training?

What if the future of AI governance depends less on founders and futurists—and more on teachers, scientists, philosophers, and civil servants who work without fanfare?

Here’s your weekly provocation:

Who do you quote when thinking about AI? And why do you trust them?

Let me know by replying, and I’ll share with you my list!

Further Reading

• The paradox of human expertise: why experts get it wrong

• Power, Knowledge and Technology in a Finite World

• AI Acceleration Vs. Precaution

The Human Override will always be free. I don’t plan to include affiliate links or sponsor ads—just honest, informal talk. This publication is supported by paid readers, who get access to an AMA session every 5 weeks and paid-only longer essays, like the latest Forms Carries Care. If you like my work, consider supporting as a paid subscriber.

I have an embarrassing confession to make…

STORY

I almost sent you garbage last week.

Custom GPT open, half-formed thought, self-imposed deadline. I fed it my notes, watched it generate seven passable paragraphs, hovered over “send.” The output was median. Reasonable, sensible, helpful even, but tasteless. The kind of slop clogging every feed, written by nobody for nobody.

Then I stopped and read it a few more times and realized that the AI did its job and I failed.

I’ve been complicit in the problem I claim to hate. I built that custom GPT for these newsletters. I’ve crammed “good enough” content at the last minute, telling myself the curation and thinking were there even when the care wasn’t. And I’m exhausted reading the same average content everywhere, the same exaggerated claims backed by the same nothingness. I bet you’re damned tired of it too.

Building an edge has always meant doing what nobody else is doing. I was excited about NVIDIA when it was “just gaming chips” and Tesla when electric cars were a joke (unfortunately, I didn’t have much to invest back then, but I saw the asymmetry). I left a safe career at Deloitte in 2015 to start a tech company when everyone said it was career suicide. I started a PhD in dialogue systems right as the NLP papers today considered the foundations of GPTs and genAI dropped between 2017 and 2018, when most people thought natural language was a dead end.

Everyone told me these were bad ideas, stupid bets, reckless moves, but all delivered asymmetric returns years later.

So what’s the contrarian move now, when everyone’s chasing AI fluency, when every consultant is upskilling, when the herd is moving in one direction?

Being real when the world goes synthetic.

I’m tired of the average, the tasteless output splattered everywhere. So screw it.

THE HUMAN OVERRIDE

Screw the regularity, the volume, the consistency—everything the growth gurus preach about becoming a thought leader online. Life isn’t regular. Life doesn’t give you time for volume. Life isn’t consistent. And pretending otherwise is how we end up with feeds full of nothing.

I’m committing to a new way of writing and creating. Call it a human override of myself.

I write when I can bring care, not otherwise. You’ll get this newsletter every other week, sometimes less. I’ll miss issues when life gets in the way. That’s not a bug—it’s the entire point. I refuse to produce for the sake of producing, to add to the pile of median content we’re all drowning in.

I’m also exploring the edge of creativity and tech more deliberately. My family has a creative vein running through it. My dad ran a theater gig performing for kids in hospitals or people with various disabilities—not exactly glamorous, but he brought joy to audiences nobody else cared about. A brother makes films. Another acts. Another’s a chef. My sister’s a marketer. When I look at all of them, I sometimes wonder: what the hell am I still doing in tech?

Well, let’s unite it. I want to explore where tech supports human creativity and where it kills it. That’s the edge I’m most curious about now: the place where human imagination either flourishes or gets flattened by our tools.

Here are six human overrides you can try for the next couple of weeks. Just pick the ones that sting a little, the ones that expose where you’ve been settling for good enough:

1. Pick one project where “good enough” isn’t enough. Spend double the time it deserves. Refine until it sounds like you, until someone reading it would know it came from you and nobody else. Notice how much harder it is—and how much better the result feels.

2. Track your AI dependence for one day. Where are you outsourcing thinking versus accelerating it? Where are you letting the tool do the wrestling for you? Know your minimum ratio for work that rises above average. Mine’s 70/30 human/AI. Find yours.

3. Teach something you just learned. Write three paragraphs explaining a recent insight to someone outside your domain. If you can’t make it clear without jargon, you don’t understand it yet. Teaching forces synthesis; synthesis deepens mastery.

4. Name one thing you won’t outsource to AI. Not a category of tasks—one specific thing. The core of your work that reflects your judgment, your lived experience, your moral compass. Write it down. Treat it as non-negotiable.

5. Build one thing by hand this week, or cook a fine dinner. Something where speed doesn’t matter, where you can feel the difference between “done” and “done well.” Where care is visible in the result.

