Floreat Britannia in the era of AI
The bottleneck to progress isn’t engineering, it’s an absurd inability to consider trade-offs that has led to decisions like spending £140m to save the life of a single salmon. AI is about to change all that, says Samuel Albanie
May Britain flourish. I mean this unironically.
To say this in early 2026, however, is to mark oneself out as a dangerous contrarian, or perhaps just someone whose internet service provider has been down since the Platinum Jubilee. I say this with the stubborn affection of a hobbyist developer trying to run Doom on a smart fridge: the hardware is eccentric, the display is glitchy, but deep down, I believe the architecture is solid.
However, to get this fridge playing smoothly, we need to acknowledge that the operating system (our national capacity for decision-making in this increasingly tenuous metaphor) is currently stuck in a boot loop. My optimism for 2026 is not based on hope (a strategy that historically performs poorly in British weather) but instead on a rather specific mechanism: AI-assisted decision-making.
Britain is not currently flourishing. Real wages grew by 33 per cent per decade from 1970 to 2007. Since 2007 they have grown by almost nothing, representing the longest wage stagnation since the Napoleonic Wars, though in fairness to the current era, Napoleon was eventually defeated and exiled to St Helena, whereas the causes of British wage stagnation remain at large and are frequently invited to speak on panels.
Reviewing the damage, Matt Clifford calculated that if Britain had continued on its pre-2008 growth trajectory, we would now be £16,000 per person per year richer. The UK ranked 27th of 35 OECD nations for wage growth in the 2010s. Since 2019, UK real wages have actually contracted by 0.17 per cent per year. while US real wages grew.
Our industrial electricity prices are the highest in Europe. Hinkley Point C will cost £46bn, making it the most expensive power station ever built. South Korea builds equivalent reactors for one-quarter the cost. The Fingleton Report analyses why, citing capital structures and safety frameworks across 162 pages of sober text. But the detail that reached my heart this year, concerns the fish.
Hinkley’s fish protection measures will cost approximately £700m. This includes an acoustic fish deterrent system referred to, apparently without irony, as the “fish disco”. Based on the developer’s own modelling, this nightclub for aquatic life is expected to save 0.083 Atlantic salmon per year. At £700m amortised over the system’s life, this values a single salmon at roughly £140m. This is approximately 700 times the fish’s weight in cocaine.
The stagnation of British growth is a sunk cost. We cannot unstagnate the 2010s. But what I want, as a citizen, is a system going forward where the primary constraint on energy is not the acoustic preferences of 0.083 salmon.
I want a system where the primary constraint on energy is not the acoustic preferences of 0.083 salmon
There are wise souls working on this. I recommend the reliably caustic Alex Chalmers, Sam Bowman’s heroic attempts to enumerate Britain’s challenges, and the non-partisan efforts of the Centre for British Progress. Like these folks, I care deeply about whether Britain flourishes by re-igniting its growth and meaningfully contributing to whatever comes next.
Where I differ is that I am a card-carrying believer in the “Compute Theory of Everything” – the idea that AI will continue to improve significantly as more computer hardware comes online.
If Hans Moravec is right, and “The performance of AI machines tends to improve at the same pace that AI researchers get access to faster hardware,” then we are likely to soon see systems that radically reduce the cost of both software and hardware. Given the human creativity and capital it possesses, I think Britain has a fighting shot at gathering the benefits of this transition. The problem is not that we have forgotten how to pour concrete. What we lack is the ability to coordinate resources without forming a committee to investigate the feasibility of a committee.
The bottleneck will move from engineering to decision-making
In future, I expect the bottleneck for our infrastructure projects to shift from engineering to decision-making and coordination. How can we tackle this?
Robin Hanson’s futarchy (policy guided by prediction markets) offers one framework: bets reveal beliefs, markets aggregate information, decisions improve. The difficulty is that the beliefs being aggregated are themselves unreliable. Forecasting complex systems is genuinely hard. Coordinating outcomes that large groups of stakeholders are happy with is hard too.
This makes me profoundly optimistic about the potential impact of AI-assisted decision-making, driven through better AI forecasting and AI-enabled coordination. If AI progress continues, the cost of generating high-quality probability estimates is about to be substantially reduced. We can move from a world where rigorous forecasting is a luxury good (available only to hedge funds, meteorologists and Dave from across the road who bets on horses) to one where it is broadly accessible. There are major challenges here too (structural risks from correlated predictions etc.), but I’m broadly optimistic.
However, high-fidelity forecasting is useless if your culture treats ‘earnestness’ as a breach of etiquette.
The problem is not that we have forgotten how to pour concrete. What we lack is the ability to coordinate resources without forming a committee to investigate the feasibility of a committee
I confess to being an unashamed admirer of the American Spirit. It is, after all, the spirit of a hero of mine, Ben Franklin, a man who looked at a lethal thunderstorm and decided it was the perfect time to fly a kite. It turns out that a nation founded on that sort of practical audacity, alongside ‘Life, Liberty and the pursuit of Happiness’, is uniquely calibrated to invent the future. My hope is that Britain stops viewing this instinct as a compliance violation and remembers that we invented the steam engine before we invented the clipboard.
Franklin didn’t confuse audacity with arson. He was a pragmatist who knew that instantaneous combustion tends to hinder one’s long-term career prospects. He respected the raw power of electricity and had the foresight to invent the lightning rod. The Compute Theory of Everything is global, but to prosper we’ll need to build both the kite and the rod. And you cannot catch lightning in a feasibility study. The storm is cresting the horizon, but the spark only favours the nations nervously holding kite strings in the rain. Fortunately, standing in the damp while tinkering with dangerous machinery is practically our national sport. If Britain can get back in the game, I see no reason why the next generation of AI shouldn’t have, at the very least, a dry sense of humour.
Samuel Albanie is a Research Scientist at Google DeepMind, but writes in a strictly personal capacity. He covers the collision of AI scaling laws and British growth. He is a stubborn optimist. He writes at https://samuelalbanie.substack.com/