Irregular: What's your pigeon?
Killer robots, nuclear armageddon, and one very good bird
Hello from Paris,
In today’s Irregular, I would like to share with you the text of the keynote address I gave at the Digital Trust Summit in Brussels earlier this week.
The conference, organised by Global Data Innovation, brought together an unusual mix of technologists from Silicon Valley, European members of parliament and senior policymakers, Hollywood producers, directors and actors, and chief executives and board members from both sides of the Atlantic.
Taking any stage after Cate Blanchett is, presumably, always nerve-wracking. (She was launching a new online tool for artists to register the degree of consent they give for their face, voice, movements and ideas to be transformed by AI.) So I decided to play it safe, avoid giving my best Bob Dylan impersonation, and stick to the theme of geopolitics and artificial intelligence, which I had been asked to speak to.
My remarks are below. I hope you enjoy.
Best wishes,
Damien
Introduction
Now for a change of pace. I will cover killer robots, nuclear armageddon, Vladimir Putin, pigeons, and how to manage geopolitical risk in an age of artificial intelligence. Sound ok?
Since it is always prudent to start at the beginning, I would like to start with my own first interaction with artificial intelligence.
Fifteen years ago, I was a lazy law student trying to find an honours thesis topic that would allow me to avoid reading turgid legislation and meandering jurisprudence. I stumbled on artificial intelligence. And, in particular, the prospect that seemed more and more real, the more and more I read, that, one day, AI would be embedded in military systems and make autonomous and lethal decisions in real life.
I created a scenario that, at the time, felt like science fiction: a robot fielded on the battlefield that killed civilians in a way that, if done by a human, would clearly constitute a war crime or a crime against humanity. I asked a simple legal question: who, if anyone, would be held responsible? The developers? The commanders? The generals? The political leaders? The state? Or even the robot itself?
At the time, DARPA – the US defence ministry’s advanced research arm – was not only throwing money towards developing autonomous weapons but also commissioning computer scientists and lawyers to try to embed the Geneva Conventions into the code governing robots’ actions so that any decisions made by an autonomous weapon would, ipso facto, be compliant with the laws of war.
My conclusion, ultimately, was that this simply could not be done. The Geneva Conventions, like many laws, inevitably demand value judgments about trade-offs to be made. All relevant legal frameworks – from domestic criminal law to torts to those of state responsibility – were ill-suited to the situation. And it was therefore incumbent on lawmakers to fill in the gaps, lest the situation arise and no-one be held responsible. That seemed wrong.
Fast-forward fifteen years and we’re just about there already. The front-lines in the Ukraine war involve largely drone-to-drone combat. The schoolchildren killed on the first day of the Iran war were targeted by systems infused by artificial intelligence, even if, it seems, man remained “in the loop”. Diplomats in Geneva have taken up the precise topic I wrote about but not reached any agreement after years of discussion. And artificial intelligence has become not only much more advanced but significantly more widely distributed across so many systems we use today.
Indeed, the presentation we have just seen – on the extraordinary pace and power of AI “agents” in hacking into corporate systems and how Anthropic’s latest model was taken off the market after, within hours of release, it had breached all the NSA’s confidential systems – speaks to the imaginary situation of one lazy law student suddenly seeming much less hypothetical.
Enjoying Geopolitical Dispatch?
The man who saved the world
I have been asked to speak to you today about AI and geopolitics.
There are so many intersections between these two topics, which are arguably the two greatest macro trends of our age. There is no way, in fifteen minutes, to cover the field. So instead, I will restrict myself to what may appear to be a few contrarian theses: that the limiting factor in this new era is not the technology; it’s us. That we have more to fear from human stupidity than artificial intelligence. And that the winners will be the ones who understand that first.
Often, when we think about dystopian futures, we immediately reach for 1984, George Orwell’s fictional version. But I would like to start with September 1983 and an altogether real situation.