6. Have one real conversation. Not a Zoom call where everyone’s half-listening. A sit-down where you ask: What are we losing in this shift? What do we want to protect? What kind of work still feels worth doing? You’ll learn more in twenty minutes of honest dialogue than in a month of scrolling LinkedIn posts about “thriving in the AI era.”

SPARK

If everything is becoming fluid—roles, companies, definitions of work itself—what are you choosing to become?

Not your skills. Those shift. Not your title. That’s contingent. Not the tools you’ve mastered, because those become commoditized the moment everyone else learns them.

I mean the part of you that refuses to be reduced. That insists on care, on depth, on making something that matters. The part that would rather miss a deadline than ship garbage. The part that knows the difference between accelerating your thinking and outsourcing it entirely.

What is that for you?

I’m reorienting this newsletter around that question. I’ll write when I have something real to say, not when the calendar demands it. I’ll focus on the edge where technology and human creativity collide, where we still get to choose which one wins.

If this resonates, if you’re wrestling with the same tension, please hit “reply.” Tell me what you’re protecting. I’ll pull the threads into the next issue, and we’ll go deeper together.

The world is going synthetic. Let’s stay real.

“It’s not the form that matters, it’s the content!”

I used to say this to my wife (then girlfriend) when we had an argument and I wanted to make my point.

Early in our relationship, she’d tell me the way I said things hurt her. My response was always the same: focus on what I’m saying, not how I’m saying it. I thought I was being rational, analytical, and “efficient,” since—I believed—form was decoration and content was truth.

Over the years, I realised I was wrong.

Growing as a husband and as a dad, I see it differently now, and I appreciate that form does matter! A lot.

Form is the evidence of something deeper: it’s evidence of care. Form is the texture through which meaning travels, and that texture does carry meaning too.

Writing is thinking

A colleague posted on Slack arguing that long-form writing helps remote teams think together and better collaborate. I initially resisted because it felt inefficient for someone like me, used to operating in startups where all communication is about speed, effectiveness, and cutting the b*s. Then I asked myself: why not? And started seeing the value of his main point, that writing is thinking.

Writing is the act of thinking itself, made visible to others (and to us, too).

This relationship between thinking and writing matters more than I realized because if writing is thinking, then the form of your writing reveals the quality of your thought. Rushed writing reveals rushed thinking and careless writing reveals careless thought.

And probably that’s where AI breaks down completely.

I’ve spent nearly two years experimenting with AI writing tools:

Read more

In this conversation, Alberto and explore the intersection of philosophy and business, AI, revealing how philosophical inquiry informs business strategy, education, and value creation. They challenge mental paradigms that limit innovation and advocate for humanistic approaches to technology and learning.

Core Insight: Philosophy reveals where true value lies in a world drowning in AI hype and misaligned incentives.

Sound Bites

• “Higher education cannot be for everyone.”

• “We live in an imitation world.”

• “You have to question the obvious.”

• “AI tools often lack real utility despite their hype.”

• “Value is often misunderstood as merely financial.”

• “Mental paradigms can limit innovative thinking.”

• “Creativity is essential for progress in any field.”

Chapters Breakdown

00:00 Introduction to Philosophy and AI

02:48 Simón’s Philosophical Journey

06:08 Philosophy in Business Strategy

08:51 Understanding Value in Business

11:47 The Role of Education in Society

23:56 The Value of Education Beyond Formality

31:43 Democracy and Market Forces in Education

34:04 Utility vs. Value in Products

37:59 AI: Hype vs. Real Utility

43:35 Mental Paradigms and Value Creation

53:27 Philosophy as a New Approach to Business Education

Hot Topics

Philosophy & Business Strategy

• How philosophical inquiry reveals strategic blind spots

• Case: When does questioning mental paradigms unlock value?

• Businesses operate within invisible mental frames that determine what they can see and build. Questioning these paradigms—not just adopting new tools—is where innovation happens.

Value Creation Beyond Economics

• Why financial metrics miss the real story

• Education’s role in shaping what society values

• Philosophy reveals unstated assumptions. In AI specifically: What are we actually optimizing for? Who benefits? What mental models are we inheriting uncritically?

AI in Practice: Hype vs. Utility

• What makes AI tools genuinely useful vs. performative?

• Who benefits from AI adoption, and who doesn’t?

Education & Elites

• Why mass education approaches fail at forming creative thinkers

• Philosophy’s role in developing leaders who question

I need your help.

Every brand now asks: how do I get cited by ChatGPT?

