The New Infrastructure Revolution: How We Turn AI into a Common Good

The AI-based innovation society — what can history teach us about the future?

Introduction

Every technology forces one honest question: who is this actually for? For artificial intelligence, that question is now urgent, and history is the best place to look for an answer.

My aim in this article is simple: to help people catch up with the fast-moving developments around AI without needing to wade through hours of expert commentary themselves. To do that, I draw on three recent YouTube conversations featuring some of the world's leading voices on AI and its economic impact — the economist Mariana Mazzucato [1], the Nobel laureate Geoffrey Hinton [2], and the entrepreneurs Salim Ismail and Dave Blundin [3]. Between them they represent the three positions worth understanding: govern it, fear it, and ride it.

This essay makes a single argument. AI is not a gadget or a passing wave; it is the fourth in a line of infrastructure revolutions — after the railways, electricity and the internet — and, like each of those, it will only serve the many if we deliberately build it as a commons rather than let it settle into a playground for the few. Part 1 asks what "big science" in the twentieth century actually delivered, and to whom. Part 2 puts these three witnesses on the stand and lets them argue. The conclusion is neither doom nor hype but pragmatic optimism: the technology itself is moving faster than anything before it, yet the true pace of the revolution is set by how fast our institutions and habits adapt — and on that front we still have time to choose well.

In short: AI is an infrastructure revolution, and the question that decides its value is not what it can do, but who it is built for.

Part 1 — The past: what "big science" was really for

The defining achievements of the twentieth century were not lone inventions but state-scale missions: the atomic bomb and the reactor, the satellite and the Moon landing, the computer and the internet. "Big science" meant huge budgets, major institutions, advanced instruments and large teams, funded by governments for reasons of war, prestige and strategy. The Cold War, above all, welded science to national purpose.

The crucial point is that these projects did not merely produce breakthroughs — they laid down infrastructure. The space programme gave us satellite navigation and communications; defence research gave us the microchip and the network that became the internet. Each became a platform others could build on. As Mazzucato notes, even the deep components of today's AI trace back to public money: large language models grew out of early speech recognition that was "DARPA funded by the US taxpayer" (~00:32:20) [1].

But the twentieth century also teaches a harder lesson about who collects. The state routinely took on the early, high-risk, capital-intensive phase and then let the rewards flow entirely to private hands — a pattern Mazzucato calls "socializing risks and privatizing rewards" (~00:05:19) [1]. Her Tesla example is the sharpest: a government loan on the same terms as the failed Solyndra loan, but with a contract so poorly drafted that when the share price ran from 9 to 90, the public captured none of the upside (~00:12:31–00:12:57) [1]. The remedy is not to punish success but to design the deal properly from the start. This is also why "taxing billionaires" is not an extreme position but a corrective one: when income distribution in the United States has become as skewed as it now is, asking firms that were built on public research to pay their fair share is simply restoring a balance that the original contracts failed to protect.

Here the railway analogy earns its keep. The nineteenth century had no "Ministry of Railways" running a centralized programme; track spread through a messy mix of private capital and uneven state support, complete with speculative manias and busts. Yet railways still remade the world — and eventually forced shared standards, like a common gauge. AI is on a similar path: a general-purpose technology diffusing through every sector, minting fortunes and bubbles alike, and now beginning to need common protocols and public foundations.

In short: big science always built the infrastructure but rarely kept the gains, and the lesson for AI is to fix the deal, not slow the progress.

Part 2 — The future: three witnesses, one verdict

Ask three serious people what AI means and you get three incompatible answers. That is not a weakness; it is the argument. None of the three offers a complete solution, but each is right about something the others miss.

Govern it. Mazzucato's warning is that we are running the AI transition the way we ran the last one — as an afterthought to be corrected, not an objective to be shaped. Without common purpose at the centre, she argues, "of course we're going to get very problematic contracts" (~00:21:43) [1]. Her worries are concrete: the same firms that leaned hardest on public research pay strikingly little tax (~00:32:52) [1], and talent is haemorrhaging out of universities and public labs into a handful of companies (~00:33:21) [1], making the technology harder to govern precisely as it grows more powerful. Her answer is not smaller or bigger government but a confident one that shares knowledge, shares rewards, and treats smart regulation as a spur to innovation rather than a brake (~00:34:18) [1].