Back then, Stanislav Petrov was the duty officer at a Soviet early-warning facility, doing the night shift. The system lights up, showing five American intercontinental ballistic missiles inbound for Russia. Protocol is clear. Petrov should have reported the information up the chain, retaliation would follow, and the Soviet Air Defence Forces should launch a barrage of nuclear-armed missiles at the United States. The world should have ended before breakfast.
But Petrov looked at the screen and did something machines cannot do. He doubted. Five missiles seemed wrong — in his judgment, if the Americans were going to launch a “first strike” against the newly labelled “Evil Empire,” they wouldn’t send just five missiles, as that would almost certainly be an act of suicide; they would fire hundreds so as to prevent any possible second strike. Petrov judged the blips on his screen a malfunction. And so he did nothing.
The reason we are all here is that Petrov was right. Sunlight glinting off clouds had fooled the satellites. One man’s judgment quite literally saved the world. And he was able to do so not only because he was in the loop of lethal decision-making, but because he used critical reasoning to question the data presented to him and, ultimately, because he was willing to defy what was meant to be essentially an automated decision.
As the world embarks on fielding more and more “AI agents” – which to me, incidentally, always conjure up Hugo Weaving’s Agent Smith in The Matrix and his disgust at “the smell” of humans, which he considers a virus – it’s worth thinking about the potential consequences for embedding, trusting and handing over control to artificial intelligence at scale, especially in the context of nuclear systems.
Putin’s thesis
My second interaction with artificial intelligence came in 2017, while I was working as a diplomat in Moscow.
I got quite excited when I heard that Vladimir Putin had told a hall of Russian schoolchildren – perhaps an inappropriate audience for the message – that “whoever controls artificial intelligence will control the world”. At last, I could dust off the old thesis.
Putin’s statement was mostly a one-liner. But his strategic sentiment has caught on. And he may be right.
There is now a commonly held view among policymakers that the nation that adopts AI fastest may have the stronger economy, the better weapons, the sharper intelligence, the tighter grip on its own population. That logic is driving the largest technology race since the Manhattan Project, with almost all world leaders having bought into the notion and now viewing not just advanced frontier models but also the mineral and energy industrial “stack” beneath them as “strategic”. Governments are accordingly pouring money into ensuring that it is they, and not their adversaries, who gain the advantage.
But there are two other possibilities Putin didn’t mention.
First, there is a good chance that, ultimately, nobody will control artificial intelligence. Not in the science-fiction sense, although that, of course, is worth bearing in mind. I mean, in the more mundane sense, that general-purpose technologies have never remained under control. They leak, diffuse, commoditise. The cat, once out of the bag, cannot be put back in. After all, what country “controls” the railway or the light bulb or electricity or any other general-purpose technology?
A second possibility is that even if someone controls artificial intelligence, there’s every chance it won’t be governments. For the first time since the atomic age, the frontier of strategically decisive technology is almost entirely within private companies, not states. Nation-states have, so far, been able to keep nuclear technology in the hands of a handful of countries owing to the complex resourcing – mineral, scientific, and industrial – required to build them. The barrier to entry, even eighty years on, is very high.
By contrast, the barrier to entry for artificial intelligence - not for frontier models but for their applications - is extraordinarily low. And whereas societies have agreed and worked hard to avoid “proliferation” of nuclear weapons, we are extremely gung-ho in our present attempt to “deploy and diffuse” artificial intelligence as quickly and as widely as possible. If I can make an app with Claude, I can assure you that anyone can, including “bad guys”.
Homing run
Another vignette.
Now we’re going further back in time. It’s the 1850s. The world is electrifying. Telegraph lines (and railways) are the AI of their day — racing across Europe, promising instant information to whoever wires up first. But there’s a gap in the transmission line between Brussels and Aachen, in Germany. A missing link of 76 miles where the lines had not yet been connected.