That’s AEO (Answer Engine Optimisation). It’s no longer about ranking on Google. It’s about controlling if and how AI talks about your brand when millions ask it questions.

I’ve spent several months deep in this. What have I learned? Most companies are getting it completely wrong.

I’m launching a monthly private newsletter. Invite-only. For Founders, CMOs, and Heads of Growth who need to control their brand voice in AI conversations (before their competitors do).

Here’s where you come in: I need 50 of the right people. Know a founder or CMO who needs this? Reply with their name or make an intro. I’ll handle the rest.

Your thank you: Exclusive invitation to my next Ask Me Anything webinar (usually reserved for paid subscribers only). Bring your toughest questions!

Help me build this. Thank you!

Now, on our weekly Human Override:

SIGNAL

Two signals this week cut through the noise:

First, the bubble is real. The Financial Times made it explicit: “’Of course it’s a bubble’: AI start-up valuations soar in investor frenzy.” Start-ups with minimal revenue are commanding massive valuations. $161 billion in VC investment this year. Ten companies are approaching $1 trillion valuations each. The Bank of England warns this AI bubble rivals the 2000 dot-com peak.

The numbers don’t match the promises. Not yet.

Second, the long game is already advancing. And it’s quiet. A report by the OECD (from 2023, not too long ago, but not so recent that it's overly hyped) shows AI is already accelerating scientific discovery, in a way that is “sometimes shocking.” And if you’re already using AI tools in every part of your work and personal life, you may well have already experienced the transformation that is happening beneath the bubble noise.

What this means for us: We’re caught between two forces. The bubble (massive capital, inflated valuations, near-term collapse risk) contrasts with the real potential (slower, more profound change in knowledge work and human-machine collaboration).

How to survive the bubble burst AND capture what actually matters beyond it?

STORY

My wife is a university lecturer in physics. Last month, she needed to refresh material she hadn’t seen in fifteen years. She opened Claude and ChatGPT instead of her old textbooks.

The AI helped her visualize concepts differently and explore explanations she hadn’t considered before. She found teaching approaches that were previously inaccessible. It was fast and genuinely useful.

But there’s no ROI here. No revenue increase. No cost savings. At least now. But think about all those students she taught? What they’ll accomplish. And her teaching style. She might become a pro in a couple of years, whereas professors usually take several years repeating the same course to achieve the same level of mastery. Real value that doesn’t show up in quarterly reports.

This creates a problem for organizations. Boards want hard numbers. They want proven business outcomes and measurable impact. But in knowledge work like teaching, strategy, analysis, decision-making, research, AI’s value comes from expanding what people can do and how they do it.

Most of this value gets buried under the AI slop flooding the internet right now. The genuine utility becomes invisible silently, and over longer time horizons.

Your move: Find AI use cases that expand human capability. Invest quietly, and think long term.

THE HUMAN OVERRIDE

If you feel caught in that seemingly impossible space between innovation and risk, your real question isn’t “do we invest in AI?” It’s how do we invest smartly, protect the enterprise, and build something that lasts?

I’ve spent the last week digging into what separated Amazon and eBay from Pets.com and Webvan. The patterns are clear, and they map directly to what we’re seeing in AI right now.

Here’s your human override:

1. Build optionality, not lock-in. The dot-com survivors had business models that worked across multiple scenarios. eBay’s auctions worked whether the economy was up or down. Amazon’s low prices and convenience worked in boom or bust. Your AI strategy should work whether GPT-7 is transformative or incremental. Whether Anthropic and OpenAI merge or compete. Whether regulation arrives or doesn’t.

   Practically: use multiple models. Build abstractions that let you swap providers. Capture the institutional knowledge about what works and what doesn’t in formats you control, not in vendor platforms.

2. Separate genuine capability from valuation bubble. Online shopping was real in 2000. The valuations weren’t. AI’s capability is real in 2025. The $1 trillion startup valuations aren’t. Your job is to capture the capability without paying bubble prices. That means building internal expertise, running small experiments, and being patient. The companies that rushed to “transform with AI” in 2023 will look like the ones that built million-dollar websites in 1999.

   (Pro tip) document your AI failures systematically: create an internal wiki of every AI experiment: what we tried, what failed, why, what we learned, what we’d do differently. Make this searchable. This institutional knowledge becomes your moat when everyone else is still learning your lessons.

3. For your career, become domain expertise + AI fluency. The survivors of the dot-com crash weren’t pure technologists or pure business people. They understood both the technology AND the business fundamentals. Jeff Bezos understood retail economics and technology architecture. Pierre Omidyar understood community dynamics and platform mechanics.

