David to the American Goliath

If your Twitter feed resembles mine, a cursory glance would reveal that in many minds our national stagnation has many simple causes; electricity is too expensive, there aren't enough trams, or house prices are so high that the idea of buying one feels like a hallucination painted on a distant horizon. While solving each of these will give us momentum, the unhappy truth is our dream destination is not a reasonable gas bill, but an image of a place: Silicon Valley.

Every nation aspires to have its own; it's wormed its way so deep into our collective psyche, that within an hour of starting a tech company, a local politician will arrive to christen the area "Silicon Roundabout", "Silicon Hill", or possibly "Silicon Junction 9 on the M1". A ceremony performed under the illusion that by uttering this incantation, the heavens shall open, wealth shall flood forth, and the stagnation will be over. We desire it so much that eyewatering prices, poor public transport, and severe "public sanitation" issues fall by the wayside. Our one desire, however, is not Silicon Valley. It's the money-printing golden goose, its being at the centre of something, a longing for the feeling that we have the means to solve our problems if not the will.

Instead of small ambitions, I propose that we go big, or go back to an overly-expensive, poorly-heated home. Due to recent advances in AI, we no longer have a reason not to. A cold piece of Californian silicon could be along to steal your job soon. Then what good would moderate growth for a couple of years be when the economy departs for its unfeeling arms? Therefore, as a nation, we must have a singular goal: to ensure that every challenger company has no option but to base themselves here in Britain.

Now, there are two scarce resources in the AI world; training data, and GPUs. The architecture and model weights have value, but Mixtral will open-source their weights, and Meta will merilly publish all, but neither will publish their training process. Therefore, we know by implication that the training process makes a company valuable, and why it's worth it to throw €105mn at a company that just gave their product away for free.

Many don't realise that there's also a unique compounding feature to AI companies, which I call the "Creator Effect". In using an AI, you create additional training data, which then can be used to train an improved AI. This positive feedback cycle means that by having trained an AI once, as your product gets used more and more, the product improves to fit the use case of your users. While AIs are often initially trained on public data, this new, created data will remain hoarded away, with the effect of the owner pulling up the ladder behind it.

How do we leverage this effect to build a British Silicon Valley, I hear you ask? It's simple. Set up a government-backed national data pool that acts as a "seed dataset" while embedding a mechanism that guarantees the private data flows back to rejoin it.

It doesn't matter what we dump in the data pool. It could be medical images liberated from dusty NHS servers, the results of a thousand medical tests, or we could convince the insurance industry to hand over every case that's ever crossed a desk. What matters is that it exists in one place, in an accessible format, while enticing founders enough that they know using it would be make or break.

Given our goal is to build a better, British Silicon Valley, access to the pool must come with a geographic tie-in. Perhaps companies that want to participate must 'buy in' with a 5-10% ownership stake transferred to a national wealth fund, or must have 75% of their staff based at a headquarters within the UK, but what they must certainly do is agree to share any training data they acquire with the national data pool.

Through every byte that returns to the pool, we leverage the creator effect to turn our seed dataset into a self-strengthening feedback loop that joins the strength of each pool member into a multiplier for all of them. At the same time, as the size of the dataset expands, companies will feel the pressure to join, while also becoming a golden handcuff gleaming from the wrist of any company that already is.

There are three clear benefits to this approach. Firstly, in the same way that the world's advertisement spending flows to California, the revenue for every AI built from the pool shall flow to Britain. Secondly, the nation benefits directly from its ownership stake in each member. Finally, data is the main barrier of entry to newcomers, a barrier we have now eliminated. Naturally, this will lead to companies being far more competitive on the fundamentals while having an ecosystem free of data rentiers.

The downside is that larger, more established companies will be reluctant to participate at such a steep buy-in. That's absolutely by design. To be the David to the American Goliath we need to be nimble, unburdened by bureaucracy, and fast-moving. That doesn't mean we shouldn't work with established British businesses but instead offer a softer buy-in. If they contribute some data, they get early access to what our pool members offer, we get an enthusiastic client list, and in the process, we rejuvenate the wider economy.

It sounds too simple, but after spending half my life programming I've learned the most effective ideas often are. If we follow this path, we have a self-strengthening system that forces cooperation between companies within our borders on the factors that limit them while directing competition outwards. Not only that, we create a national advantage for developing AIs that would be hard for a far less interventionist nation like America to implement while restoring our dynamism. Of course, we may not be able to catch our desire, but certainly, we must chase it.