A young media entrepreneur named Paul Julius Reuter looks at this gap, and while everyone else waits for the cable, he buys a fleet of 45 homing pigeons. Reuter’s pigeons, carrying stock prices, crossed that gap six hours faster than the mail train. Reuter beat electricity — with birds. The old, the natural, the unfashionable “technology”, deployed with creativity, gave him the edge. And he turned that edge into the news empire that still bears his name. By the time the telegraph lines had been connected, Reuter had established his firm not only as fast but also as trustworthy.
Note what Reuter’s advantage actually was. It wasn’t the telegraph — everyone was getting the telegraph. His advantage was in seeing what the shiny new technology couldn’t yet do and bridging the gap by other means. Where everyone zigged toward the wires, Reuter zagged toward birds. And once everyone had access to the most advanced technology of the day, Reuters had developed its “moat” – it wasn’t tech; it was trust, sticky clients, and distribution.
Since so many of you in the audience are business leaders, I would encourage you to ask: while everyone races to adopt AI, what is your homing pigeon?
Know someone who’d enjoy this?
Racy thoughts
Now, the geoeconomics. Earlier, I mentioned how states are racing to achieve supremacy in artificial intelligence. Many have, quite rightly, likened this to other “arms races” and “technology races” throughout history. That is not wrong. But there is often a hidden assumption that the prize will go to whoever spends the most, whether governments or corporates. I would just offer one caution about that way of thinking, based on the history of geopolitics.
This time, back to the USSR. We all know the Soviets poured their economy into the space race. They had genuine wins — Sputnik was the first satellite in space, Gagarin was the first man in space, Laika was the first animal (and a very cute one at that) to orbit the Earth. And they often achieved these wins for less money than the Americans spent. But the competition itself, the reflex of matching investment every time, everywhere, on everything, helped break them. You may have heard the (possibly apocryphal) story of American private firms spending millions developing a space pen while the Soviets simply used a pencil. The lessons? Not everything requires newfangled technology; you can win many technological battles but still lose the strategic war; and it’s easy to go bust trying to outspend your competitor.
And think about general-purpose technologies and their business applications throughout history. Did the firms that adopted personal computers earliest permanently outcompete their rivals — or did the advantage even out once everyone had one on their desk? Did the inventors of the lightbulb or the steam engine become the dynasties of their age — or were the benefits diffused across whole economies? Did the countries at the frontier of those technologies prosper most, or the countries that incorporated them widely, broadly, and deeply into ordinary life? Did Reuters or Bloomberg win by having the best technology or by having the most customers?
It seems to me that general-purpose technologies are ultimately levellers. The durable advantage rarely goes to the inventor but rather to the “deep adopter”. While a Brit invented the steam train, it was the United States that rolled out rail at a scale that allowed it to become the world’s most powerful nation. And so it is worth “sitting with” – as Claude might advise – the question of whether the United States and Europe, as they “race” for supremacy in frontier technology, will do better than China, whose December 2025 plan aims for 90% economy-wide adoption within the next five years.
It’s also probably worth thinking about how the experience of financial market exuberance over railways and electrification ultimately panned out for the leading firms of the day. Debating whether we are in the middle of a bubble about to burst I’ll leave for another day.
What’s in a name
Part of our confusion around artificial intelligence comes down to its name.
At the founding workshop at Dartmouth in the 1950s, the “inventors” of AI debated what to call this new field. They had two alternatives: one was the rather banal “complex information processing”. The other, which they obviously chose, was “artificial intelligence” — sexier, more fundable, more human. And we’ve been anthropomorphising it ever since. So much so that one of today’s leading firms is called “Anthropic,” which is very deliberately programming its models to “think” and “act” in a human way.
But strip away the branding, and you see what artificial intelligence actually does: it processes complex information. Point it at complex information that genuinely needs processing, and it’s remarkable. Point it at simple information, or at complexity that doesn’t need processing — and it’s a hammer in search of nails. That’s true whether the complex information processing is directed by humans, or by “agents” without blood and bone, or – as seems to be the next frontier – by “loops” of agents interacting with one another.
How does this relate to geopolitics? Well, because geopolitics is not, mostly, a complex information processing problem.