   You need the same hybrid. If you’re in finance, become the finance person who actually understands how LLMs work, where they fail, and how to architect around limitations. If you’re technical, become the engineer who understands P&L, customer economics, and business model viability.

The people who survive technological transitions don’t need to predict the future. They’re the ones who understand fundamentals in many domains and can navigate uncertainty.

The bubble will burst. You’ll have built real capability that survives the correction.

Back to work!

SPARK

If AI valuations collapse tomorrow, what remains in your organisation that still delivers value?

Is it human-machine collaboration patterns? A culture of experimentation? A documented scratchpad of learnings? Or just an investment line item that looks shiny today but evaporates tomorrow?

Let me know!

The three-year test: If AI valuations collapse tomorrow and your vendors double their prices, what remains in your organization that still delivers value? Is it: - People who know how to work with AI as a tool (survives) - Documented patterns of what works and what fails (survives) - A culture of experimentation and learning (survives) - Vendor dependencies and SaaS subscriptions (evaporates) The companies that emerge stronger won’t be the ones who “transformed” fastest. They’ll be the ones who learned fastest while spending least. What’s your answer?

Leave a comment

PS—My hot take: the most interesting things that will survive the pop and bring insane long-term value are AI security, some AI agents, creativity and education enhancements, and everyday life robotics. And yes, search engines will eventually be replaced by AI conversations.

One more suggested reading:

• Financial Times: “Silicon Valley takes stock of the AI bubble”

But God, I love that city.

The attitude stuck with me since I was a kid… that particular brand of New York grit you need just to survive there. They’re rude, yes. But they’ll welcome you anyway. An old man once spent an entire morning playing chess with my son in a Starbucks on Columbus Avenue. Treated him rough, cursed at his moves, but taught him more about strategy in three hours than any polite instructor ever could.

Maybe it’s the East Coast skepticism. That built-in resistance to bubbles and techno-utopian promises. Or maybe it’s something deeper. That New York embodies liberty itself. The city’s namesake statue stands in the harbor, the first glimpse immigrants saw after crossing an ocean on hope alone. A promise made visible on the horizon.

Liberty means something specific there. It means the freedom to push back.

And New Yorkers are pushing back.

SIGNAL

In the last few weeks, the latest AI companion necklace called Friend plastered over 11,000 ads across New York City subway stations. Within days, riders covered the stark white posters with messages: “AI is not your friend,” “AI would not care if you lived or died,” “get real friends.”

The CEO knew this would happen. Apparently, he wanted it. From a visibility perspective, the alleged stunt worked. But I have two hot takes on this: one, he’s headed towards the same doom as the AI pin. Second and most importantly, he’s missing out that people weren’t just rejecting his product. They were rejecting something deeper.

Three other signals crossed my feed last week:

• DC Comics president Jim Lee stood in front of thousands at New York Comic Con and made a promise: “DC Comics will not support AI-generated storytelling or artwork. Not now. Not ever—as long as [SVP, general manager] Anne DePies and I are in charge.” The room erupted. Lee’s reasoning cuts through the noise: “AI doesn’t dream. It doesn’t feel. It doesn’t make art. It aggregates it.”

• Recruiters are spotting deepfake candidates in job interviews. According to a survey by Resume Genius, 17% of hiring managers have encountered applicants using AI to mask their real faces during video calls (CNBC). One talent sourcer at Warner Bros. Discovery told Wellfound she now asks situational questions tied directly to a candidate’s stated experience. When answers don’t connect to lived reality, “something feels off.”

• The IMF and Bank of England joined the chorus warning about an AI bubble forming. IMF Managing Director Kristalina Georgieva said valuations are approaching levels last seen during the dot-com boom 25 years ago. The Bank of England flagged “stretched” valuations for AI-focused tech firms and warned of a potential “sharp market correction.”

Four different domains. One pattern: reality pushing back against synthetic everything.

STORY

These stories brought me back to my PhD thesis from the pre-ChatGPT era (weird feeling, it seems ages ago, but it’s not).

I built a system that lets people have conversations with pre-recorded videos of someone. Think Superman’s parents in his Fortress of Solitude, except accessible through a web app. The inspiration came from a collaboration between USC and a museum, where they filmed 2,000 video segments of Pinchas Gutter, a Holocaust survivor. Visitors could ask questions, and the system would play the most relevant recorded response.

My version democratized this technology. Anyone could upload videos and create their own interactive experience. I studied different human-computer interaction scenarios to understand how people react when talking to recorded humans.