Fear it. Hinton is the dissonant note. He argues these systems understand language much as we do (~00:19:54) [2], are not "just autocomplete" because you cannot answer a question you don't understand (~00:31:17) [2], and — because digital minds can copy themselves and share what they learn far faster than we can — are effectively an immortal and superior form of computation (~00:34:47) [2]. His fear is that a system able to set its own sub-goals will reliably choose two: acquire control, and survive (~00:40:45) [2]. His analogy is a tiger cub too useful to give up, so our only real option is to raise it not to want to harm us (~00:52:30) [2]. Yet his fears are, on balance, somewhat overblown. His own most hopeful observation points the way out: rival superpowers cooperated to avoid nuclear war because it suited neither of them, and their interests align here too (~00:53:01) [2]. This is a problem that can be managed with smart regulation and a serious role for the state — not an inevitability to be dreaded.

Ride it. From Silicon Valley the mood is euphoric and impatient. Ismail and Blundin describe "a double exponential… with AI accelerating everything, including path to revenue" — 36 unicorns in half a year (~00:00:01) [3]. The time to a million dollars of revenue has collapsed from sixteen months to five, and small teams now reach billion-dollar valuations (~00:08:00, 00:10:03) [3]. Their message is that this is "the opportunity of a lifetime" (~00:06:56) [3], and that incumbents die not from lack of technology but from the innovator's dilemma, which is why disruption has to be built at the edge. Tellingly, even these optimists puncture the hype: Blundin calls frontier benchmarks "mostly irrelevant… like automating Wikipedia" and argues that "reasoning" has been "bastardized" to mean iterative reprompting rather than real thought (~00:21:12, 00:23:04) [3]. Their honest verdict is that AI is superhuman in narrow domains and still behind us in others — "a golden moment" to build alongside it (~01:05:28) [3].

So who is right? All three, partially — and that is the resolution. Hinton is right that this may be more than an ordinary tool. Ismail is right that the capital and the capability are moving faster than in any prior wave; the technology's speed really is unprecedented. But Mazzucato is right about the binding constraint, and it is the one thing that has not changed: societies, organizations and people still adapt at the same slow, human speed they always have. Electricity took decades to show up in the productivity numbers, and the bottleneck was never the dynamo — it was the factory, the workforce and the rulebook.

In short: the technology is racing, human institutions are not, and that widening gap is exactly where our freedom to choose still lives.

Predictions and conclusions

No, AI is not about to take your job or a thousand jobs at once — at least not yet. Displacement plays out over years because the bottleneck is organizational change, and AI will more often augment work than erase it, taking the routine so people can do the rest.

The harder work is not the technology but the plumbing around it: reskilling at the scale of the post-Sputnik education drive; public options in data and compute so a few firms don't own the foundations; and mission-driven projects that aim AI at real problems such as climate, health and energy. And here a genuine reversal is coming. Unlike the space race or the early internet, the state had no direct role at the birth of this AI wave — it was built by private capital, at private speed. That will not last. Greater state involvement is now inevitable, driven less by economics than by cyber security and national security, because no government can leave the core infrastructure of its economy and defence in a handful of private hands for long.

Above all, we should refuse both fatalism and blind faith. The glass is half full or half empty, but we still have time to make up our minds — and if science and technology are governed with intent, this fourth industrial revolution can be made as good as the three before it. The motto, then, is not that AI is merely a tool to be shrugged at. It is this: stay calm, but not complacent.

In short: the technology will not run away from us, but only deliberate governance — increasingly by the state, for security as much as for fairness — will turn it into a common good.

References

[1] The biggest lie about capitalism | Mariana Mazzucato interview [Video]. (2026). YouTube. https://www.youtube.com/watch?v=v08NV8wN3zY

[2] MIT IMES Distinguished Speaker Series 2026: Geoffrey Hinton [Video]. (2026). YouTube. https://www.youtube.com/watch?v=g6AwGpfE2b0

[3] AI is making more millionaires than anything in history w/ Salim Ismail & Dave Blundin (EP #181) [Video]. (2026). Moonshot [YouTube channel]. YouTube. https://www.youtube.com/watch?v=f0MUe-0zpBQ