The questions that will move markets and borders this decade are not “Big Data” questions. They are what I call “Small Data” questions. What is in the head of Donald Trump? Of Vladimir Putin? Of Xi Jinping? What was agreed in a room, in a glance, between the lines of an official statement? What do the small number of “great powers” agree on, or disagree on? I would submit that those “data points” – the vital few inaccessible to artificial intelligence – matter more for understanding the trajectory of international relations than a synthesis of everything written on the World Wide Web.
More information, in short, is not better information. Back in Reuter’s day, more, better and quicker information about world affairs was extremely valuable. But today, most of the time, less is more. The task for corporate leaders like you is not to read longer reports, to consume more analysis, or engage in chats that become increasingly sycophantic the more the model knows you. Instead, the task is to improve one’s own judgment of what on earth is going on.
I use that phrase – what on earth is going on – very deliberately. Consider what today’s models actually consume: text, the written record of humanity, the content of the internet. While this is an extraordinarily vast store of information, albeit of inconsistent quality, the written record is nevertheless a fraction of reality – a thin slice of what reality consists of and what our five human senses perceive, and often the least honest slice. So while artificial intelligence may be able to read the internet, it cannot read a mind, read a room, or read a statement delivered with a wink or a nudge.
That means, if you’re not careful, relying on artificial intelligence for your geopolitical analysis can become like running faster and faster in the wrong direction — ever more confident, and ever further from what actually matters.
So-called “world models” — AI trained on the physical world rather than its written shadow — may change this. Who knows. But we’re not there yet.
People power
One final theme I would like to touch on is that technology, while undoubtedly an important factor in shaping society and moving history, is far from the only one.
AI may well have major societal, political and geopolitical implications. It holds potential to boost productivity, change the nature of work, and disrupt the social contract. And it may ultimately be the case that whichever countries diffuse the productivity benefits of artificial intelligence broadly across their economies will have greater economic success than those that do not.
But when you look at the rise and fall of nations throughout history, people seem to matter a great deal more than technology. Even Elon Musk — not a man given to underselling technology — says the “limiting factor is people”.
However AI pans out, when considering the future of geopolitics and which parts of the world will do well, it’s very useful to look at where the demographic booms and busts will be.
According to the United Nations, by 2050, India will have another 250 million people, on top of its already world’s largest population. Nigeria will add another 200 million. Countries like the DRC, Angola and Pakistan will see major population growth. Half of the world’s population will be born in just eight countries, with five of those in Sub-Saharan Africa.
Meanwhile, China is projected to have 150-200 million fewer people than today. Russia and Japan will see declines of 15-20 million. Europeans also aren’t mating enough. And extraordinary declines in fertility rates, caused, according to some researchers, by the omnipresence of microplastics and other toxins in modern life, may mean the UN’s predictions will one day be judged optimistic.
I mention demographics because geopolitics is driven by national power. This is an amorphous concept. But, like the American Supreme Court judge who refused to define “pornography” except to say “I know it when I see it”, a country’s power is plainly a composite of its people and its technology. People are the denominator, technology is the multiplier.
But a multiplier on a shrinking denominator gives you a smaller number. AI can help you get more out of people — but it is people who use it, people who build with it, and people who buy goods and services. Robots are not and, I would hazard, never will be, customers.
Similarly, there is an important material dimension of the digital revolution. While artificial intelligence can feel like it exists only in the ether, it is built very much in the real world: data centres devouring energy and water, infrastructure devouring capital, and human beings to build it, run it, and check (rather than write) the code. The countries that thrive won’t simply be those that invent the most advanced technologies of the future. They’ll be those with the right environment — physical, legal, financial, cultural, human — to absorb it. Good old geography, not just the patent register, will play a big role in sorting the winners from the also-rans.
Conclusion
Let me close with this.
While your competitors play with AI — and they are playing, mostly — the highest-value moves available to you may be stubbornly old-fashioned. This is very likely true for all business questions. But it is especially true for geopolitical risk management.