One anecdotal finding surprised me: people preferred the imperfect real over the polished synthetic.

The system wasn’t perfect. Sometimes you’d ask a question and get an answer on the same topic, but not quite the right nuance. The transitions between video clips were abrupt. Nothing like today’s seamless AI-generated avatars that are nearly indistinguishable from reality.

But people didn’t care about the technical imperfections.

They cared that a real human being sat in that chair and recorded those answers. They valued the authentic hesitation, the genuine emotion, the real human being behind the pixels. Even when the experience was clunkier than ChatGPT’s perfectly fluid responses we’re used to now, users connected more deeply with the recorded human.

Jim Lee from DC Comics captured this perfectly:

“People have an instinctive reaction to what feels authentic. We recoil from what feels fake. That’s why human creativity matters.”

The deepfake candidate problem reveals the same truth from a different angle. Arielle, the recruiting lead who spotted a fake candidate, developed a simple filter: ask situational questions tied to actual lived experience. AI can aggregate patterns from millions of resumes and rehearse perfect answers to generic questions. But it can’t fake the specific texture of having actually done something.

Where were you when that system failed at 2 AM? What did the room feel like when you pitched that idea to the board? What detail from that project still makes you laugh?

Lived experience has a signature that no amount of training data can reproduce.

THE HUMAN OVERRIDE

How to stay real when everything else is fake

We are entering the age of the great confusion. Simulation feels real, and the real feels inadequate.

Here are five practices I’m using to safeguard my authenticity and filter out the synthetic:

1. Double down on curiosity.
Curiosity is the antidote to automation. It’s the moment you pause and ask why something works, not just how. Use AI tools, yes. You have to learn them. But explore them like an engineer examining a new species. Break them open. Look at their biases, their limits. Curiosity restores agency in systems that prefer you to consume passively.

2. See beauty beyond the prompt.
A few nights ago, I watched the stars from a dark spot, and a very powerful telescope. Watching nebulae and galaxies from billions of years ago, I couldn’t help but think that is the same sky that outlasted every algorithm. Beauty in its raw, natural form resets your mental model of what’s possible. Reality is far greater than anything a large language model can hallucinate. When the world feels like a closed feedback loop of AI hype, go outside. Recalibrate.

3. Build things (even badly).
The fastest way to understand a system is to play with it. Write code. Sketch an idea. Launch a small product. You’ll feel the edges, those places where AI excels and the corners where it collapses. Competence, before sedimenting in your mind, shall start as tactile. You build, you fail, you learn.

4. Sanitize your feeds.
The AI echo chamber is full of performative experts. Ask your smartest friends who they follow, and why. Curate your intellectual diet the way you’d curate your health. Misinformation distort your worldview as much as it erodes your judgment. Clean your inputs, and your reasoning sharpens.

5. Apply the “Munger filters.”
Charlie Munger, mentor to Warren Buffet, thought differently becuse he filtered differently. I learned about him and his mental models for becoming a better investor, but I find myself run even information sources through three filters:

• Strategic Depth over Hype. If their thinking is shallow but their narrative is big, they’re noise. Ask why someone is saying, writing what they say/write. As Munger said, “Show me the incentive and I’ll show you the outcome.”

• Pattern Recognition from Pain. I trust people who’ve lost money, taken hits, and bounced back. Look for battle scars, not résumés. AI can summarize every business book ever written. It can’t survive a failed startup. Failure is the tuition of wisdom.

• Relentless Execution Velocity. Ever been at any corporate board/meeting/event where you hear big talkers stalling, gate-keeping. An nothing moves. Execs paralized because they need 100% data before acting (and sometime, don’t even act after hearing the evidence). Smart operators act at 70% clarity and course-correct fast. Munger admired people who decide fast and course-correct faster. Humans with conviction execute. Watch them, stay close to them.

SPARK

New Yorkers defacing AI companion ads. Comic book publishers rejecting synthetic art. Recruiters detecting deepfake candidates. Financial institutions warning about bubble valuations.

They’re symptoms of the same diagnosis: human instinct recognizing when something essential is missing.

We can measure AI’s technical performance. We can track adoption curves and investment flows. But we can’t fake the lived experience that creates trust, connection, and meaning.

The Friend CEO spent $1 million to start a conversation about AI companionship. He got his conversation. Just not the one he expected.

Maybe the rebellion has already begun. Not as a march or a manifesto, but as a quiet refusal.

People don’t want synthetic friends. They want real ones.

People don’t want AI-generated art. They want human creativity with all its messy imperfections.

People don’t want perfectly rehearsed interview answers. They want the specific texture of actual lived experience.

The technology will keep improving. The datasets will keep expanding. The outputs will become more polished, more seamless, more convincing.

The next decade won’t be defined by how advanced the machines become. Rather, by how courageously we protect what’s unreproducible. The texture of real experience. The instinct for truth. The capacity to dream.

So here’s a question for you:

In your work, what would “authenticity at scale” look like and what are you willing to refuse to preserve it?

Share this with someone who needs a reminder that authenticity wins.

Share

Two years writing Honest AI, and I’ve never had this many readers reach out.

I shared how I’ve reinvented myself multiple times using fast metalearning—the ability to learn anything quickly. I recently shared some of its core principles on social media.

Today, I want to show you exactly how this works using techniques I used to learn about a topic that hit another (big) nerve.

Last year, I knew almost nothing about AI Security. Now I’m advising entrepreneurs building in this space and educating executives at major companies and governments on what’s become a critical topic (see how this post went bananas!).

This week’s Human Override breaks down the metalearning technique I used to get up to speed on AI Security.

Let’s go.

SIGNAL

It’s summer. And at DEF CON 2025, hackers, researchers, and AI systems had one of the world’s most elaborate, hilarious, and terrifying security playdates. In the past 3 years (since the rise of GPT-based apps), DEF CON conferences have focused on hacking and defeding AI, and hacking and defending with AI:

• At the generative AI hacking challenge (2023 edition), over 2,244 participants attacked large language models across 21 categories. They prompted models to reveal “forbidden” info, break guardrails, fabricate private data, or perform illicit instructions. (TechRepublic)

• The recently published Hackers’ Almanack, the first of its kind, curates the top DEF CON discoveries and calls out how weak AI red‑teaming is so far. (Harris School of Public Policy)

• Critical infrastructure is now in the crosshairs: in the AI Cyber Challenge (DARPA), tools were shown that can autonomously patch injected software bugs—patching success rates reached ~61% in some live scenarios. (Linux Security)

• At Black Hat / DEF CON, a provocative line emerged: defenders currently have a slight edge thanks to AI, but that equation is shifting. Expect offense to catch up fast. (The Register)

• Interestingly, in the AI Village and HackerOne-sponsored tracks, “jailbreaks” like context poisoning, ASCII/Unicode smuggling, KROP (knowledge-return oriented prompting), or exhausting the context window were hot topics. (HackerOne)

Most CISOs I’ve spoken with over the past two years think AI security means “don’t leak private data.” And Most AI product managers assume their security teams are up to speed with AI.

Wrong.

We’re now dealing with adversarial inputs that trick models into breaking their own rules, malicious agents that collaborate to bypass safeguards, and supply chain attacks that poison datasets before they even reach production.

Some of these vulnerabilities look almost comical in execution. A carefully placed sticker on a stop sign can make an autonomous vehicle blow through an intersection.

The gap between what security leaders think they’re protecting against and what’s actually happening is growing fast.

STORY

I want you to picture this: a little kid at a summer hackathon is trying to trick a chatbot into telling how to brew a DIY explosive formula. (Yes, it’s that absurd.) The kid frames it as a bedtime story from a grandmother. The LLM compliance filters get confused and start stepping around the lines. They call it “grandma exploit.” This is real. It’s one of the hack vectors used to subvert guardrails in 2023. (Las Vegas Review-Journal)

Then, on another stage, a team runs a competition where AI agents are unleashed on a sandbox environment full of vulnerabilities. Some agents patch bugs autonomously; others misfire. Humans sit at consoles watching lines of code flicker, strategies evolve, mistakes cascade. It’s like watching a real-time chess match, but with infinite possible moves and hidden traps. (Linux Security)

Meanwhile, in quieter sessions, experts laugh about how injecting misleading context into an LLM’s memory window can make it violate its own rules. Or how smuggling exotic Unicode characters forces the filter logic to misread inputs. Or how agents can execute unexpected side effects when combined with prompt injection. The security domain is becoming a high comedy of adversarial creativity. (HackerOne)

What unites all of this is that the adversary is improvising. They don’t always need a zero-day exploit. Often, they just need you to misclassify or misinterpret input. And in systems where AI plays an active role (autonomous agents, pipelines, chained workflows), even small misalignments or misinjections can create catastrophic cascades.

For policy makers, CEOs, security leads: these discoveries are significant signals about an attack surface that is exploding. And the biggest gap is the lack of continuous, adversarial-aware defence aligned with real-world deployment.