Instead of relying on Claude as your Machiavelli, it would be much wiser to convene your leadership. To map your exposures. To understand the specific threats that governments could pose to your business. To quantify your risks, not through any complex modelling but through simple probability and impact estimation. To focus ruthlessly on what matters to you. None of that requires “Big Data”. And none of it requires a huge degree of processing.
Your chief task is not to understand the world. Which is good, as that’s very difficult.
Rather, your task is to understand how the world affects you. As the old diplomatic adage goes, “foreign policy starts at home”. The richest intelligence available to most organisations isn’t on the internet — it’s inside your buildings, scattered across functions, countries, and the accumulated knowledge of your own people. Draw on it. Go to your people. Ask them for their views, their judgment, what they’re hearing and seeing and feeling and thinking. Build resilience with your people, not the synthetic simulacrum of analysis that sounds right but is often built on the wrong data, “hallucinates” often enough it cannot be taken on trust, and regularly tells you only what you want to hear.
Beyond that, it may be worth doubling down on your comparative advantage — the oldest idea in economics and still the best. Specialise and excel at what you are uniquely good at. The alternative — using AI to become mediocre at everything — is now available to all of us at remarkably low cost. Indeed, with AI already the fastest diffusing technology in history and the cost of information asymptotically approaching zero, the value of generated text may soon approach zero too.
Stanislav Petrov’s advantage was not better technology than the machine in front of him, but good old-fashioned human judgment that the machine could never have. Paul Reuter’s advantage was not the telegraph everyone was relying on, but the pigeons nobody respected and the creativity to use them. The Soviets’ greatest wins in the Space Race came from simplicity, and their downfall came from overspending. And artificial intelligence, for all its extraordinary power, may simply not be the right tool either to understand the world or how it impacts you.
The age of artificial intelligence, in conclusion, may well belong to those who remember what it cannot do. In other words, time to find your pigeon.
Find your pigeon, together
I closed the speech by asking what your homing pigeon is: the old, human, unfashionable advantage you reach for while everyone else races toward the shiny new tool.
For a board or leadership team, the answer is rarely another model, another feed, or another long report — and certainly not analysis generated by artificial intelligence. The answer, in our view, is shared judgment: the right people reading the same handful of things that matter, at the same time, and asking the same question — what does this mean for us?
That is what we built GD Corporate Membership to do.
As readers of Geopolitical Dispatch know, our analysis rests on judgment developed over decades — as diplomats, as advisers to prime ministers and leaders of international organisations, and now as advisers to boards, executives and investors navigating an increasingly geopolitical world.
Each business day, we sift hundreds of developments, apply the analytic discipline we learned in government, and identify the five most likely to matter: what they mean, why they matter, and how they may affect those running businesses and managing portfolios.
Each Daily Dispatch is designed to be read in under ten minutes. Every Saturday, in Week Signals, we step back from the headlines to the larger forces taking shape. And once a month, members join us behind closed doors for a private roundtable on the developments that defined it — and what’s coming next.
GD Corporate Membership puts all of that in front of ten people across your organisation — board, executive, risk, strategy, legal, finance, communications, government affairs — so your leadership starts each day from the same understanding of the world, rather than ten partial ones. It includes full access for ten, Quarterly Outlook Briefings, and internal distribution rights, so the analysis can circulate where useful.
It also opens the door to our broader advisory work — geopolitical risk assessments, scenario planning, in-person crisis simulations and board briefings. Judgment applied to your specific exposures, not the world’s.
At $1,900 a year — ten seats, around $190 a head — it costs less than a single management report, and a great deal less than being caught off guard.
If your organisation should be reading this together but you’re not the person who decides, forward this to whoever is. Or simply reply, and we’ll talk through what would work best.
Prefer to start on your own? Individual access — the Daily Dispatch, Week Signals and Private Monthly Roundtables — is $190/year, or $19/month.
 | A guest post by
|