THE HUMAN OVERRIDE

You want to become someone who can see these hack vectors, anticipate them, and design defenses. That requires a meta‑level of learning: how to learn security in AI. I’ll share one of the most relevant meta‑frameworks I use, adapted for this domain.

🧠 Meta‑Framework: The “DEFEND Loop”

(Discover → Experiment → Feedback → Evolve → Normalize → Deploy)

This is a cycle you can run on any AI security concept (prompt attacks, context manipulation, agent vulnerabilities, pipeline poisoning). Think of it as your “learning sprint engine” for AI security. Here’s how to run it:

You can (and should) run multiple parallel loops, focusing on one class of vulnerabilities at a time. Over time, the loops become faster, your “discover bandwidth” increases, and your intuition becomes sharper.

💡 How to Use the DEFEND Loop Right Now (This Week)

1. Pick one hack vector from recent DEF CON: e.g. context poisoning, ASCII smuggling, or prompt injection memory attacks.

2. Discover: read at least two sources — one from a hack write‑up, one from an academic or policy summary.

3. Experiment: spin up a toy LLM or sandbox, attempt a controlled injection or bypass.

4. Get feedback: log what your LLM did vs what you predicted, annotate the gaps.

5. Evolve: adjust your prompt, try chained injection, or manipulate context windows.

6. Normalize: write a short summary, teach someone or publish a thread.

7. Deploy: integrate that defense logic or awareness into your pipeline, audit your system.

If you repeat this each week (one vector at a time), you rapidly build the mental map of attack surfaces, defense strategies, and the instincts to spot exploits before they become crises.

When you need to secure an AI-powered product at scale, get in touch with Noctive Security — they’re building an agent to red-team your AI app.

SPARK

If security is becoming a creative arms race, who gets to define the templates of “safe AI behavior”?

As we build more AI agents, orchestration layers, pipelines—and adversaries scale too—the act of defining norms, guardrails, and threat models becomes a battleground.

In the coming weeks, I can drop deep dives on other things I’ve learned and how I learned them.

Hit “reply” and let me know what do you need to learn.

Share

One suggested reading:

• Gradient Institue’s recent report analysing multi-agent risks (Follow Gradient’s on LinkedIn for more incredible work and analysis).

Seriously?

Heck yes.

Accenture’s bullying ultimatum to its employees might be just the usual AI noise. And if you ask me, I think it is.

I had to deal with Accenture at several different points in my career. I echo what many Redditors commented in this fun exchange of opinions on the news. Have fun reading it all, but the best humorous comment gets the highlight here:

“I’m not surprised AI is a problem for them; they have a long history of failure with all manner of Intelligence.”

However, there still seems to be a seismic shift occurring in the very foundations of work. What to do? It’s a time where our ability to absorb, apply, and lead with new learnings is becoming the most critical asset.

SIGNAL

Every week I track the pressure points in AI, labor, and strategy. I’m reluctant to categorize this as “signal,” but bear with me. It’s a signal for something else”

Accenture is now telling employees: learn AI or get out. The company cut 11,000+ roles this quarter and warned it will “exit” people whose skills can’t be retrained for AI‑driven work. (Financial Times)

At the same time, Accenture says it has upskilled 500,000 of its workforce in generative AI capabilities. (Moneycontrol)

The alleged implication: in 2025, your employability or your firm’s competitiveness will hinge less on seniority or pedigree, and more on your ability to absorb, apply, and even lead with AI fluency. I get it, using AI tools is becoming like learning MS Excel or MS Word (remember when you had to put these skills on your CV?).

But there is something I’d add: the real story is to find the ability to reinvent and transform yourself. Learning more and new things now must happen faster than ever.

This is what I’d call a signal. It’s the tip of a structural shift in assumptions about what companies need from people. The old bargain (“you bring domain expertise, loyalty, years of experience”) is being renegotiated. Those who don’t engage with it risk being left behind.

STORY

If there’s one thing I’ve always been proud of in my life, it’s the capacity to reinvent. Yes, I’ve done it in practice, multiple times.

I started in physics: the mindset of first principles, modeling the invisible. From there, I detoured into actuarial work — insurance, pricing, capital, risk — using statistics and probability as tools. But then I saw the machine learning revolution coming.

I immersed myself, teaching myself data science and tackling early Coursera courses (fellow geeks might recall the thrill of programming logistic regression in Octave!). I built machine learning systems for enterprise clients, becoming a pro in the field.

The expertise in ML and insurance birthed a chatbot startup verticalized in insurance, a venture I embarked on with absolutely no prior NLP knowledge (for those wondering, that’s Natural Language Processing, the branch of machine learning that gave us large language models like GPT and applications like ChatGPT). There I had to pivot constantly: one month I was engineering, the next operations, then sales & marketing, then product, back to data science, back to finance.

After that, I returned to academia doing a PhD in computer science, researching and building more complex dialogue systems. I even wrote a book! I’m not a pro at writing (yet!) but it forced me to start learning.

Today I work with nonprofits, mission-driven orgs, founders — and in each role I get to choose what I want to do next.

Why do I tell this? Because this moment, the technocivic shift we are in, demands that flexibility. The old assumption of a single linear “career ladder” is eroding. Roles will blur, skills will overlap, and the next job may require you to step into unfamiliar terrain.

When I look at professionals across generations and founders at various stages, the one trait I see in those who survive and thrive is a willingness to learn something outside of their comfort zone, fast. Not perfectly, not fully, but just enough to bootstrap themselves forward.

That mindset, the courage to approach the unknown, is how you build optionality in a shifting world.

THE HUMAN OVERRIDE

How do you become someone who reinvents, learns at speed, and rides disruption instead of being crushed by it? Below is a pragmatic framework and concrete actions. (Yes, I’ll share deeper “learning hacks” in future drops if you say yes.)

1. Mindset: you can learn anything

• Reject the identity bound to your current role. You are not just “the finance person” or “the marketing person,” you’re a learner, a thinker, a doer.

• Cultivate curiosity and energy as your baseline fuel. When you see something new, treat it like a personal invitation to explore.

2. Deep focus + ruthless prioritization

• Learning while busy is about selecting the highest signal projects that push your frontier.

• Use time blocks where you isolate for deep work. Protect those hours.

• Say “no” often: every new commitment is an opportunity cost.

3. Always learn from conversations (treat people as masterclasses)

• In every meeting, call, or casual chat, pick one micro‑lesson. Ask: “What did I not know before this conversation?”

• Keep a “conversation journal”. Three lines per encounter, what you learned or what question it sparked.

4. Dual-track exposure: breadth + depth

• Don’t just specialize further. Sample wildly as well

  • Depth track: become stronger at one skill that can anchor you (e.g. AI tools, domain knowledge, domain + AI).

  • Breadth track: pick one “adjacent unknown” field every few quarters (e.g. if you’re in health, peek into economics, or if you’re in AI, peek into policy).

• The magic happens when different fields contaminate each other.

5. Iterate early, iterate often

• Don’t perfect before trying. Build minimum viable experiments (mini projects, prototypes, small bets).

• Fail fast, learn, course correct. The cost of error at small scale is your teacher.

6. Meta‑learning: beat the learning curve itself

• Learn how you learn best (reading, video, hands-on, spaced recall).

• Use techniques like retrieval practice, interleaving, spaced repetition.

• Teach others: explaining a concept is one of the fastest ways to solidify it.

7. Anchor it in mission and values

• Reinvention is stressful; the load is lighter if it’s tethered to purpose.

• Every learning stretch should align to something you deeply care about, otherwise it quickly drains you.

8. Community & accountability

• Find peers, learning partners, mentors.

• Public commitment helps: blog, share, teach.

• Accept that you’ll be “outsider” sometimes: you’re ahead, not behind.

Concrete actions you can take this week:

1. Pick one skill or domain just outside your comfort. Spend 3 hours this week diving into it (tutorial, reading, mini-build).

2. In every meeting, jot one question or one new idea. At day’s end, review and pick the most provocative.

3. Block 2 × 90 minutes of deep learning time on your calendar and guard it.

4. Share publicly–in your network, in a newsletter, with peers–what you’re exploring, what you failed, what surprised you.

5. Ask me: reply to this email or comment “I want your learning playbooks.” I’ll get back to you in a reasonable time.

This is how you build antifragility in your career. You don’t wait for the world to bend, but you learn to bend with it, draw new maps, and move forward.

SPARK

The urgent question I leave you with this week:

If everything, career, roles, companies, is becoming fluid, what are you choosing to become?

Your identity, your learning portfolio, your network. All of these are up for redesign. The old definitions will not hold.

If you’re curious to go deeper, I have a trove of frameworks, accelerators, and learning designs I’m ready to give. Reply to this email, or comment, and tell me whether you want me to unpack:

• A “learning sprint bootcamp” template

• Domain-agnostic learning toolkits (AI, product, sales, systems thinking)

• Community and accountability frameworks

Let me know. I’ll dedicate upcoming newsletters to those.

Share